Free Essay: Ns2 Introduction to Network Simulator - 112068 Words

Ns2 Introduction to Network Simulator

Introduction to Network Simulator NS2

Teerawat Issariyakul • Ekram Hossain

Introduction to Network Simulator NS2

123

Teerawat Issariyakul TOT Public Company Limited 89/2 Moo 3 Chaengwattana Rd. Thungsonghong, Laksi Bangkok, Thailand 10210 teerawas@tot.co.th iteerawat@hotmail.com

Ekram Hossain Department of Electrical & Computer Engineering University of Manitoba 75A Chancellor’s Circle Winnipeg MB R3T 5V6 Canada ekram@ee.umanitoba.ca

ISBN: 978-0-387-71759-3 DOI: 10.1007/978-0-387-71760-9

e-ISBN: 978-0-387-71760-9

Library of Congress Control Number: 2008928147 c 2009 Springer Science+Business Media, LLC All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Springer Science+Business Media, LLC, 233 Spring Street, New York, NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden. The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are not identiﬁed as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights. Printed on acid-free paper springer.com

To our families

Preface

NS2 is an open-source event-driven simulator designed speciﬁcally for research in computer communication networks. Since its inception in 1989, NS2 has continuously gained tremendous interest from industry, academia, and government. Having been under constant investigation and enhancement for years, NS2 now contains modules for numerous network components such as routing, transport layer protocol, application, etc. To investigate network performance, researchers can simply use an easy-to-use scripting language to conﬁgure a network, and observe results generated by NS2. Undoubtedly, NS2 has become the most widely used open source network simulator, and one of the most widely used network simulators. Unfortunately, most research needs simulation modules which are beyond the scope of the built-in NS2 modules. Incorporating these modules into NS2 requires profound understanding of NS2 architecture. Currently, most NS2 beginners rely on online tutorials. Most of the available information mainly explains how to conﬁgure a network and collect results, but does not include suﬃcient information for building additional modules in NS2. Despite its details about NS2 modules, the formal documentation of NS2 is mainly written as a reference book, and does not provide much information for beginners. The lack of guidelines for extending NS2 is perhaps the greatest obstacle, which discourages numerous researchers from using NS2. At this moment, there is no guide book which can help the beginners understand the architecture of NS2 in depth. The objective of this textbook is to act as a primer for NS2 beginners. The book provides information required to install NS2, run simple examples, modify the existing NS2 modules, and create as well as incorporate new modules into NS2. To this end, the details of several built-in NS2 modules are explained in a comprehensive manner. NS2 by itself contains numerous modules. As time elapses, researchers keep developing new NS2 modules. This book does not include the details of all NS2

VIII

Preface

modules, but does so for selected modules necessary to understand the basics of NS2. For example, it leaves out the widely used modules such as wireless node or web caching. We believe that once the basics of NS2 are grasped, the readers can go through other documentations, and readily understand the details of other NS2 components. The details of Network AniMator (NAM) and Xgraph are also omitted here. We understand that these two tools are nice to have and could greatly facilitate simulation and analysis of computer networks. However, we believe that they are not essential to the understanding of the NS2 concept, and their information are widely available through most of the online tutorials. This textbook can be used by researchers who need to use NS2 for communication network performance evaluation based on simulation. Also, it can be used as a reference textbook for laboratory works for a senior undergraduate level course or a graduate level course on telecommunication networks oﬀered in Electrical and Computer Engineering and Computer Science Programs. Potential courses include “Network Simulation and Modeling”, “Computer Networks”, “Data Communications”, “Wireless Communications and Networking”, “Special Topics on Telecommunications”. In a ﬁfteen-class course, we suggest the ﬁrst class for an introduction to programming (Appendix A), and other 14 classes for each of the 14 chapters. Alternately, the instructor may allocate 10 classes for teaching and 5 classes for term projects. In this case, we suggest that the materials presented in this book are taught in the following order: Chapters 1–2, 3, 12, 4–5, 6, 7–8, 9–11, 13 and 14. When the schedule is really tight, we suggest the readers to go through Chapters 2, 4–7, and 9–10. The readers may start by getting to know NS2 in Chapter 2, and learn the main concepts of NS2 in Chapters 4–5. Chapters 6–7 and 9–10 present the details of most widely used NS2 modules. From time to time, the readers may need to visit Chapter 3, 8, and 12 for further information. If tracing is required, the readers may also have to go through Chapter 13. Finally, Chapter 14 would be useful for those who need to extend NS2 beyond it scopes. We recommend the readers who intend to go through the entire book to proceed chapter by chapter. A summary of all the chapters in this book is provided below. As the opening chapter, Chapter 1 gives an introduction to computer networks and network simulation. The emphasis is on event-driven simulation from which NS2 is developed. An overview of Network Simulator 2 (NS2) is discussed in Chapter 2. Here, we brieﬂy show the two-language NS2 architecture, NS2 directory and the conventions used in this book, and NS2 installation guidelines for UNIX and Windows systems. We also present a three-step simulation formulation as well as a simple example of NS2 simulation. Finally, we demonstrate how to use the make utility to incorporate new modules into NS2. Chapter 3 explains the details of the NS2 two language structure, which consists of the following six main C++ classes: Tcl, Instvar, TclObject,

Preface

IX

TclClass, TclCommand, and EmbeddedTcl. Chapters 4–5 present the very main simulation concept of NS2. While Chapter 4 explains implementation of event-driven simulation in NS2, Chapter 5 focuses on network objects as well as packet forwarding mechanism. Chapters 6–11 present the following six most widely used NS2 modules. First, nodes (Chapter 6) act as routers and computer hosts. Secondly, links, particularly SimpleLink objects (Chapter 7), deliver packets from one network object to another. They model packet transmission time as well as packet buﬀering. Thirdly, packets (Chapter 8) contain necessary information in its header. Fourthly, agents (Chapters 9–10) are responsible for generating packets. NS2 has two main transport-layer agents: TCP and UDP agents. Finally, applications (Chapter 11) model the user demand for data transmission. Chapter 12 presents three helper modules: timers, random number generators, and error models. It also discusses the concepts of two bit-wise operations, namely, bit masking and bit shifting, which are used throughout NS2. Chapter 13 summarizes the post-simulation process, which consists of three main parts: debugging, variable and packet tracing, and result compilation. After discussing all the NS components, Chapter 14 demonstrates how a new module is developed and integrated into NS2 through two following examples: Automatic Repeat reQuest (ARQ) and packet schedulers. Appendices A and B provide programming details which could be useful for the beginners. These details include an introduction to Tcl, OTcl, and AWK programming languages as well as a review of the polymorphism OOP concept. As the ﬁnal words, we would like to express sincere gratitude to our colleagues, especially, Surachai Chieochan, at the University of Manitoba, and the colleagues at TOT Public Company Limited, Bangkok, Thailand, for their continuous support. Last but not the least, we would like to acknowledge our families as well as our partners – Wannasorn and Rumana – for their incessant moral support and patient understanding throughout this endeavor.

TOT Public Company Limited University of Manitoba July 2008

Teerawat Issariyakul Ekram Hossain

Contents

1

Simulation of Computer Networks . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.1 Computer Networks and the Layering Concept . . . . . . . . . . . . . . 1 1.1.1 Layering Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.1.2 OSI and TCP/IP Reference Models . . . . . . . . . . . . . . . . . 3 1.2 System Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1.2.1 Analytical Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 1.2.2 Simulation Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 1.3 Basics of Computer Network Simulation . . . . . . . . . . . . . . . . . . . . 6 1.3.1 Simulation: The Formal Deﬁnition . . . . . . . . . . . . . . . . . . 7 1.3.2 Elements of Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 1.4 Time-Dependent Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.4.1 Time-Driven Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.4.2 Event-Driven Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . 10 1.5 A Simulation Example: A Single-Channel Queuing System . . . . 12 1.6 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 Introduction to Network Simulator 2 (NS2) . . . . . . . . . . . . . . . 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Basic Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Installation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1 Installing an All-In-One NS2 Suite on Unix-Based Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.2 Installing an All-In-One NS2 Suite on Windows-Based Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Directories and Convention . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.1 Directories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.2 Convention . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 Running NS2 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5.1 NS2 Program Invocation . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5.2 Main NS2 Simulation Steps . . . . . . . . . . . . . . . . . . . . . . . . 2.6 A Simulation Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 19 20 21 22 22 23 23 23 26 26 26 27

2

XII

Contents

2.7 Including C++ Modules into NS2 and the make Utility . . . . . . 2.7.1 An Invocation of a Make Utility . . . . . . . . . . . . . . . . . . . . 2.7.2 A Make Descriptor File . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.7.3 NS2 Descriptor File . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.8 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Linkage Between OTcl and C++ in NS2 . . . . . . . . . . . . . . . . . . . 3.1 The Two-Language Concept in NS2 . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Class Tcl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 Obtain a Reference to the Tcl Instance . . . . . . . . . . . . . . 3.2.2 Invoking a Tcl Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.3 Pass or Receive Results to/from the Interpreter . . . . . . 3.2.4 Reporting Error and Quitting the Program . . . . . . . . . . 3.2.5 Retrieve the Reference to TclObjects . . . . . . . . . . . . . . . 3.3 Class InstVar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.1 Real and Integer Variables . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.2 Bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.3 Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.4 Boolean . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 Class TclObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.1 Reference to a TclObject . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.2 Creating and Destroying a Shadow TclObject . . . . . . . . 3.4.3 Binding Variables in the Compiled and Interpreted Hierarchies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.4 OTcl Commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5 Class TclClass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.1 An Overview of Class TclClass . . . . . . . . . . . . . . . . . . . . 3.5.2 TclObject Creation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.3 Naming Convention for Class TclClass . . . . . . . . . . . . . 3.5.4 Instantiation of Mapping Variables . . . . . . . . . . . . . . . . . 3.6 Class TclCommand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6.1 Invoking a TclCommand . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6.2 Creating a TclCommand . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6.3 Deﬁning Your Own TclCommand . . . . . . . . . . . . . . . . . . 3.7 Class EmbeddedTcl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.8 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Implementation of Discrete-Event Simulation in NS2 . . . . . . 4.1 NS2 Simulation Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Events and Handlers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 An Overview of Events and Handlers . . . . . . . . . . . . . . . 4.2.2 Class NsObject: A Child Class of Class Handler . . . . . 4.2.3 Classes Packet and AtEvent: Child Classes of Class Event . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 The Scheduler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.1 Main Components of the Scheduler . . . . . . . . . . . . . . . . .

33 33 33 35 36 37 38 42 42 43 44 45 45 46 47 47 48 48 49 50 50 54 56 61 61 62 62 63 63 63 64 65 66 67 69 70 70 70 71 72 74 74

4

Contents

XIII

4.3.2 Data Encapsulation and Polymorphism Concepts . . . . . 4.3.3 Main Functions of the Scheduler . . . . . . . . . . . . . . . . . . . 4.3.4 Dynamics of the Unique ID of an Event . . . . . . . . . . . . . 4.3.5 Scheduling-Dispatching Mechanism . . . . . . . . . . . . . . . . . 4.3.6 Null Event and Dummy Event Scheduling . . . . . . . . . . . 4.4 The Simulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4.1 Main Components of a Simulation . . . . . . . . . . . . . . . . . . 4.4.2 Retrieving the Instance of the Simulator . . . . . . . . . . . . 4.4.3 Simulator Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4.4 Running Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5 Instprocs of OTcl Class Simulator . . . . . . . . . . . . . . . . . . . . . . . . . 4.6 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Network Objects: Creation, Conﬁguration, and Packet Forwarding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Overview of NS2 Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.1 Functionality-Based Classiﬁcation of NS2 Modules . . . . 5.1.2 C++ Class Hierarchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 NsObjects: A Network Object Template . . . . . . . . . . . . . . . . . . . . 5.2.1 Class NsObject . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.2 Packet Forwarding Mechanism of NsObjects . . . . . . . . . 5.3 Connectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.1 Class Declaration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.2 OTcl Conﬁguration Commands . . . . . . . . . . . . . . . . . . . . 5.3.3 Packet Forwarding Mechanism . . . . . . . . . . . . . . . . . . . . . 5.4 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

75 75 77 78 78 80 81 82 83 83 84 84 87 87 87 88 90 90 91 91 92 93 96 98

6

Nodes as Routers or Computer Hosts . . . . . . . . . . . . . . . . . . . . . 99 6.1 An Overview of Nodes in NS2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 6.1.1 Architecture of a Node . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 6.1.2 Related Instproc of Class Node . . . . . . . . . . . . . . . . . . . . 101 6.1.3 Default Nodes and Node Conﬁguration Interface . . . . . 102 6.2 Routing Mechanism in NS2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 6.3 Route Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 6.4 Classiﬁers: Multi-target Packet Forwarders . . . . . . . . . . . . . . . . . . 107 6.4.1 Class Classifier and Its Main Components . . . . . . . . 107 6.4.2 Hash Classiﬁers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 6.4.3 Port Classiﬁers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 6.4.4 Installing Classiﬁers in a Node . . . . . . . . . . . . . . . . . . . . . 116 6.5 Routing Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 6.5.1 An Overview of Routing Modules . . . . . . . . . . . . . . . . . . 118 6.5.2 C++ Class RoutingModule . . . . . . . . . . . . . . . . . . . . . . . . 121 6.5.3 OTcl Class RtModule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 6.5.4 C++ Class BaseRoutingModuleand OTcl class RtModule/Base . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

XIV

Contents

6.6 Node Object Conﬁguration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 6.6.1 Relationship Among Instvars module list , reg module , rtnotif , and ptnotif . . . . . . . . . . . . . . 126 6.6.2 Adding/Deleting a Routing Entry . . . . . . . . . . . . . . . . . . 127 6.6.3 Agent Attachment/Detachment . . . . . . . . . . . . . . . . . . . . 127 6.6.4 Node Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128 6.6.5 Route Conﬁguration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 6.7 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137 7 Link and Buﬀer Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 7.1 Introduction to SimpleLink Objects . . . . . . . . . . . . . . . . . . . . . . . . 139 7.1.1 Main Components of a SimpleLink . . . . . . . . . . . . . . . . . 139 7.1.2 Instprocs for Conﬁguring a SimpleLink Object . . . . . . 140 7.1.3 The Constructor of Class SimpleLink . . . . . . . . . . . . . . . 142 7.2 Modeling Packet Departure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 7.2.1 Packet Departure Mechanism . . . . . . . . . . . . . . . . . . . . . . 143 7.2.2 C++ Class LinkDelay . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144 7.3 Buﬀer Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146 7.3.1 Class PacketQueue: A Model for Packet Buﬀering . . . . 147 7.3.2 Queue Handler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148 7.3.3 Queue Blocking and Callback Mechanism . . . . . . . . . . . 149 7.3.4 Class DropTail: A Child Class of Class Queue . . . . . . . 151 7.4 A Sample Two-Node Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152 7.4.1 Network Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 7.4.2 Packet Flow Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . 154 7.5 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154 Packets, Packet Headers, and Header Format . . . . . . . . . . . . . . 157 8.1 An Overview of Packet Modeling Principle . . . . . . . . . . . . . . . . . . 157 8.1.1 Packet Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 8.1.2 A Packet as an Event: A Delayed Packet Reception Event . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160 8.1.3 A Linked List of Packets . . . . . . . . . . . . . . . . . . . . . . . . . . 161 8.1.4 Free Packet List . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 8.2 Packet Allocation and Deallocation . . . . . . . . . . . . . . . . . . . . . . . . 163 8.2.1 Packet Allocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163 8.2.2 Packet Deallocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166 8.3 Packet Header . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168 8.3.1 An Overview of First Level Packet Composition: Oﬀseting Protocol Speciﬁc Header on the Packet Header . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 8.3.2 Common Packet Header . . . . . . . . . . . . . . . . . . . . . . . . . . . 170 8.3.3 IP Packet Header . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 8.3.4 Packet Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172 8.3.5 Protocol Speciﬁc Headers . . . . . . . . . . . . . . . . . . . . . . . . . 174

8

Contents

XV

Packet Header Access Mechanism . . . . . . . . . . . . . . . . . . 178 Packet Header Manager . . . . . . . . . . . . . . . . . . . . . . . . . . . 180 Protocol Speciﬁc Header Composition and Packet Header Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 8.4 Data Payload . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187 8.5 Customizing Packets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191 8.5.1 Creating Your Own Packet . . . . . . . . . . . . . . . . . . . . . . . . 191 8.5.2 Activate/Deactivate a Protocol Speciﬁc Header . . . . . . 192 8.6 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194 9 Transport Control Protocols Part 1 – An Overview and User Datagram Protocol implementation . . . . . . . . . . . . . . . . . . 197 9.1 UDP and TCP Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197 9.1.1 UDP Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197 9.1.2 TCP Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198 9.2 Basic Agents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203 9.2.1 Applications, Agents, and a Low-level Network . . . . . . . 203 9.2.2 Agent Conﬁguration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 9.2.3 Internal Mechanism for Agents . . . . . . . . . . . . . . . . . . . . . 206 9.2.4 Guidelines to Deﬁne a New Transport Layer Agent . . . 210 9.3 UDP (User Datagram Protocol) and Null Agents . . . . . . . . . . . . 210 9.3.1 Null (Receiving) Agents . . . . . . . . . . . . . . . . . . . . . . . . . . . 211 9.3.2 UDP (Sending) Agent . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211 9.3.3 Setting Up a UDP Connection . . . . . . . . . . . . . . . . . . . . . 215 9.4 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215

8.3.6 8.3.7 8.3.8

10 Transport Control Protocols Part 2 – Transmission Control Protocol (TCP) . . . . . . . . . . . . . . . . . . . . . 217 10.1 An Overview of TCP Agents in NS2 . . . . . . . . . . . . . . . . . . . . . . . 217 10.1.1 Setting Up a TCP Connection . . . . . . . . . . . . . . . . . . . . . 217 10.1.2 Packet Transmission and Acknowledgment Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218 10.1.3 TCP Header . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219 10.1.4 Deﬁning TCP Sender and Receiver . . . . . . . . . . . . . . . . . 219 10.2 TCP Receiver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222 10.2.1 Class Acker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224 10.2.2 Class TcpSink . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228 10.3 TCP Sender . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230 10.4 TCP Packet Transmission Functions . . . . . . . . . . . . . . . . . . . . . . . 230 10.4.1 Function sendmsg(nbytes) . . . . . . . . . . . . . . . . . . . . . . . . 231 10.4.2 Function send much(force,reason,maxburst) . . . . . . 232 10.4.3 Function output(seqno,reason) . . . . . . . . . . . . . . . . . . 234 10.4.4 Function send one() . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236 10.5 ACK Processing Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237 10.5.1 Function recv(p,h) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238 10.5.2 Function recv newack helper(pkt) . . . . . . . . . . . . . . . . 239

XVI

Contents

10.5.3 10.6 Timer 10.6.1 10.6.2 10.6.3 10.6.4 10.6.5 10.6.6 10.6.7 10.6.8 10.6.9

Function newack(pkt) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241 Related Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242 RTT Sample Collection . . . . . . . . . . . . . . . . . . . . . . . . . . . 242 RTT Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244 Overview of State Variables . . . . . . . . . . . . . . . . . . . . . . . 244 Retransmission Timer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246 Function Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247 Function rtt update(tao) . . . . . . . . . . . . . . . . . . . . . . . . 248 Function rtt timeout() . . . . . . . . . . . . . . . . . . . . . . . . . . 250 Function rtt backoff() . . . . . . . . . . . . . . . . . . . . . . . . . . 251 Function set rtx timer()and Function reset rtx timer(mild,backoff) . . . . . . . . . . . . . . . . . . 252 10.6.10 Function newtimer(pkt) . . . . . . . . . . . . . . . . . . . . . . . . . . 253 10.6.11 Function timeout(tno) . . . . . . . . . . . . . . . . . . . . . . . . . . . 253 10.7 Window Adjustment Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255 10.7.1 Function opencwnd() . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256 10.7.2 Function slowdown(how) . . . . . . . . . . . . . . . . . . . . . . . . . . 256 10.8 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259

11 Application: User Demand Indicator . . . . . . . . . . . . . . . . . . . . . . . 261 11.1 Relationship Between an Application and a Transport Layer Agent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261 11.2 Details of Class Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265 11.2.1 Functions of Classes Application and Agent . . . . . . . . 265 11.2.2 Public Functions of Class Application . . . . . . . . . . . . . 266 11.2.3 Related Public Functions of Class Agent . . . . . . . . . . . . 267 11.2.4 OTcl Commands of Class Application . . . . . . . . . . . . . 268 11.3 Traﬃc Generators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268 11.3.1 An Overview of Class TrafficGenerator . . . . . . . . . . . 268 11.3.2 Main Mechanism of a Traﬃc Generator . . . . . . . . . . . . . 270 11.3.3 Built-in Traﬃc Generators in NS2 . . . . . . . . . . . . . . . . . . 272 11.3.4 Class CBR Traffic: An Example Traﬃc Generator . . . 275 11.4 Simulated Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277 11.4.1 FTP (File Transfer Protocol) . . . . . . . . . . . . . . . . . . . . . . 278 11.4.2 Telnet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278 11.5 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279 12 Related Helper Classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281 12.1 Timers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281 12.1.1 Implementation Concept of Timer in NS2 . . . . . . . . . . . 281 12.1.2 OTcl Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283 12.1.3 C++ Based Class Implementation . . . . . . . . . . . . . . . . . . 285 12.1.4 Guidelines for Implementing Timers in NS2 . . . . . . . . . . 295 12.2 Implementation of Random Numbers in NS2 . . . . . . . . . . . . . . . . 296 12.2.1 Random Number Generation . . . . . . . . . . . . . . . . . . . . . . 296 12.2.2 Seeding a Random Number Generator . . . . . . . . . . . . . . 296

Contents

XVII

12.2.3 OTcl and C++ Implementation . . . . . . . . . . . . . . . . . . . . 298 12.2.4 Randomness in Simulation Scenarios . . . . . . . . . . . . . . . . 300 12.2.5 Random Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303 12.2.6 Guidelines for Random Number Generation in NS2 . . . 306 12.3 Built-in Error Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307 12.3.1 OTcl Implementation: Error Model Conﬁguration . . . . 309 12.3.2 C++ Implementation: Error Model Simulation . . . . . . . 312 12.3.3 Guidelines for Implementing a New Error Model in NS2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320 12.4 Bit Operations in NS2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321 12.4.1 Bit Masking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321 12.4.2 Bit Shifting and Decimal Multiplication . . . . . . . . . . . . . 323 12.5 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324 13 Processing an NS2 Simulation: Debugging, Tracing, and Result Compilation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327 13.1 Debugging: A Process to Remove Programming Errors . . . . . . . 327 13.1.1 Types of Programming Errors . . . . . . . . . . . . . . . . . . . . . 327 13.1.2 Debugging Guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329 13.2 Variable Tracing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332 13.2.1 Activation Process for Variable Tracing . . . . . . . . . . . . . 332 13.2.2 Instvar Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334 13.2.3 TracedVar Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335 13.2.4 Tracers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335 13.2.5 Connections Among a TclObject, a TracedVar Object, a Tracer, and a Trace File . . . . . . . . . . . . . . . . . . 338 13.2.6 Trace File Format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339 13.3 Packet Tracing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341 13.3.1 OTcl Conﬁguration Interfaces . . . . . . . . . . . . . . . . . . . . . . 342 13.3.2 C++ Main Packet Tracing Class Trace . . . . . . . . . . . . . . 347 13.3.3 C++ Helper Class BaseTrace . . . . . . . . . . . . . . . . . . . . 350 13.3.4 Various Types of Packet Tracing Objects . . . . . . . . . . . . 352 13.3.5 Packet Trace Format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355 13.4 Compilation of Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . 356 13.5 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360 14 Developing New Modules for NS2 . . . . . . . . . . . . . . . . . . . . . . . . . 363 14.1 Automatic Repeat reQuest (ARQ) . . . . . . . . . . . . . . . . . . . . . . . . . 363 14.1.1 The Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364 14.1.2 C++ Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365 14.1.3 OTcl Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371 14.1.4 ARQ Under a Delayed (Error-Free) Feedback Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374 14.2 Packet Scheduling for Multi-Flow Data Transmission . . . . . . . . . 376 14.2.1 The Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376

XVIII Contents

14.2.2 C++ Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377 14.2.3 OTcl Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381 14.3 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387 A Programming Essentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389 A.1 Tcl Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389 A.1.1 Program Invocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389 A.1.2 A Simple Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390 A.1.3 Variables and Data Types . . . . . . . . . . . . . . . . . . . . . . . . . 391 A.1.4 Input/Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394 A.1.5 Mathematical Expressions . . . . . . . . . . . . . . . . . . . . . . . . . 396 A.1.6 Control Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397 A.1.7 Procedures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399 A.2 Objected Oriented Tcl (OTcl) Programming . . . . . . . . . . . . . . . . 400 A.2.1 Class and Inheritance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 400 A.2.2 Class Member Procedures and Variables . . . . . . . . . . . . 401 A.2.3 Object Construction and the Constructor . . . . . . . . . . . 402 A.2.4 Related Instprocs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403 A.3 AWK Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405 A.3.1 Program Invocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405 A.3.2 An AWK Script . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406 A.3.3 AWK Programming Structure . . . . . . . . . . . . . . . . . . . . . 407 A.3.4 Pattern Matching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407 A.3.5 Basic Actions: Operators and Output . . . . . . . . . . . . . . . 408 A.3.6 Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 410 A.3.7 Control Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411 A Review of the Polymorphism Concept in OOP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413 B.1 Fundamentals of Polymorphism . . . . . . . . . . . . . . . . . . . . . . . . . . . 413 B.2 Type Casting and Function Ambiguity . . . . . . . . . . . . . . . . . . . . . 416 B.3 Virtual Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417 B.4 Abstract Classes and Pure Virtual Functions . . . . . . . . . . . . . . . . 418 B.5 Class Composition: An Application of Type Casting Polymorphism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419 B.6 Programming Polymorphism with No Type Casting: An Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420 B.7 A Scalability Problem Caused by Non Type Casting Polymorphism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421 B.8 The Class Composition Programming Concept . . . . . . . . . . . . . . 422

B

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427 General Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429 Code Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433

1 Simulation of Computer Networks

People communicate. One way or another, they exchange some information among themselves all the times. In the past several decades, many electronic technologies have been invented to aid this process of exchanging information in an eﬃcient and creative way. Among these are the creation of ﬁxed telephone networks, the broadcasting of television and radio, the advent of computers, and the emergence of wireless sensation. Originally, these technologies existed and operated independently, serving their very own purposes. Not until recently that these technological wonders seem to converge, and it is a well-known fact that a computer communication network is a result of this convergence. This chapter presents an overview of computer communication networks, and the basics of simulation of such a network. Section 1.1 introduces a computer network along with the reference model which is used for describing the architecture of a computer communication network. A brief discussion on designing and modeling a complex system such as a computer network is then given in Section 1.2. In Section 1.3, the basics of computer network simulation are discussed. Section 1.4 presents one of the most common type of network simulation-time-dependent simulation. An example simulation is given in Section 1.5. Finally, Section 1.6 summarizes the chapter.

1.1 Computer Networks and the Layering Concept
A computer network is usually deﬁned as a collection of computers interconnected for gathering, processing, and distributing information. Computer is used as a broad term here to include devices such as workstations, servers, routers, modems, base stations, wireless extension points, etc. These computers are connected by communications links such as copper cables, ﬁber optic cables, and microwave/satellite/radio links. A computer network can be built as a nesting and/or interconnection of several networks. The Internet is a good example of computer networks. In fact, it is a network of networks,
T. Issariyakul, E. Hossain, Introduction to Network Simulator NS2, DOI: 10.1007/978-0-387-71760-9 1, c Springer Science+Business Media, LLC 2009

2

1 Simulation of Computer Networks

within which, tens of thousands of networks interconnect millions of computers worldwide. 1.1.1 Layering Concept A computer network is a complex system. To facilitate design and ﬂexible implementation of such a system, the concept of layering is introduced. Using a layered structure, the functionalities of a computer network can be organized as a stack of layers. There is a peer-to-peer relationship (or virtual link) between the corresponding layers in two communicating nodes. However, actual data ﬂow occurs in a vertical fashion – from the highest layer to the lowest layer in a node, and then through the physical link to reach the lowest layer at the other node, and then following upwards to reach the highest layer in the stack. Each layer represents a well-deﬁned and speciﬁc part of the system and provides certain services to the above layer. Accessible (by the upper layers) through so-called interfaces, these services usually deﬁne what should be done in terms of network operations or primitives, but does not speciﬁcally deﬁne how such things are implemented. The details of how a service is implemented is deﬁned in a so-called protocol. For example, the transmitter at a source computer can use a speciﬁc protocol (e.g., a data encoding scheme) at the physical layer to transmit information bits to the receiving computer, which should be able to decode the received information based on the protocol rules. The beauty of this layering concept is the layer independency. That is, a change in a protocol of a certain layer does not aﬀect the rest of the system as long as the interfaces remain unchanged. Here, we highlight the words services, protocol, and interface to emphasize that it is the interaction among these components that makes up the layering concept. Figure 1.1 graphically shows an overall view of the layering concept used for communication between two computer hosts: a source host and a
Source M Layer 4 Protocol Destination M

H3 M1

H3 M2

Layer 3 Protocol

H3 M1

H3 M2

Layer 2 Protocol H2 H3 M1 H2 H3 M2 Layer 1 Protocol H2 H3 M1 H2 H3 M2

H1 H2 H3 M1

H1 H2 H3 M2

H1 H2 H3 M1

H1 H2 H3 M2

Physical Medium

Fig. 1.1. Data ﬂow in a layered network architecture.

1.1 Computer Networks and the Layering Concept

3

destination host. In this ﬁgure, the functionality of each computer host is divided into four layers.1 When virtually linked with the same layer on another host, these layers are called peers.2 Although not directly connected to each other, these peers virtually communicate with one another using a protocol represented by an arrow. As has already been mentioned, the actual communication needs to propagate down the stack and use the above layering concept. Suppose an application process running on Layer 4 of the source generates data or messages destined for the destination. The communication starts by passing a generated message M down to Layer 3, where the data are segmented into two chunks (M1 and M2), and control information called header (H3) speciﬁc to Layer 3 is appended to M1 and M2. The control information are, for example, sequence numbers, packet sizes, and error checking information. These information are understandable and used only by the peering layer on the destination to recover the data (M). The resulting data (e.g., H3+M1) is handed to the next lower layer, where some protocol-speciﬁc control information are again added to the message. This process continues until the message reaches the lowest layer, where transmission of information is actually performed over a physical medium. Note that, along the line of these processes, it might be necessary to further segment the data from upper layers into smaller segments for various purposes. When the message reaches the destination, the reverse process takes place. That is, as the message is moving up the stack, its headers are ripped oﬀ layer by layer. If necessary, several messages are put together before being passed to the upper layer. The process continues until the original message (M) is recovered at Layer 4. 1.1.2 OSI and TCP/IP Reference Models The OSI (Open Systems Interconnection) model was the ﬁrst reference model developed by ISO (International Standards Organization) to provide a standard framework in order to describe the protocol stacks in a computer network. Its consists of seven layers where each layer is intended to perform a well-deﬁned function [1]. These are physical layer, data link layer, network layer, transport layer, session layer, presentation layer, and application layer. The OSI model only speciﬁes what each layer should do; it does not specify the exact services and protocols to be used in each layer. Although not implemented in current systems, the OSI model philosophy (i.e., the layering concept) lays a strong foundation for further developement in computer networking.

1

2

For the sake of illustration only four layers are shown. In the real world systems, the number of layers may vary, depending on the functionality and objectives of the networks. A peering host of a source and a destination are the destination and the source, respectively.

4

1 Simulation of Computer Networks

The TCP (Transmission Control Protocol)/IP (Internet Protocol) reference model, which is based on the two primary protocols, namely, TCP and IP, is used in the current Internet. These protocols have proven very powerful, and as a result, have experienced widespread use and implementation in the existing computer networks. It was developed for ARPANET, a research network sponsored by the U.S. Department of Defense, which is considered as the grandparent of all computer networks. In the TCP/IP model, the protocol stack consists of ﬁve layers – physical, data link, network, transport, and application – each of which is responsible for certain services as will be discussed shortly. Note that the application layer in the TCP/IP model can be considered as the combination of session, presentation, and application layers of the OSI model. Application Layer The application layer sits on top of the stack, and uses services from the transport layer (discussed below). This layer supports several higher-level protocols such as HTTP (Hypertext Transfer Protocol) for World Wide Web applications, SMTP (Simple Mail Transfer Protocol) for electronic mail, TELNET for remote virtual terminal, DNS (Domain Name Service) for mapping comprehensible host names to their network addresses, and FTP (File Transfer Protocol) for ﬁle transfer. Transport Layer The objective of a transport layer is to transport the messages from the application layer of the source host to that of the destination host. To accomplish this goal, two well-known protocols, namely, TCP and UDP (User Datagram Protocol), are deﬁned in this layer. While TCP is responsible for a reliable and connection-oriented communication between the two hosts, UDP supports an unreliable connectionless transport. TCP is ideal for applications that prefer accuracy over prompt delivery and the reverse is true for UDP. Generally, control information related to ﬂow control and error control need to be embedded into the messages. Also, before adding any header, fragmentation is usually performed to break a long message into segments. For this reason, the protocol data units in this layer are normally called segments. Network Layer This layer provides routing services to the transport layer. Network layer is designed to deliver the data units, usually called packets, along the paths they are meant to traverse from a source host to a destination host. Again, to facilitate routing, headers containing information such as source and destination network addresses are added to the transport protocol data units to formulates network-layer data unit.

1.2 System Modeling

5

Link Layer The packets are generally routed through several communication links and nodes before they actually reach the destination node. To successfully route these packets all the way to the destination, a mechanism is required for nodeto-node delivery across each of the communication links. A link layer protocol is responsible for data delivery across a communication link. A link layer protocol has three main responsibilities. First, ﬂow control regulates the transmission speed in a communication link. Secondly, error control ensures the integrity of data transmission. Thirdly, ﬂow multiplexing/demultiplexing combines multiple data ﬂows into and extracts data ﬂows from a communication link. Choices of link layer protocols may vary from host to host and network to network. Examples of widely-used link layer protocols/technologies include Ethernet, Point-to-Point Protocol (PPP), IEEE 802.11 (i.e., WiFi), and Asynchronous Transfer Mode (ATM). Physical Layer The physical layer deals with the transmission of data bits across a communication link. Its primary goal is to ensure that the transmission parameters (e.g., transmission power, modulation scheme) are set appropriately to achieve the required transmission performance (e.g., to achieve the target bit error rate performance). Finally, we point out that the ﬁve layers discussed above are common to the OSI layer. As has already been mentioned, the OSI model contains two other layers sitting on top of the transport layer, namely, session and presentation layers. The session layer simply allows users on diﬀerent computers to create communication sessions among themselves. The presentation layer basically takes care of diﬀerent data presentations existing across the network. For example, a uniﬁed network management system gathers data with diﬀerent format from diﬀerent computers and converts their format into a uniform format.

1.2 System Modeling
System modeling refers to an act of representing an actual system in a simply way. System modeling is extremely important in system design and development, since it gives an idea of how the system would perform if actually implemented. With modeling, the parameters of the system can be changed, tested, and analyzed. More importantly, modeling, if properly handled, can save costs in system development. To model a system, some simplifying assumptions are often required. It is important to note that too many assumptions would simplify the modeling but may lead to an inaccurate representation of the system.

6

1 Simulation of Computer Networks

Traditionally, there are two modeling approaches: analytical approach and simulation approach. 1.2.1 Analytical Approach The general concept of analytical modeling approach is to ﬁrst come up with a way to describe a system mathematically with the help of applied mathematical tools such as queuing and probability theories, and then apply numerical methods to gain insight from the developed mathematical model. When the system is simple and relatively small, analytical modeling would be preferable (over simulation). In this case, the model tends to be mathematically tractable. The numerical solutions to this model in eﬀect require lightweight computational eﬀorts. If properly employed, analytical modeling can be cost-eﬀective and can provide an abstract view of the components interacting with one another in the system. However, if many simplifying assumptions on the system are made during the modeling process, analytical models may not give an accurate representation of the real system. 1.2.2 Simulation Approach Simulation is widely-used in system modeling for applications ranging from engineering research, business analysis, manufacturing planning, and biological science experimentation, just to name a few. Compared to analytical modeling, simulation usually requires less abstraction in the model (i.e., fewer simplifying assumptions) since almost every possible detail of the speciﬁcations of the system can be put into the simulation model to best describe the actual system. When the system is rather large and complex, a straightforward mathematical formulation may not be feasible. In this case, the simulation approach is usually preferred to the analytical approach. In common with analytical modeling, simulation modeling may leave out some details, since too much details may result in an unmanageable simulation and substantial computation eﬀort. It is important to carefully consider a measure under consideration and not to include irrelevant detail into the simulation. In the next section, we describe the basic concepts of simulation in more detail with particular emphasis on simulation of a computer network.

1.3 Basics of Computer Network Simulation
A simulation is, more or less, a combination of art and science. That is, while the expertise in computer programming and the applied mathematical tools account for the science part, the very skill in analysis and conceptual model

1.3 Basics of Computer Network Simulation

7

formulation usually represents the art portion. A long list of steps in executing a simulation process, as given in [2], seems to reﬂect this popular claim. Basically, all these steps can be put into three main tasks each of which carries diﬀerent degrees of importance. According to Shannon [2], it is recommended that 40 percent of time and eﬀort be spent on deﬁning a problem, designing a corresponding model, and devising a set of experiments to be performed on the simulation model. Further, it was pointed out that a portion of 20 percent should be used to program the conceptual elements obtained during the ﬁrst step. Finally, the remaining 40 percent should be utilized in verifying/validating the simulation model, experimenting with designed inputs (and possibly ﬁne-tuning the experiments themeselves), and analyzing the results. We note that this formula is in no way a strict one. Any actual simulation may require more or less time and eﬀort, depending on the context of interest and, deﬁnitely, on the modeler himself/herself. A simulation can be thought of as a ﬂow process of network entities (e.g., nodes, packets). As these entities move through the system, they interact with other entities, join certain activities, trigger events, cause some changes to the state of the system, and leave the process. From time to time, they contend or wait for some type of resources. This implies that there must be a logical execution sequence to cause all these actions to happen in a comprehensible and manageable way. An execution sequence plays an important role in supervising a simulation and is sometimes used to characterize the types of simulation (see Section 1.4). 1.3.1 Simulation: The Formal Deﬁnition According to Shannon [2], simulation is “the process of designing a model of a real system and conducting experiments with this model for the purpose of understanding the behavior of the system and/or evaluating various strategies for the operation of the system.” With the dynamic nature of computer networks, we thus actually deal with a dynamic model of a real dynamic system. 1.3.2 Elements of Simulation According to Ingalls [3], the structural components of a simulation consist of the following: Entities Entities are objects which interact with one another in a simulation program to cause some changes to the state of the system. In the context of a computer network, entities may include computer nodes, packets, ﬂows of packets, or non-physical objects such as simulation clocks. To distinguish the diﬀerent entities, unique attributes are assigned to each of them. For instance, a packet entity may have attributes such as packet length, sequence number, priority, and the header.

8

1 Simulation of Computer Networks

Resources Resources are a part of complex systems. In general, a limited supply of resources has to be shared among a certain set of entities. This is usually the case for computer networks, where bandwidth, air time, the number of servers, for instance, represent network resources which have to be shared among the network entities. Activities and Events From time to time, entities engage in some activities. This engaging creates events and triggers changes in the system states. Common examples of activities include delay and queuing. When a computer needs to send a packet but ﬁnd the medium busy, it waits until the medium is free. In this case, the packet is to be sent over the air but the medium is busy, the packet is said to be engaged in a waiting activity. Scheduler A scheduler maintains the list of events and their execution time. During a simulation, it runs a simulation clock creates events, and executes them. Global Variables In simulation, a global variable is accessible by any function or entity in the system, and basically keeps track of some common values of the simulation. In the context of computer networks, such variables might represent, for example, the length of the packet queue in a single-server network, the total busy air time of the wireless network, or the total number of packets transmitted. Random Number Generator A Random number generator (RNG) is required to introduce randomness in a simulation model. Random numbers are generated by sequentially picking numbers from a deterministic sequence of psudo-random number [4], yet the numbers picked from this sequence appear to be random. In most case, a psudo-random sequence is predeﬁned and is used by every RNG. In many situations, several statistically results are required. An RNG needs to start picking numbers from diﬀerent location (i.e., seed) in the (same) predeﬁned psudo-random sequence. Otherwise, the results for every run would be the same. In an actual implementation, an RNG is initialized with a seed. A seed identiﬁes the starting location in a psudo-random sequence, where an RNG starts picking numbers. Diﬀerent simulation initialized with diﬀerent seeds therefore generates diﬀerent results (but statistically identical). In a computer network simulation, for example, a packet arrival process, waiting process, and service process are usually modeled as random processes.

1.4 Time-Dependent Simulation

9

A random process is expressed by sequences of random variables. These random process are usually implemented with the aids of an RNG. For a comprehensive treatment on random process implementation (e.g., those having the uniform, exponential, Gaussian, Poisson, Binomial distribution functions), the readers are referred to [5, 6]. Statistics Gatherer The main responsibility of a statistics gatherer is to collect data generated by the simulation so that meaningful inferences can be drawn from such data.

1.4 Time-Dependent Simulation
A main type of simulation is time dependent simulation which proceeds chronologically. This type of simulation maintains a simulation clock which keeps track of the current simulation time. In most cases, the simulation is run until the clock reaches a predeﬁned threshold. Time-dependent simulation can be further divided into time-driven simulation and event-driven simulation. A time-driven simulation induces and executes events for every ﬁxed time interval. In other words, the simulation advances from one time interval to another, and executes events (if any) until it reaches a certain limit. An event-driven simulation, on the other hand, induces events at arbitrary time. The simulation moves from one event to another, and again executes the event (if any) until the simulation terminates. There is an important note for time-dependent simulation: The simulation must progress in a chronological order. While this note is fairly straightforward for a time-driven simulation, [7] speciﬁes two important points for an implementation of event-driven simulation. First, every new event scheduled into the event list must be tagged with a timestamp equal to or greater than that of the current event. In other words, no outdated events can be scheduled. Secondly, the next event the simulation always executes is that event with the smallest timestamp in the event list. It will never jump over chronologically ordered events or Jump back to the past event. 1.4.1 Time-Driven Simulation In time-driven simulations, the simulation clock is advanced exactly by a ﬁxed interval of Δ time units. After each advancement of the clock, the simulation looks for events that may have occurred during this ﬁxed interval. If so, such events are treated as if they occurred at the end of this interval. Figure 1.2 shows the basic idea behind time advancement in a time-driven simulation. The curved arrows here represent such advances, and a, b, and c mark the occurrences of particular events. During the ﬁrst interval, no event occurs, whereas the second interval contains event a, which is not handled

10

1 Simulation of Computer Networks

until the end of the interval. One disadvantage of time-driven simulation is illustrated in the ﬁfth interval, where events b and c are considered to occur exactly at the end of the interval (at time 5Δ). This calls for a procedure that determines which event should be handled ﬁrst. One solution to get around this situation is to narrow down a simulation time interval such that every interval contains only one event. This, however, puts substantial computational burden on the simulator. Time-driven simulation is therefore not recommended for system models whose events tend to occur over a random period of time.
2 3 4 6 b c

a

0

time

5

Fig. 1.2. Clock advancement in a time-driven simulation.

Example 1.1. Program 1.1 shows time-driven simulation pseudo codes. Lines 1 and 2 initializes the system state variables and the simulation clock, respectively. Line 3 speciﬁes the stopping criterion. Here, Lines 4-7 are run as long as the simulation clock (i.e., simClock) is less than a predeﬁned threshold (i.e., stopTime). These lines collect statistics, executes events, and advance the simulation to the current event time. Program 1.1 Skeleton of the event-processing loop in a time-driven simulation.
1 2 3 4 5 6 7 8 initialize {system states} SimClock := startTime; while {SimClock < stopTime} collect statistics from current state; execute all events that occurred during [SimClock, SimClock + step]; SimClock := SimClock + step; end while

1.4.2 Event-Driven Simulation As the name suggests, an event-driven simulation is initiated and run by a set of events. A list of all scheduled events are usually maintained and updated throughout the simulation process. Technically speaking, the main loop in the simulation program actually has to sequence through this list, and handle one event after another until either the list is empty or the stopping criterion is

1.4 Time-Dependent Simulation

11

To the next event

0

a

b

c

time

Fig. 1.3. Clock advancement in an event-driven simulation.

satisﬁed. The mechanism of handling events is shown graphically in Fig. 1.3, where events a, b, and c are executed in order. The time gap between two events is not ﬁxed. The simulation advance from one event to another, as opposed to one interval to another in a time-driven simulation. Except for the time advancing mechanism, the event-driven simulation is quite similar to the time-driven mechanism. In an event-driven simulation, all the events in an entire simulation may not be created at the initialization. As the simulation advances, one event may induce one or more events. The new event is usually inserted into the chain (i.e., list) of events arranged chronologically. An event-driven simulation ignores the intervals of inactivity by advancing the simulation clock from one event time to another. This process goes on and on until all the events are executed, or until the system reaches a speciﬁc state (e.g., the simulation time reaches a predeﬁned value). Along the way, we certainly need a way to gather some statistics or states of the system for analysis purposes. This process of gathering information can take place right after every event execution. Alternatively, it can be done using a specialized entity which gathers statistics during the simulation. Example 1.2. Program 1.2 shows the skeleton of a typical event-driven simulation program. Lines 1 and 2 initializes the system state variables and the list of events, respectively. Line 3 speciﬁes the stopping criterion. Lines 4–6 are executed as along as Line 3 returns true. Here, the previously executed event is removed from the list, the simulation clock is set to the scheduled time of the current event, and the current event is executed. Within such a loop, the system state variables may be modiﬁed to capture those changes that occur in the system according to the executed event. Program 1.2 Skeleton of the event-processing loop in an event-driven simulation.
1 2 3 4 5 6 7 initialize {system states} initialize {list of events} while {state != finalState} % or while {this.event != Null} expunge the previous event from list of events; set SimClock := time of current event; execute this.event end while

12

1 Simulation of Computer Networks

1.5 A Simulation Example: A Single-Channel Queuing System
This section demonstrates a simulation of a single-channel queuing system, as an example. Consider a point-to-point wired communication link as shown in Fig. 1.4. For simplicity, we consider only a one-way communication from node A to node B. In particular, we are interested in an intra-station packet queuing system at node A, where a packet is retrieved from the queue and transmitted (or served) one at a time – the transmission time depends on the bandwidth or capacity of the link. Futhermore, we assume that packets, whose inter-arrival time follows some probability distribution, are unlimited and randomly generated from a set of applications. Since a packet can be of any random length and the conditions of the channel may vary, the service time of each packet is also random and follows some probability distribution. In our case, it is deﬁned as the elapsed time from the moment a packet is transmitted to the moment it is successfully received by node B. Next, the queuing discipline employed at node A is First-In-First-Out (FIFO), i.e., packets are enqueued and transmitted (served) in the order of their arrival. For simplicity, the queuing mechanism at node B is ignored. Additionally, for the system to be stable, we assume that the arrival rate is less than the service rate. Otherwise, the queue will build up with no bound. Entities The primary entities in this simulation include the following: • Server (medium availability) with idle and busy attributes,
NODE A APP APP APP NODE B

Buffer

FIFO Channel

Fig. 1.4. Illustration of a single-channel queuing system.

1.5 A Simulation Example: A Single-Channel Queuing System

13

• Packets with arrival time and service time attributes, and • Queue with empty and non-empty attributes. Resource Obviously, the only resource in this example is the transmission time in the channel. System State Variables and Events • Two system state variables: (i) num_system is the number of packets in the system, i.e., the one being served and those waiting in the queue. (ii) channel_free is the status of the channel (server) which is either idle or busy. • Two events: (i) pkt_arrival corresponds to a packet arrival event. This event occurs when a packet arrives at the queue. As shown in Fig. 1.5, once entered, the packet may either go directly to service or wait in the queue, depending on whether the channel is busy or idle. (ii) pkt_complete corresponds to a successful packet transmission event. This event indicates that a packet has been received successfully by node B. At the completion, node A begins to transmit (serve) another packet waiting in the queue. If there is no more packet to be sent, the channel becomes idle. The ﬂow diagram of such a process is shown in Fig. 1.6.

Packet Arrival Event

Transmit Packet

yes

Channel Idle?

no

Enqueue Packet

Fig. 1.5. Packet arrival event.

Successful Packet Transmission

Begin Channel Idle Time

yes

Queue Empty?

no

Dequeue and transmit a packet from the buffer

Fig. 1.6. Successful packet transmission (service completion) event.

14

1 Simulation of Computer Networks

Two other important elements in an event-driven simulation are a simulation clock and an event list. A simulation clock maintains the current simulation time, as the simulation advances. An event list is a chain of scheduled events (e.g., packet arrival and successful packet transmission) connecting in a chronological order. Again, the simulation executes an event after another down the event list, and updates the simulation clock based on the time speciﬁed in the executed event. Simulation Performance Measures Here, we consider three following performance measures are: • Mean waiting time is the average time that a packet spends in the queue. In the simulation, we deﬁne a global variable which keeps track of the total time all the transmitted packets spent in the queue. At the end of the simulation, we divide this value by the total number of packets transmitted to obtain the mean waiting time. • Mean packet transmission latency is the average time that a packet spends (from its arrival to its departure) in the system. It is the total time of all the packets spend the system divided with the total number of transmitted packets. • Mean server utilization is the percentage time where the server is busy. During the simulation we measure the time where the server is busy. At the end of the simulation, we divide this busy time by the total simulation time, and obtain the mean server utilization. It is important to note that all the above measures are the average values taken over time, implying that the longer the simulation, the more accurate the statistics. Program 1.3 shows a skeleton of the simulation program that can be used to implement the single-channel queuing system described above. The program starts with the initialization of system state variables as deﬁned above. Additionally, we deﬁne num_queue (Line 3) and num_system (Line 4) to store the number of waiting packets and the number of all packets currently in the system (i.e., both the queue and the channel), respectively. The variable SimClock is also initialized to zero at the beginning of the simulation. Next, Line 7 creates an event list by invoking the procedure create_list(). We assume that this function automatically generates packets and associates each packet with the random inter-arrival and service times. Further we assume that the event_list here is implemented using some appropriate data structure that usually indicates the event type (arrival or completion) and the associated timestamp (i.e., either inter-arrival time and service time). Initially, only the arrival events are put into the event_list. Now we deﬁne a main loop which continuously checks whether the simulation should be terminated. The stopping criteria in Line 9 are (1) the event list is exhausted and (2) the simulation clock has reached a predeﬁned threshold.

1.5 A Simulation Example: A Single-Channel Queuing System

15

Program 1.3 Simulation skeleton of a single-channel queuing system.
1 2 3 4 5 6 7 % Initialize system states channel_free = true; %Channel is idle num_queue = 0; %Number of packets in queue num_system = 0; %Number of packets in system SimClock = 0; %Current time of simulation %Generate packets and schedule their arrivals event_list = create_list();

8 % Main loop 9 while {event_list != empty} & {SimClock < stopTime} 10 expunge the previous event from event list; 11 set SimClock := time of current event; 12 call current event; 13 end while 14 %Define events 15 pkt_arrival(){ 16 if(channel_free) 17 channel_free = false; 18 num_system = num_system + 1; 19 % Update "event_list": Put "successful packet tx event" 20 % into "event_list," T is random service time. 21 schedule event "pkt_complete" at SimClock + T; 22 else 23 num_queue = num_queue + 1; %Place packet in queue 24 num_system = num_queue + 1; 25 } 26 pkt_complete(){ 27 num_system = num_system - 1; 28 num_queue = num_queue - 1; 29 if(num_queue > 0) 30 schedule event "pkt_complete" at SimClock + T; 31 else 32 channel_free = true; 33 num_system = 0; 34 num_queue = 0; 35 }

If not, Lines 10–12 keep on executing the next event by invoking either the procedure pkt_arrival() in Lines 15–25 or the procedure pkt_complete() in Lines 26–35. The procedure pkt_arrival() (Lines 15–25) checks whether the channel (server) is idle when a packet arrives. If it is idle, the channel is set to busy,

16

1 Simulation of Computer Networks

and a successful packet transmission event is inserted into the event_list for future execution. The timestamp associated with the event is equal to the current clock time (SimClock) plus the packet’s randomly generated service time (T). If the channel is busy, on the other hand, the packet is simply put in the queue whose counter (num_queue) is incremented by one unit. The number of packets in the system is also updated accordingly. When SimClock advances to a successful packet transmission event, the procedure pkt_complete() is executed (Lines 26–35). Here, the number of packets in the system (num_system) is updated. The queue counter num_queue is decremented by one unit. Upon any successful packet transmission, it is also necessary to check whether the queue is empty. If not, the head-of-the-line packet will be served. This is done by feeding the packet to the channel and scheduling it for transmission completion at time SimClock + T. However, if the queue is empty, the channel is set to idle and the numbers of packets in the queue and system are set to zero. Suppose that the inter-arrival time and the service time comply with the probability mass functions speciﬁed in Table 1.1. Table 1.2 shows the simulation results for 10 packets. The inter-arrival time and service time of each packet are shown in the ﬁrst and second columns, respectively. The third and fourth columns specify the time where the packet arrives and starts to be served. The ﬁfth column represents the packet waiting time, the time that a packet spends in the queue. It is computed as the time diﬀerence between when the service starts and when the packet arrives. Finally, the sixth column represents the packet transmission latency, the time that a packet spends in both the queue and the channel. It is computed as the summation of the waiting time and the service time. Based on the result in Table 1.2, we compute the average waiting time and the average packet transmission latency by averaging the sixth and seventh columns (i.e., adding all the values and dividing the result by 10). The average waiting time and the average packet transmission latency are therefore 1.0 and 3.5 time units, respectively.

Table 1.1. Probability mass functions of inter-arrival time and service time. Time unit 1 2 3 4 5 6 7 Inter-arrival (probability mass) 0.2 0.2 0.2 0.2 0.1 0.05 0.05 Service (probability mass) 0.5 0.3 0.1 0.05 0.05

1.5 A Simulation Example: A Single-Channel Queuing System Table 1.2. Simulation of a single-channel queuing system. Packet Interarr. time Service time Arrival time Service starts Time spent in-queue 0 3 3 3 0 0 0 0 1 0 10 Packet transmission latency 5 7 4 4 3 1 1 4 4 2 3.5

17

1 2 3 4 5 6 7 8 9 10

2 4 1 6 7 2 1 3 5

5 4 1 1 3 1 1 4 3 2

0 2 6 7 13 20 22 23 26 31

0 5 9 10 13 20 22 23 27 31

Table 1.3. Evolution of number of packets in the queue over time. Event Arrival Arrival Completion Arrival Arrival Completion Completion Completion Arrival Completion Packet No. 1 2 1 3 4 2 3 4 5 5 Simulation clock 0 2 5 6 7 9 10 11 13 16

Based on the information in Table 1.2, we also show in Table 1.3 how each event for the ﬁrst ﬁve packets occurs in a chronological order with respect to the Simulation Clock (SimClock). Figure 1.7 depicts the evolution of the number of packets in the queue over time, which is also shown in Table 1.3. As shown in Fig. 1.7, at various instances, the number of packets in the system diﬀers. When the ﬁrst packet is being transmitted, another packet arrives in the queue at time 2. The number of packets in the system becomes 2. That is, it includes the one that has been served plus the one just arrived. In Fig. 1.7, this event causes a jump in the graph at time 2. At time 5, when the ﬁrst packet is successfully received at

18

1 Simulation of Computer Networks
Number of Packets in the System 3 2 1 2 4 6 8 10 12 14 16

Simulation time

Fig. 1.7. Number of packets in the system at various instances.

node B, the next packet in queue is transmitted. Therefore, the graph drops to level 1, indicating that there is only one packet in the system. This dynamics continues until all the packets are transmitted. Based on Fig. 1.7, the mean server utilization can be computed from the ratio of the time where the server is in use and the simulation time, which is 14/16 = 0.875 in this case.

1.6 Chapter Summary
A computer network is a complex system that requires a careful treatment in design and implementation. Simulation, regarded as one of the most powerful performance analysis tools, is usually used in carrying out such a treatment to complement the analytical tools. This chapter focuses mainly on time-dependent simulation, which advances in a time domain. The time-dependent simulation can be divided into two categories. Time-driven simulation advances the simulation by ﬁxed time intervals, while event-driven simulation proceeds from one event to another. NS2 is an event-driven simulation tool. Designing event-driven simulation models using NS2 is the theme of the rest of the book.

2 Introduction to Network Simulator 2 (NS2)

2.1 Introduction
Network Simulator (Version 2), widely known as NS2, is simply an eventdriven simulation tool that has proved useful in studying the dynamic nature of communication networks. Simulation of wired as well as wireless network functions and protocols (e.g., routing algorithms, TCP, UDP) can be done using NS2. In general, NS2 provides users with a way of specifying such network protocols and simulating their corresponding behaviors. Due to its ﬂexibility and modular nature, NS2 has gained constant popularity in the networking research community since its birth in 1989. Ever since, several revolutions and revisions have marked the growing maturity of the tool, thanks to substantial contributions from the players in the ﬁeld. Among these are the University of California and Cornell University who developed the REAL network simulator,1 the foundation which NS is based on. Since 1995 the Defense Advanced Research Projects Agency (DARPA) supported development of NS through the Virtual InterNetwork Testbed (VINT) project [9].2 Currently the National Science Foundation (NSF) has joined the ride in development. Last but not the least, the group of researchers and developers in the community are constantly working to keep NS2 strong and versatile. Again, the main objective of this book is to provide the readers with insights into the NS2 architecture. This chapter gives a brief introduction to NS2. NS2 Beginners are recommended to go thorough the detailed introductory online resources. For example, NS2 oﬃcial website [10] provides NS2 source code as well as detailed installation instruction. The web pages in [11] and [12] are among highly recommended ones which provide tutorial and
1

2

REAL was originally implemented as a tool for studying the dynamic behavior of ﬂow and congestion control schemes in packet-switched data networks. Funded by DARPA, the VINT project aimed at creating a network simulator that will initiate the study of diﬀerent protocols for communication networking.

T. Issariyakul, E. Hossain, Introduction to Network Simulator NS2, DOI: 10.1007/978-0-387-71760-9 2, c Springer Science+Business Media, LLC 2009

20

2 Introduction to Network Simulator 2

examples for setting up basic NS2 simulation. A comprehensive list of NS2 codes contributed by researchers can be found in [13]. These introductory online resources would be helpful in understanding the material presented in this book. In this chapter an introduction to NS2 is provided. In particular, Section 2.2 presents the basic architecture of NS2. The information on NS2 installation is given in Section 2.3. Section 2.4 shows NS2 directories and conventions. Section 2.5 shows the main steps in NS2 simulation. A simple simulation example is given in Section 2.6. Section 2.7 describes how to include C++ modules in NS2. Finally, Section 2.8 concludes the chapter.

2.2 Basic Architecture
Figure 2.1 shows the basic architecture of NS2. NS2 provides users with an executable command ns which takes on input argument, the name of a Tcl simulation scripting ﬁle. Users are feeding the name of a Tcl simulation script (which sets up a simulation) as an input argument of an NS2 executable command ns. In most cases, a simulation trace ﬁle is created, and is used to plot graph and/or to create animation. NS2 consists of two key languages: C++ and Object-oriented Tool Command Language (OTcl). While the C++ deﬁnes the internal mechanism (i.e., a backend) of the simulation objects, the OTcl sets up simulation by assembling and conﬁguring the objects as well as scheduling discrete events (i.e., a frontend). The C++ and the OTcl are linked together using TclCL. Mapped to a C++ object, variables in the OTcl domains are sometimes referred to as handles. Conceptually, a handle (e.g., n as a Node handle) is just a string (e.g., _o10) in the OTcl domain, and does not contain any functionality. Instead, the functionality (e.g., receiving a packet) is deﬁned in the mapped C++ object (e.g., of class Connector). In the OTcl domain, a handle acts as a frontend which interacts with users and other OTcl objects. It may deﬁnes its own procedures and variables to facilitate the interaction. Note that the member procedures and variables in the OTcl domain are called instance procedures

Tcl Simulation Script

Simulation Objects C++

TclCL

Simulation Objects OTcl

Simulation Trace File

NS2 Shell Executable Command (ns)

NAM (Animation)

Xgraph (Plotting)

Fig. 2.1. Basic architecture of NS.

2.3 Installation

21

(instprocs) and instance variables (instvars), respectively. Before proceeding further, the readers are encouraged to learn C++ and OTcl languages. We refer the readers to [14] for the detail of C++, while a brief tutorial of Tcl and OTcl tutorial are given in Appendices A.1 and A.2, respectively. NS2 provides a large number of built-in C++ objects. It is advisable to use these C++ objects to set up a simulation using a Tcl simulation script. However, advance users may ﬁnd these objects insuﬃcient. They need to develop their own C++ objects, and use a OTcl conﬁguration interface to put together these objects. After simulation, NS2 outputs either text-based or animation-based simulation results. To interpret these results graphically and interactively, tools such as NAM (Network AniMator) and XGraph are used. To analyze a particular behavior of the network, users can extract a relevant subset of text-based data and transform it to a more conceivable presentation.

2.3 Installation
NS2 is a free simulation tool, which can be obtained from [9]. It runs on various platforms including UNIX (or Linux), Windows, and Mac systems. Being developed in the Unix environment, with no surprise, NS2 has the smoothest ride there, and so does its installation. Unless otherwise speciﬁed, the discussion in this book is based on a Cygwin (UNIX emulator) activated Windows system. NS2 source codes are distributed in two forms: the all-in-one suite and the component-wise. With the all-in-one package, users get all the required components along with some optional components. This is basically a recommended choice for the beginners. This package provides an “install” script which conﬁgures the NS2 environment and creates NS2 executable ﬁle using the “make” utility. The current all-in-one suite consists of the following main components: • • • • NS release 2.30, Tcl/Tk release 8.4.13, OTcl release 1.12, and TclCL release 1.18.

and the following are the optional components: • NAM release 1.12: NAM is an animation tool for viewing network simulation traces and packet traces. • Zlib version 1.2.3: This is the required library for NAM. • Xgraph version 12.1: This is a data plotter with interactive buttons for panning, zooming, printing, and selecting display options. The idea of the component-wise approach is to obtain the above pieces and install them individually. This option save considerable amount of downloading

22

2 Introduction to Network Simulator 2

time and memory space. However, it could be troublesome for the beginners, and is therefore recommended only for experienced users. 2.3.1 Installing an All-In-One NS2 Suite on Unix-Based Systems The all-in-one suite can be installed in the Unix-based machines by simply running the install script and following the instructions therein. The only requirement is a computer with a C++ compiler installed. The following commands show how the all-in-one NS2 suite can be installed and validated, respectively: shell>./install shell>./validate Validating NS2 involves simply running a number of working scripts that verify the essential functionalities of the installed components. 2.3.2 Installing an All-In-One NS2 Suite on Windows-Based Systems To run NS2 on Windows-based operating systems, a bit of tweaking is required. Basically, the idea is to make Windows-based machines emulate the functionality of the Unix-like environment. A popular program that performs this job is Cygwin.3 After getting Cygwin to work, the same procedure as that of Unix-based installation can be followed. For ease of installation, it is recommended that the all-in-one package be used. The detailed description of Windows-based installation can be found online at NS2’s Wiki site [9], where the information on post-installation troubles can also be found. Note that by default Cygwin does not install all packages neccessary to run NS2. A user needs to manually install the addition packages shown in Table 2.14 .

Table 2.1. Additional Cygwin packages required to run NS2. Category Development Utils X11 Packages gcc, gcc-objc, gcc-g++, make patch xorg-x11-base, xorg-x11-devel

3

4

Cygwin is available online and comes free. Information such as how to obtain and install Cygwin is available online at the Cygwin website (www.cygwin.com). Diﬀerent versions may install diﬀerent default packages. Users may need to install more or less packages depending on the version of Cygwin.

2.4 Directories and Convention

23

2.4 Directories and Convention
2.4.1 Directories Suppose that NS2 is installed in directory nsallinone-2.30. Figure 2.2 shows the directory structure under directory nsallinone-2.30. Here, directory nsallinone-2.30 is on the Level 1. On the Level 2, directory tclcl-1.18 contains classes in TclCL (e.g., Tcl, TclObject, TclClass). All NS2 simulation modules are in directory ns-2.30 on the Level 2. Hereafter, we will refer to directories ns-2.30 and tclcl-1.18 as ˜ns/ and ˜tclcl /, respectively.

LEVEL 1 All NS2 simulation modules

nsallinone-2.30

... TclCL classes ns-2.30 tcl8.4.13 tclcl-1.18

LEVEL 2

...
LEVEL 3 common tools tcp queue trace tcl Modules in the ... interpreted hierarchy

LEVEL 4

Commonly-used modules in the interpreted hierarchy

lib

rtglib

Fig. 2.2. Directory structure of NS2 [12].

On Level 3, the modules in the interpreted hierarchy are under directory tcl. Among these modules, the frequently-used ones (e.g., ns-lib.tcl, ns-node.tcl, ns-link.tcl) are stored under directory lib on Level 4. Simulation modules in the compiled hierarchy are classiﬁed in directories on Level 2. For example, directory tools contains various helper classes such as random variable generators. Directory common contains basic modules related to packet forwarding such as the simulator, the scheduler, connector, packet. Directories queue, tcp, and trace contain modules for queue, TCP (Transmission Control Protocol), and tracing, respectively.

2.4.2 Convention The terminologies and formats which are used in NS2 and in this book hereafter are shown below:

24

2 Introduction to Network Simulator 2

Terminology • An NS2 simulation script (e.g., myfirst_ns.tcl) is referred to as a Tcl simulation script . • C++ and OTcl class hierarchies, which have one-to-one correspondence, are referred to as the compiled hierarchy and the interpreted hierarchy, respectively. Class (or member) variables and class (or member) functions are the variables and functions which belong to a class. In the compiled hierarchy, they are referred to simply as variables and functions, respectively. Those in the interpreted hierarchy are referred to as instance variables (instvars) and instance procedures (instprocs), respectively. As we will see in Section 3.4.4, command, is a special instance procedure, whose implementation is in the compiled hierarchy (i.e., written in C++). An OTcl object is, therefore, associated with instance variables, instance procedures, and commands, while a C++ object is associated with variables and functions. • Despite their minor diﬀerences, the terms “OTcl” and “interpreted” are used interchangeably throughout the book. Likewise, “C++” and “compiled” are used interchangeably. These terms can be used as adjectives to indicate the domain under consideration. For example, both OTcl variables and interpreted variables refer to variables in the interpreted hierarchy. Similarly, both C++ functions and compiled functions refer to functions in the compiled hierarchy. Also, we will refer to the C++ compiler and the OTcl interpreter simply as the compiler and the interpreter, respectively. • A “MyClass” object is a shorthand for an object of class MyClass. A “MyClass” pointer is a shorthand for a pointer which points to an object of class MyClass. For example, based on the statements “Queue q” and “Packet* p”, “q” and “p” are said to be a “Queue” object and a “Packet pointer”, respectively. Also, suppose further that class DerivedClass and AnotherClass derive from class MyClass. Then, the term a MyClass object refers to any object which is instantiated from class MyClass or its derived classes (i.e., DerivedClass or AnotherClass). • Objects and instances are instantiated from a C++ class and an OTcl class, respectively. However, the book uses these two terms interchangeably. • NS2 consists of two languages. Suppose that objects “A” and “B” are written in each language and correspond to one another. Then, “A” is said to be the shadow object of “B”. Similarly “B” is said to be the shadow object of “A”. • Consider two consecutive nodes in Fig. 3.2. In this conﬁguration, an object (i.e., node) on the left always sends packets to the object on the right. The object on the right is referred to as a downstream object or a target, while the object on the right is referred to as an upstream object. In a general case, an object can have more than one target. However, a packet must be forwarded to one of these targets. From the perspective of an upstream object, a downstream object which receive the packet is also referred to as a forwarding object.

2.4 Directories and Convention

25

Notations • As in C++, we use “::” to indicate the scope of functions and instprocs (e.g., TcpAgent::send(...)). • Most of the texts in this book are written in regular letters. NS2 codes are written in “this font type”. The quotation marks are omitted if it is clear from the context. For example, the Simulator is a general term for the simulating module in NS2, while a Simulator object is an object of class Simulator. • A value contained in a variable is embraced with . For example, if a variable var stores an integer 7, will be 7. • A command prompt or an NS2 prompt is denoted by “>>” at the beginning of a line. • In this book, codes shown in ﬁgures are partially excerpted from NS2 ﬁle. The ﬁle name from which the codes is excerpted is shown in the ﬁrst line of the ﬁgure. For example, the codes in Program 2.1 are from ﬁle “myfirst_ns.tcl”. • A class name may consist of several words. All the words in a class name are capitalized. In the interpreted hierarchy, a derived class is named by having the name of its parent class and a slash character (“/”) as a preﬁx, while that in the compiled hierarchy is named by having the name of its base class as a suﬃx. Examples of NS2 naming convention are given in Table 2.2. • In the interpreted hierarchy, an instproc name is written in lower-case. If the instproc name consists of more than one word, each word except for the ﬁrst one will be capitalized. In the compiled hierarchy, all the words are written in lower case and separated by an underscore “_” (see Table 2.2). • The naming convention for variables is similar to that for functions and instprocs. However, the last character of the names of class variables in both the hierarchies is always an underscore (“_”; see Table 2.2). Note that this convention is only a guideline that a programmer should (but does not have to) follow.

Table 2.2. Examples of NS2 naming convention The interpreted hierarchy Base class Agent Derived class Agent/TCP Derived class (2nd level) Agent/Tcp/Reno Class functions installNext Class variables windowOption_ The compiled hierarchy Agent TcpAgent RenoTcpAgent install_next wnd_option_

26

2 Introduction to Network Simulator 2

Exercise 2.1. Design C++ and OTcl classes (e.g., Class My TCP). Derive this class from the TCP Reno classes shown in Table 2.2. Use the convention deﬁned above to name the class names, variables/instvars, and functions/instprocs in both the domain.

2.5 Running NS2 Simulation
2.5.1 NS2 Program Invocation After the installation and/or recompilation (see Section 2.7), an executable ﬁle ns is created in the NS2 home directory. NS2 can be invoked by executing the following statement from the shell environment: >>ns [] [] where and are optional input argument. If no argument is given, the command will bring up an NS2 environment, where NS2 waits to interpret commands from the standard input (i.e., keyboard) line-by-line. If the ﬁrst input argument is given, NS2 will interpreted the input scripting (i.e., a so-called Tcl simulation script) according to the Tcl syntax. The detail for writing a Tcl scripting ﬁle is given in Appendix A.1. Finally, the input arguments , each separated by a white space, are fed to the Tcl ﬁle . From within the ﬁle , the input argument is stored in the built-in variable argv (see Appendix A.1.1). 2.5.2 Main NS2 Simulation Steps The followings show the three key step guideline in deﬁning a simulation scenario in a NS2: Step 1: Simulation Design The ﬁrst step in simulating a network is to design the simulation. In this step, the users should determine the simulation purposes, network conﬁguration and assumptions, the performance measures, and the type of expected results. Step 2: Conﬁguring and Running Simulation This step implements the design in the ﬁrst step. It consists of two phases: • Network conﬁguration phase: In this phase network components (e.g., node, TCP and UDP) are created and conﬁgured according to the simulation design. Also, the events such as data transfer are scheduled to start at a certain time.

2.6 A Simulation Example

27

• Simulation Phase: This phase starts the simulation which was conﬁgured in the Network Conﬁguration Phase. It maintains the simulation clock and executes events chronologically. This phase usually runs until the simulation clock reached a threshold value speciﬁed in the Network Conﬁguration Phase. In most cases, it is convenient to deﬁne a simulation scenario in a Tcl scripting ﬁle (e.g., ) and feed the ﬁle as an input argument of an NS2 invocation (e.g., executing “ns ”). Step 3: Post Simulation Processing The main tasks in this steps include verifying the integrity of the program and evaluating the performance of the simulated network. While the ﬁrst task is referred to as debugging, the second one is achieved by properly collecting and compiling simulation results (see Chapter 13).

2.6 A Simulation Example
We demonstrate a network simulation through a simple example. Again, a simulation process consists of three steps. Step 1: Simulation Design Figure 2.3 shows the conﬁguration of a network under consideration. The network consists of ﬁve nodes n0 to n4. In this scenario, node n0 sends constantbit-rate (CBR) traﬃc to node n3, and node n1 transfers data to node n4 using
CBR

UDPAgent n0 node Transport agent Application UDP Flow TCP Flow

100 Mbps 5 ms Delay n2

TCPSink 54 Mbps 10 ms Delay n4 10 Mbps 15 ms Delay 54 Mbps 10 ms Delay n3 NullAgent

100 Mbps 5 ms Delay n1 TCPAgent

FTP

Fig. 2.3. A sample network topology.

28

2 Introduction to Network Simulator 2

a ﬁle transfer protocol (FTP). These two carried traﬃc sources are carried by transport layer protocols User Datagram Protocol (UDP) and Transmission Control Protocol (TCP), respectively. In NS2, the transmitting object of these two protocols are a UDP agent and a TCP agent, while the receivers are a Null agent and a TCP sink agent, respectively. Step 2: Conﬁguring and Running Simulation Programs 2.1–2.2 show two portions of a Tcl simulation script which implements the scenario in Fig. 2.3. Consider Program 2.1. This program creates a simulator instance in Line 1. It creates a trace ﬁle and a NAM trace ﬁle in Lines 2–3 and 4–5, respectively. It deﬁnes procedure finish{} in Lines 6–13. Finally, it creates nodes and links them together in Lines 14–18 and 19–24, respectively. The Simulator is created in Line 1 by executing “new Simulator”. The returned Simulator handle is stored in a variable ns. Lines 2 and 4 open ﬁles out.tr and out.nam, respectively, for writing. The variables myTrace and myNAM are the ﬁle handles for these two ﬁles, respectively. Lines 3 and 5 inform NS2 to collect all trace information for a regular trace and a NAM trace, respectively. The procedure finish{} is invoked immediately before the simulation terminates. The keyword global informs the Tcl interpreter that the variables ns, myTrace, myNAM are those deﬁned in the global scope (i.e., deﬁned outside the procedure). Line 8 ﬂushes the buﬀer of the packet tracing variables. Lines 9–10 close the ﬁle associated with handles myTrace and myNAM. Line 11 executes the statement “nam out.nam &” from the shell environment. Finally, Line 12 tells NS2 to exit with code 0. Lines 14–18 creates Nodes using the instproc node of the Simulator whose handle is ns. Lines 19–23 connects each pair of nodes with a bi-directional link using an instproc duplex-link {src dst bw delay qtype} of class Simulator, where src is a beginning node, dst is an terminating node, bw is the link bandwidth, delay is the link propagation delay, and qtype is the type of the queues between the node src and the node dst. Similar to the instproc duplex-link{...}, Line 23 create a uni-directional link using an instproc simplex-link{...} of class Simulator. Finally, Line 24 sets the queue size of the queue between node n2 and node n3 to be 40 packets. Next, consider the second portion of the Tcl simulation script in Program 2.2. A UDP connection, a CBR traﬃc source, a TCP connection, and an FTP session are created and conﬁgured in Lines 25–30, 31–34, 35–40, and 41–42, respectively. Lines 43–47 schedules discrete events. Finally, the simulator is started in Line 48 using the instproc run{} associated with the simulator handle ns. To create a UDP connection, a sender udp and a receiver null are created in Lines 25 and 27, respectively. Taking a node and an agent as input

2.6 A Simulation Example

29

Program 2.1 First NS2 Program
# myfirst_ns.tcl # Create a Simulator 1 set ns [new Simulator] # Create a trace file 2 set mytrace [open out.tr w] 3 $ns trace-all $mytrace # Create a NAM trace file 4 set myNAM [open out.nam w] 5 $ns namtrace-all $myNAM # Define a procedure finish 6 proc finish { } { 7 global ns mytrace myNAM 8 $ns flush-trace 9 close $mytrace 10 close $myNAM 11 exec nam out.nam & 12 exit 0 13 } # Create Nodes 14 set n0 [$ns 15 set n1 [$ns 16 set n2 [$ns 17 set n3 [$ns 18 set n4 [$ns

node] node] node] node] node]

# Connect Nodes with Links 19 $ns duplex-link $n0 $n2 100Mb 5ms 20 $ns duplex-link $n1 $n2 100Mb 5ms 21 $ns duplex-link $n2 $n4 54Mb 10ms 22 $ns duplex-link $n2 $n3 54Mb 10ms 23 $ns simplex-link $n3 $n4 10Mb 15ms 24 $ns queue-limit $n2 $n3 40

DropTail DropTail DropTail DropTail DropTail

argument, an instproc attach-agent{...} of class Simulator in Line 26 attaches a UDP agent udp and a node n0 together. Similarly, Line 28 attaches a Null agent null to a node n3. The instproc connect{from_agt to_agt} in Line 29 informs an agent from_agt to send the generated traﬃc to an agent to_agt. Finally, Line 30 sets the UDP ﬂow ID to be 1. The construction of a TCP connection in Lines 35–40 is similar to that of a UDP connection in Lines 25–30.

30

2 Introduction to Network Simulator 2

Program 2.2 First NS2 Program (Continued)
# 25 26 27 28 29 30 # 31 32 33 34 # 35 36 37 38 39 40 Create a UDP agent set udp [new Agent/UDP] $ns attach-agent $n0 $udp set null [new Agent/Null] $ns attach-agent $n3 $null $ns connect $udp $null $udp set fid_ 1 Create a CBR traffic source set cbr [new Application/Traffic/CBR] $cbr attach-agent $udp $cbr set packetSize_ 1000 $cbr set rate_ 2Mb Create a TCP agent set tcp [new Agent/TCP] $ns attach-agent $n1 $tcp set sink [new Agent/TCPSink] $ns attach-agent $n4 $sink $ns connect $tcp $sink $tcp set fid_ 2

# Create an FTP session 41 set ftp [new Application/FTP] 42 $ftp attach-agent $tcp # 43 44 45 46 47 Schedule events $ns at 0.05 "$ftp start" $ns at 0.1 "$cbr start" $ns at 60.0 "$ftp stop" $ns at 60.5 "$cbr stop" $ns at 61 "finish"

# Start the simulation 48 $ns run

A CBR traﬃc source is created in Line 31. It is attached to a UDP agent udp in Line 32. The packet size and generation rate of the CBR connection are set to 1000 bytes and 2 Mbps, respectively. Similarly, an FTP session handle is created in Line 41 and is attached to a TCP agent tcp in Line 42. In NS2, discrete events can be scheduled using an instproc at of class Simulator, which takes two input arguments: time and str. This instproc schedules an execution of str when the simulation time is time. Lines 43 and 44 start the FTP and CBR traﬃc at 0.05th second and 1st second, respectively. Lines 45 and 46 stop the FTP and CBR traﬃc at 60.0th second

2.6 A Simulation Example

31

and 60.5th second, respectively. Line 47 terminates the simulation by invoking the procedure finish{} at 61st second. Note that the FTP and CBR traﬃc source can be started and stopped by invoking its commands start{} and stop{}, respectively. We run the above simulation script by executing >>ns myfirst_ns.tcl from the shell environment. At the end of simulation, the trace ﬁles should be created and NAM should be running (since it is invoked from within the procedure finish{}). Step 3: Post Simulation Processing–Packet Tracing Packet tracing records the detail of packet ﬂow during a simulation. It can be classiﬁed into a text-based packet tracing and a NAM packet tracing. Text-Based Packet Tracing Text-based packet tracing records the detail of packets passing through network checkpoints (e.g., nodes and queues). A part of the text-based trace obtained by running the above simulation (myfirst_ns.tcl) is shown below. ... + 0.110419 1 2 tcp 1040 ------- 2 1.0 4.0 5 12 + 0.110419 1 2 tcp 1040 ------- 2 1.0 4.0 6 13 - 0.110431 1 2 tcp 1040 ------- 2 1.0 4.0 5 12 - 0.110514 1 2 tcp 1040 ------- 2 1.0 4.0 6 13 r 0.11308 0 2 cbr 1000 ------- 1 0.0 3.0 2 8 + 0.11308 2 3 cbr 1000 ------- 1 0.0 3.0 2 8 - 0.11308 2 3 cbr 1000 ------- 1 0.0 3.0 2 8 r 0.11316 0 2 cbr 1000 ------- 1 0.0 3.0 3 9 + 0.11316 2 3 cbr 1000 ------- 1 0.0 3.0 3 9 - 0.113228 2 3 cbr 1000 ------- 1 0.0 3.0 3 9 r 0.115228 2 3 cbr 1000 ------- 1 0.0 3.0 0 6 r 0.115348 1 2 tcp 1040 ------- 2 1.0 4.0 3 10 + 0.115348 2 4 tcp 1040 ------- 2 1.0 4.0 3 10 - 0.115348 2 4 tcp 1040 ------- 2 1.0 4.0 3 10 r 0.115376 2 3 cbr 1000 ------- 1 0.0 3.0 1 7 r 0.115431 1 2 tcp 1040 ------- 2 1.0 4.0 4 11 ... Figure 2.4 shows the format of each trace line, which consists of 12 columns. The general format of each trace line is shown in Fig. 2.4, where 12 columns make up a complete trace line. The type identiﬁer ﬁeld corresponds to four possible event types that a packet has experienced: r (received), + (enqueued), - (dequeued), and d (dropped). The time ﬁeld denotes the time at which

32

2 Introduction to Network Simulator 2

Fig. 2.4. Format of each line in a normal trace ﬁle.

such event occurs. Fields 3 and 4 are the starting and the terminating nodes, respectively, of the link at which a certain event takes place. Fields 5 and 6 are packet type and packet size, respectively. The next ﬁeld is a series of ﬂags, indicating any abnormal behavior. Note the output "-------" denotes no ﬂag. Following the ﬂags is a packet ﬂow ID. Fields 9 and 10 mark the source and the destination addresses, respectively, in the form of node.port. For correct packet assembly at the destination node, NS also speciﬁes a packet sequence number in the second last ﬁeld. Finally, to keep track of all packets, a packet unique ID is recorded in the last ﬁeld. Now, having this trace at hand would not be useful unless meaningful analysis is performed on the data. In post-simulation analysis, one usually extracts a subset of the data of interest and further analyzes it. For example, the average throughput associated with a speciﬁc link can be computed by extracting only the columns and ﬁelds associated to that link from the trace ﬁle. Two of the most popular languages that facilitate this process are AWK and Perl. The basic structures and usage of these languages are described in Appendix A. Text-based packet tracing is activated by executing “$ns trace-all $file”, where ns is the Simulator handle and file is a handle associated with the ﬁle which stores the tracing text. This statement simply informs NS2 of the need to trace packets. When an object is created, a tracing object is also created to collect the detail of traversing packets. Hence, the “trace-all” statement must be executed prior to object creation. We shall discuss the detail of text-based packet tracing later in Chapter 13. Network AniMation (NAM) Trace NAM trace is records simulation detail in a text ﬁle, and uses the text ﬁle the play back the simulation using animation. NAM trace is activated by the command “$ns namtrace-all $file”, where ns is the Simulator handle and file is a handle associated with the ﬁle (e.g., out.nam in the above example) which stores the NAM trace information. After obtaining a NAM trace ﬁle, the animation can be initiated directly at the command prompt through the following command (See Line 11 in Program 2.2): >>nam filename.nam Many visualization features are available in NAM. These features are for example animating colored packet ﬂows, dragging and dropping nodes (positioning), labeling nodes at a speciﬁed instant, shaping the nodes, coloring a speciﬁc link, and monitoring a queue.

2.7 Including C++ Modules into NS2 and the make Utility

33

2.7 Including C++ Modules into NS2 and the make Utility
In developing an NS2 simulation, very often it is necessary to create the customized C++ modules to complement the existing libraries. As such, the developer is faced with the task of keeping track of all the created ﬁles as a part of NS2. When a change is made to one ﬁle, usually it requires recompilation of some other ﬁles that depend on it. Manual recompilation of each of such ﬁles may not be practical. In Unix, a utility tool called make is available to overcome such diﬃculties. In this section we introduce this tool and discuss how to use it in the context of NS2 simulation development. As a Unix utility tool make is very useful for managing the development of software written in any compilable programming language including C++. Generally, the make program automatically keeps track of all the ﬁles created throughout the development process. By keeping track, we mean recompiling or relinking wherever interdependencies exist among these ﬁles, which may have been modiﬁed as a part of the development process. 2.7.1 An Invocation of a Make Utility A “make” utility can be invoked form a UNIX shell with the following command: >>make [-f mydescriptor] where “make” is mandatory, while the text inside the bracket is optional. By default (i.e., without optional input arguments), the make utility recompiles and relinks the source codes according to what speciﬁed in the default descriptor ﬁle Makefile. If the descriptor ﬁle mydescriptor is speciﬁed, the utility is use this ﬁle in place of the default ﬁle Makefile. 2.7.2 A Make Descriptor File A descriptor ﬁle contains an instructor of how the codes should be recompiled and relinked. Again, the default descriptor ﬁle is the ﬁle named “Makefile”. A descriptor ﬁle contains the names of the ﬁles that make up the executable, their interdependencies, and how each ﬁle should be rebuilt or recompiled. Such descriptions are speciﬁed through a series of so-called “dependency rules”. Each rule takes three components, i.e., targets, dependencies, and commands. The following is the format of the dependency rule: [ ...] : [ ...] [] A target with a colon sign is mandatory. Everything else inside the brackets are optional. A target is usually the name of the ﬁle which needs to be remade if any modiﬁcation is done to dependency ﬁles speciﬁed after the mandatory colon (:). If any change is noticed, the second line executes to regenerate the target ﬁle.

34

2 Introduction to Network Simulator 2

Example 2.2 (Example of a Descriptor File). Assume that we have a main executable ﬁle channel consisting of three separate source ﬁles named main.c, fade.c, and model.c. Also assume that model.c depends on model.h. The Makefile corresponding to this example is shown below. # makefile of channel channel : main.o fade.o model.o cc -o channel main.o fade.o model.o main.o : main.c cc -c main.c fade.o : fade.c cc -c fade.c model.o : model.c model.h cc -c model.c clean : rm main.o fade.o model.o The ﬁrst line is a comment beginning with a pound (“#”) sign. When make is invoked, it starts checking the targets one by one. The target channel is examined ﬁrst, and make ﬁnds that channel depends on the object ﬁles main.o, fade.o, and model.o. The make utility next checks to see if any of these object ﬁles is designated as a target ﬁle. If this is the case, make further checks the main.o object ﬁle’s dependency, and ﬁnds that it depends on main.c. Again, make proceeds to check whether main.c is listed as a target. If not, the command under the main.o target is executed if any change is made to main.c. In the command line “cc -c main.c”,5 main.c is simply compiled to obtain the main.o object. Next, make proceeds in a similar manner with the fade.o and model.o targets. Once any of these object ﬁles is updated, make returns to the channel target and executes its command, which merely compiles all of its dependent objects. Finally, we note a special target known as phony target which is not really the name of any ﬁle in the dependency hierarchy. This target is “clean”, and usually performs a housekeeping function such as cleaning up all the object ﬁles no longer needed after the compilation and linking. In Example 2.2 we notice several occurrences of certain sequences such as main.o fade.o model.o. To avoid a repetitive typing, which may introduce typos or omissions, a macro can be deﬁned to represent such a long sequence.
5

The UNIX command “cc -c file.c” compiles the ﬁle file.c and creates an object ﬁle file.o, while the command “cc -o file.o” links the object ﬁle file.o and create an executable ﬁle file.

2.7 Including C++ Modules into NS2 and the make Utility

35

For example, we may deﬁne a macro to represent main.o fade.o model.o as follows: OBJS = main.o fade.o model.o After deﬁning the macro, we refer to “main.o fade.o model.o” by either parentheses or curly brackets and precede that with a dollar sign (e.g., $(OBJS) or ${OBJS}). With this macro, Example 2.2 becomes a bit more handy as shown in Example 2.3. Example 2.3 (Example of makeﬁle/Makeﬁle with Macros.). # makefile of channel OBJS = main.o fade.o model.o COM = cc channel : ${OBJS} ${COM} -o channel ${OBJS} main.o : main.c ${COM} -c main.c fade.o : fade.c ${COM} -c fade.c model.o : model.c model.h ${COM} -c model.c clean : rm ${OBJS}

2.7.3 NS2 Descriptor File The NS2 descriptor ﬁle is deﬁned in a ﬁle Makefile located in the home directiory of NS2. It contains the details needed to recompile and relink NS2. The key relevant details are those beginning with the following keywords. • INCLUDES = : The items behind this keyword are the directory which should be included into the NS2 environment. • OBJ_CC = and OBJ_STL = : The items behind these two keywords constitute the entire NS2 object ﬁles. When a new C++ module is developed, its corresponding object ﬁle name should be added here. • NS_TCL_LIB = : The items bind this keywords are the Tcl ﬁle of NS2. Again, when a new OTcl module is developed, its corresponding Tcl ﬁle name should be added here. Suppose a module consisting of C++ ﬁles myc.cc and myc.h and a Tcl ﬁle mytcl.tcl. Suppose further that these ﬁles are created in a directory myfiles

36

2 Introduction to Network Simulator 2

under the NS2 home directory. Then this module can be incorporated into NS2 using the following steps: (i) Include a string “-I./myfiles” into the Line beginning with INCLUDES = in the Makefile. (ii) Include a string “myfile/myc.o” into the Line beginning with OBJ_CC = or OBJ_STL = in the Makefile. (iii) Include a string “myfile/mytcl.tcl” into the Line beginning with NS_TCL_LIB = in the Makefile. (iv) Run make from the shell. After running “make”, an executable ﬁle ns is created. We can now use this ﬁle ns to run simuation.

2.8 Chapter Summary
This chapter introduces Network Simulator (Version 2), NS2. In particular, information on the installation of NS2 in both Unix and Windows-based systems is provided. The basic architecture of NS2 is described. These materials are essential for understanding NS2 as a whole and would help to get one started working with NS2. NS2 consists of OTcl and C++. The C++ objects are mapped to OTcl handles using TclCl. To run a simulation, a user needs to deﬁne a network scenario in a Tcl Simulation script, and feeds this script as an input to an executable ﬁle ns. During the simulation, the packet ﬂow information can be collected through text-based tracing or NAM tracing. After the simulation, an AWK program or a perl program can be used to analyze a text-based trace ﬁle. The NAM program, on the other hand, utilizes a NAM trace ﬁle to replay the network simulation using animation. Simulation using NS2 consists of three main steps. First, the simulation design is probably the most important step. Here, we need to clearly specify the objectives and assumptions of the simulation. Secondly, conﬁguring and running simulation implements the concept designed in the ﬁrst step. This step also includes conﬁguring the simulation scenario and running simulation. The ﬁnal step in a simulation is to collect the simulation result and trace the simulation if necessary. Written mainly in C++, NS2 employs a make utility to compile the source code, to link the created object ﬁles, and create an executable ﬁle ns. It follows the instruction speciﬁed in the default descriptor ﬁle Makefile. The make utility provides a simple way to incorporate a newly developed modules into NS2. After developing a C++ source code, we simply add an object ﬁle name into the dependency, and re-run make.

3 Linkage Between OTcl and C++ in NS2

NS2 is an object oriented simulator written in OTcl and C++ languages. While OTcl acts as the frontend (i.e., user interface), C++ acts as the backend running the actual simulation (Fig. 3.1). As can be seen from Fig. 3.1, class hierarchies of both languages can be either standalone or linked together using an OTcl/C++ interface called TclCL [15]. There are two types of classes in each domain. The ﬁrst type includes classes which are linked between the C++ and OTcl domains. In the literature, these OTcl and C++ class hierarchies are referred to as the interpreted hierarchy and the compiled hierarchy, respectively. The second type includes OTcl and C++ classes which are not linked together. These classes are neither a part of the interpreted hierarchy nor a part of compiled hierarchy. This chapter discusses how OTcl and C++ languages constitute NS2.

C++

class

class class class

OTcl class class

class

class

The compiled hierarchy

class

class

class

class

The interpreted hierarchy

one-to-one correspondence

Fig. 3.1. Two language structure of NS2 [12]. Class hierarchies in both the languages may be standalone or linked together. OTcl and C++ class hierarchies which are linked together are called the interpreted hierarchy and the compiled hierarchy, respectively.

Written in C++, TclCL consists of the following six main classes: • Class Tcl provides methods to access the interpreted hierarchy (from the compiled hierarchy; Deﬁned in ﬁles ˜tclcl /tclcl.h and ˜tclcl /Tcl.cc).
T. Issariyakul, E. Hossain, Introduction to Network Simulator NS2, DOI: 10.1007/978-0-387-71760-9 3, c Springer Science+Business Media, LLC 2009

38

3 Linkage Between OTcl and C++ in NS2

• Class InstVar binds member variables in both the hierarchies together (Deﬁned in ﬁle ˜tclcl /Tcl.cc). • Class TclObject is the base class for all C++ simulation objects in the compiled hierarchy (deﬁned in ﬁle ˜tclcl /Tcl.cc). • Class TclClass maps class names in the interpreted hierarchy to class names in the compiled hierarchy (Deﬁned in ﬁles ˜tclcl /tclcl.h and ˜tclcl /Tcl.cc). • Class TclCommand provides a global access to the compiled hierarchy from the interpreted hierarchy (Deﬁned in ﬁles ˜tclcl /tclcl.h and ˜tclcl /Tcl.cc). • Class EmbeddedTcl translates OTcl scripts into C++ codes (Deﬁned in ﬁles ˜tclcl /tclcl.h, ˜tclcl /Tcl.cc, and ˜tclcl /tclAppInit.cc). The organization of this chapter is as follows. Section 3.1 describes the concept behind the two language structure of NS2. Sections 3.2 through 3.7 discuss the six main components of TclCL, namely, class Tcl, class InstVar, class TclObject, class Tclclass, class TclCommand, and class EmbeddedTcl. Finally, the chapter summary is given in Section 3.8.

3.1 The Two-Language Concept in NS2
Why two languages? Loosely speaking, NS2 uses OTcl to create and conﬁgure a network, and uses C++ to run simulation. All C++ codes need to be compiled and linked to create an executable ﬁle. Since the body of NS2 is fairly large, the compilation time is not negligible. A typical Pentium 4 computer requires few seconds (long enough to annoy most programmers) to compile and link the codes with a small change such as including “int i=0;” into the codes. OTcl, on the other hand, is an interpreter, not a compiler. Any change in a OTcl ﬁle does not need compilation. Nevertheless, since OTcl does not convert all the codes into machine language, each line needs more execution time. In summary, C++ is fast to run but slow to change. It is suitable for running a large simulation. OTcl, on the other hand, is slow to run but fast to change. It is therefore suitable to run a small simulation over several repetitions (each may have diﬀerent parameters). NS2 is constructed by combining the advantages of these two languages. NS2 manual provides the following guidelines to choose a coding language: • Use OTcl – for conﬁguration, setup, or one time simulation, or – to run simulation with existing NS2 modules. This option is preferable for most beginners, since it does not involve complicated internal mechanism of NS2. Unfortunately, existing NS2 modules are fairly limited. This option is perhaps not suﬃcient for most researchers. • Use C++ – when you are dealing with a packet, or

3.1 The Two-Language Concept in NS2

39

– when you need to modify existing NS2 modules. This option perhaps discourages most of the beginners from using NS2. This book particularly aims at helping the readers understand the structure of NS2 and feel more comfortable in modifying NS2 modules. In principle, one can develop a C++ program in three styles. The ﬁrst style–namely “Basic C++”–is the simplest form and involves basic C++ instructions only. This style has a ﬂexibility problem, since any change in system parameters requires a compilation (which takes non-negligible time) of the entire program. Addressing the ﬂexibility problem, the second coding style– namely “C++ coding with input arguments”–takes the system parameters as input arguments. As the system parameters change, we can simply change the input arguments, and do not need to recompile the entire program. The main problem of the second style is that the invocation could be quite lengthy for a large number of input arguments. The last coding style–“C++ coding with conﬁguration ﬁles”–puts all system parameters in a conﬁguration ﬁle, and has the C++ code read the system parameters from the conﬁguration ﬁle. This style does not have the ﬂexibility problem, and it facilitates program invocation. To change system parameters, we can simply change the content of the conﬁguration ﬁle. In fact, this style acts as a foundation from which NS2 develops. Recall from Section 2.5 that we write a Tcl simulation script and feed it as an input argument to NS2 when running a simulation (e.g., executing “ns myfirst_ns.tcl”). Here, “ns” is a C++ executable ﬁle obtained from the compilation, while myfirst_ns.tcl is an input conﬁguration ﬁle specifying system parameters and conﬁguration such as nodes, link, and how they are connected. Analogous to reading a script ﬁle through C++, NS2 reads the system conﬁguration from the Tcl simulation script. Again, when we change the parameters, we do not need to re-compile the entire NS2 code. All we have to do is to modify the Tcl simulation script and re-run the simulation. Example 3.1. Consider the network topology in Fig. 3.2. Deﬁne overall packet delivery delay as the time needed to carry a packet from the leftmost node to the rightmost node, where delay in link i is d_i and total number of nodes is num_nodes. We would like to measure the overall packet delivery delay and show the result on the screen.

packet

Link delay

...
Overall packet delivery delay

Fig. 3.2. A chain topology for network simulation.

40

3 Linkage Between OTcl and C++ in NS2

Basic C++ Coding

Program 3.1 Basic C++ codes which simulates Example 3.1, where the delay for each of the links is 1 unit and the number of nodes is 11.
1 2 3 4 5 6 7 //sim.cc main(){ float delay = 0, d_i = 1; int i, num_nodes = 11; for(i = 1; i < num_nodes; i++) delay += d_i; printf("Overall Packet Delay is %2.1f seconds.\n",delay); }

Suppose that every link has the same delay of 1 second (i.e., d_i = 1 second for all i), and the number of nodes is 11 (num_nodes = 11). Program 3.1 shows the C++ codes written in this style (the ﬁlename is “sim.cc”). Since the link delay is ﬁxed, we simply increment delay for num_nodes-1 times (Lines 4-5). After compiling and linking ﬁle sim.cc, we obtain an executable ﬁle sim. By executing “./sim” at the command prompt, we will see the following statement on the screen: >>./sim Overall Packet Delay is 10.0 seconds. Despite its simplicity, this coding style has a ﬂexibility problem. Suppose link delay is changed to 2 seconds. Then, we need to modify, compile, and link the ﬁle sim.cc to create a new executable ﬁle sim. After that, we can run “./sim” to generate another result (for d_i = 2 seconds). C++ Coding with Input Arguments We can avoid the above need for re-compilation and re-linking by feeding system parameters as input arguments of the program. Program 3.2 shows the codes which feed link delay and the number of nodes as the ﬁrst and the second arguments, respectively. Line 1 speciﬁes that the codes take input arguments. Variable argc is the number of input arguments. Variable argv is an argument vector which contains all input arguments provided by the caller (See the details on C++ coding with input arguments in [14]). With this style, we only need to compile and link the program once. After obtaining an executable ﬁle sim, we can obtain results by simply changing the input arguments. For example, >> ./sim 1 11

3.1 The Two-Language Concept in NS2

41

Overall Packet Delay is 10.0 seconds. >> ./sim 2 11 Overall Packet Delay is 20.0 seconds.

Program 3.2 C++ coding with input arguments: C++ codes which simulate Example 3.1. The ﬁrst and second arguments are link delay (d i) and the number of nodes (num nodes), respectively.
1 2 3 4 5 6 7 //sim.cc int main(int argc, char* argv[]) { float delay = 0, d_i = atof(argv[0]); int i, num_nodes = atoi(argv[1]); for(i = 1; i < num_nodes; i++) delay += d_i; printf("Overall Packet Delay is %2.1f seconds\n",delay); }

Though this coding style solves the ﬂexibility problem, it suﬀers from a large number of input arguments. For example, if delays in all the links in Example 3.1 are diﬀerent, we will have to type in all values of link delay every time we run the program. C++ Coding with Conﬁguration Files

Program 3.3 C++ coding style with conﬁguration ﬁles: C++ code which simulates Example 3.1. A sample conﬁguration ﬁle (config.txt) is given in Lines 10–11.
1 2 3 4 5 6 7 8 9 //sim.cc int main(int argc, char* argv[]) { float delay = 0, d[10]; FILE* fp = fopen(argv[1],"w"); int i, num_nodes = readArgFromFile(fp,d); for(i = 1; i < num_nodes; i++) delay += d[i-1]; printf("Overall Packet Delay is %2.1f seconds\n", delay); fclose(fp); }

//config.txt 10 Number of node = 11 11 Link delay = 1 2 3 4 5 6 7 8 9 10\vspace*{-3pt}

42

3 Linkage Between OTcl and C++ in NS2

Program 3.3 shows C++ simulation codes for Example 3.1. The program takes only one input argument which is the conﬁguration ﬁle name (See C++ ﬁle input/output in [14]). Function readArgFromFile(fp,d) reads the conﬁguration ﬁle associated with a ﬁle pointer fp, and sets variables num_node and d (the details are not shown here). In this case, the conﬁguration ﬁle (config.txt) is shown in Lines 10–11. When invoking “./sim config.txt”, the screen will show the following result. >>./sim config.txt Overall Packet Delay is 55.0 seconds. To change the system parameters, we can simply modify the ﬁle “config.txt”. Clearly, this coding style removes the necessity for compiling the entire code and the lengthy invocation process.

3.2 Class Tcl
Class Tcl is a C++ class which acts as an interface to the OTcl domain. Declared in ﬁle ˜tclcl/Tcl.cc, it provides methods for the following operations: (i) Obtain the Tcl instance (using function instance), (ii) Invoke OTcl (instance) procedures from within the C++ domain (using functions eval(...), evalc(...), and evalf(...)), (iii) Pass or receive results to/from the interpreter (using functions result(... ) and resultf(...)), (iv) Report error and quit the program in a uniform manner (using function error(...)), and (v) Retrieve the reference to TclObjects (using functions enter(...), delete (...), and lookup(...)). 3.2.1 Obtain a Reference to the Tcl Instance In C++, class functions are invoked through a class object (e.g., function “fn” can be invoked by “object.fn”). To invoke the above functions (e.g., eval(...) and result(...)) of class Tcl, we need to have an object of class Tcl. Class Tcl provides function “instance()” to obtain a static Tcl variable: Tcl& tcl = Tcl::instance(); Here, function instance() of class Tcl returns the static variable instance_ of class Tcl. Since it is static, in a simulation, there is only one Tcl object, instance_. Therefore, any attempt to retrieve a Tcl object using the above statement returns the same Tcl object. After obtaining the Tcl object, we can invoke class functions through the Tcl instance (e.g., eval(...) and result(...)).

3.2 Class Tcl

43

3.2.2 Invoking a Tcl Procedure We may need to invoke an OTcl instance procedure (instproc) when programming in C++. For example, we may obtain the current simulation time (see the deﬁnition in Chapter 4) by invoking instproc now{} of class Simulator in the interpreted hierarchy. Class Tcl provides four following functions to invoke OTcl procedures. For example, the following C++ codes tell OTcl to print out “Overall Packet Delay is 10.0 seconds” on the screen1 . • Tcl::eval(char* str): executes the command string stored in a variable “str” through the interpreter. For example, Tcl& tcl = Tcl::instance(); char s[128]; strcpy(s,"puts [Overall Packet Delay is 10.0 seconds]"); tcl.evalc(s); • Tcl::evalc(const char* str): executes the command string “str”. For example, Tcl& tcl = Tcl::instance(); tcl.eval("puts [Overall Packet Delay is 10.0 seconds]"); This function is diﬀerent from the former one in that the former one takes a “string variable” as an input variable (char*), while this one take a “string” as an input variable (const char*). • Tcl::eval(): executes the command which has already been stored in the internal variable bp_. For example, Tcl& tcl = Tcl::instance(); char s[128]; sprintf(tcl.buffer(),"puts [Overall Packet Delay is 10.0 seconds]"); tcl.eval(); where tcl::buffer() returns the internal variable bp_. The third line above prints the string stored in the variable bp_. • Instproc Tcl::evalf(const char* fmt,...): uses the format fmt of printf(...) in C++ to formulate a command string, and passes the formulated string to the interpreter. For example, Tcl& tcl=Tcl::instance(); float delay = 10.0; tcl.evalf("puts [Overall Packet Delay is %2.1f seconds]",delay);
1

You can save the sample codes in any C++ ﬁle and compile NS2 to create an executable ns ﬁle. When NS2 is invoked, the message “Overall Packet Delay is 10.0 seconds” should appear on the screen.

44

3 Linkage Between OTcl and C++ in NS2

3.2.3 Pass or Receive Results to/from the Interpreter After executing few statements, we may need to pass or receive values to/from the interpreter. For example, in Example 3.1, we may want to pass the value of the overall packet delivery delay to the interpreter instead of printing it to the screen. Class Tcl provides three functions to pass values back and forth between the two hierarchies. • Tcl::result(const char* fmt): passes the string result as the result to the interpreter. For example, the following statement returns 10 to the interpreter. Tcl& tcl=Tcl::instance(); tcl.result("10"); return TCL_OK; • Tcl::resultf(const char* result,...): uses the format of printf (...) in C++ to formulate a result string, and passes the formulated string to the interpreter. Example 3.2. Let command returnDelay of class Chain returns the value in C++ variable delay with one decimal digit to the interpreter. The implementation of the command returnDelay is given below: Tcl& tcl=Tcl::instance(); tcl.resultf("%1.1f",delay); return TCL_OK; From OTcl, the following statement stores the value of the variable “delay” of the C++ Chain object in the variable “d”. set chain [new Chain] set d [$chain returnDelay] sets the variable d to be the same as the variable delay in C++. • Tcl::result(void): retrieves the result from the interpreter as a string. For example, the following statements stores the value of the OTcl variable “d” in the C++ variable “delay”. Tcl& tcl=Tcl::instance(); tcl.evalc("$d"); char* delay = tcl.result(); Class Tcl uses a private member variable tcl_->result(...) to pass results between the two hierarchies. Here, tcl_ is a member of class Tcl, and is a pointer to a Tcl_Interp object. NS2 protects tcl_->result(...) from being accessed externally, and provides three functions to access this variable. Functions Tcl::result(const char* result) and Tcl::resultf(const char* result) set the value of tcl->result(...). After setting the value of

3.2 Class Tcl

45

tcl->result(...) in C++, NS2 may return to OTcl with a certain return value (e.g., TCL_OK, TCL_ERROR) We will discuss the details of this return mechanism in Section 3.4.4. After returning to OTcl, the interpreter reads the value of tcl->result(...) for a certain purpose (e.g., setting the delay value or reporting error). Similarly, after executing an OTcl statement (e.g., tcl.evalc("$delay")), the execution result is stored in the variable tcl->result(...). Function Tcl::result(void) in the compiled hierarchy returns the value stored in tcl->result(...) by the interpreter. 3.2.4 Reporting Error and Quitting the Program Class Tcl provides function “error(. . .)” to exit the program in a uniform way. This function simply prints a string stored in “str” and tcl->result(...) to the screen, and exits with code 1. Tcl::error(const char* str) The diﬀerence between Tcl::error(str) and return TCL_ERROR is as follows. Function Tcl::error(str) simply prints out the error message and exits. When returning TCL_ERROR, NS2 traps the error, which may occur in more than one point. In the end, the user may use the trapped errors to recover from the error, to locate the error, or to print all error messages in the error stack. 3.2.5 Retrieve the Reference to TclObjects Recall that an interpreted object always has a shadow compiled object. In some cases, we may need to obtain a shadow compiled object which corresponds to an interpreted object. NS2 creates the association between objects in two hierarchies by means of a hash table. Class Tcl provides the following functions to enter, delete, and retrieve an entry to/from the hash table. • Tcl::enter(TclObject* o): inserts the object “*o” to the hash table, and associates “*o” to the OTcl name string stored in a protected variable name_. This function is invoked by function TclClass:create_shadow(... ), when an object is created. • Tcl::delete(TclObject* o): deletes the entry associated with TclObject “*o” from the hash table. This function is invoked by function TclClass:delete_shadow(...), when an object is destroyed. • Tcl::lookup(char* str): returns the TclObject whose name is “str”. Example 3.3. Consider the C++ codes in Program 3.4. Here, argv[2] is an input argument passed from OTcl (in this case argv[2] is an interpreted object). Line 8 uses function TclObject::lookup(argv[2]) to retrieve the

46

3 Linkage Between OTcl and C++ in NS2

Program 3.4 Function Connector::command.
//~ns/common/connector.cc 1 int Connector::command(int argc, const char*const* argv) 2 { 3 Tcl& tcl = Tcl::instance(); 4 ... 5 if (argc == 3) { 6 if (strcmp(argv[1], "target") == 0) { 7 ... 8 target_ = (NsObject*)TclObject::lookup(argv[2]); 9 ... 10 } 11 ... 12 } 13 return (NsObject::command(argc, argv)); 14 }

shadow compiled object corresponding to the interpreted object argv[2]. The retrieved object is converted to an object of type NsObject and stored in variable∗ target_. Note that the details of function command will be discussed later in Section 3.4.4.

3.3 Class InstVar
Class InstVar acts as a glue which binds a member variable of a C++ class to an instproc of an OTcl class. When a C++ variable is bound to an OTcl instvar, any change in the C++ variable or the OTcl instvar will result in an automatic update the OTcl instvar or the C++ variable, respectively. NS2 supports variable binding for 5 following NS2 data types: real, integer, bandwidth, time, and boolean. These 5 data types are neither a C++ data type nor an OTcl data type.2 They are deﬁned here to facilitate NS2 value assignment. As shown in Table 3.1, these data types are deﬁned in the C++ classes InstVarReal, InstVarInt, InstVarBandwidth, InstVarTime, and InstVarBool, respectively, which derive from class InstVar. Among these ﬁve data types, real, bandwidth, and time data types make use of a double C++ data type, while integer and boolean employ int and bool C++ data types, respectively.

2

As indicated in Appendix A.1.3, Tcl stores everything in strings. Therefore, OTcl variables have no data type.

3.3 Class InstVar Table 3.1. OTcl bindable data types and C++ binding classes. OTcl data type Real Integer Bandwidth Time Boolean C++ binding class InstVarReal InstVarInt InstVarBandwidth InstVarTime InstVarBool

47

3.3.1 Real and Integer Variables These two NS2 data types are speciﬁed as real-valued and integer-valued, respectively. Optionally, we can also use “e” as “×10”, where denotes the value stored in the variable x. Example 3.4. Let realvar and intvar be instvars of an OTcl object “obj” and be of real and integer NS2 data types, respectively. Diﬀerent ways of setting3 realvar and intvar to 1200 are shown below. $obj set realvar 1.2e3 $obj set realvar 1200 $obj set intvar 1200 3.3.2 Bandwidth Bandwidth is speciﬁed as real-valued. By default, the unit of bandwidth is bits per second (bps). Optionally, we can add the following suﬃxes to bandwidth setting. • “k” or “K” means kilo or ×103, • “m” or “M” means mega or ×106 , and • “B” changes the unit from bits per second to bytes per second. NS2 only considers leading character of valid suﬃxes. Therefore, the sufﬁxes “M” and “Mbps” are the same to NS2. Example 3.5. In Example 3.4, let bwvar be an instvar of “obj” whose NS2 data type is bandwidth. The followings show diﬀerent ways of setting bwvar to be 8 Mbps (megabits per second). $obj $obj $obj $obj $obj
3

set set set set set

bwvar bwvar bwvar bwvar bwvar

8000000 8m 8Mbps 800k 1MB

See the OTcl value assignment in Appendix A.2.

48

3 Linkage Between OTcl and C++ in NS2

3.3.3 Time Time is speciﬁed as real-valued. By default, the unit of time is second. Optionally, we can add the following suﬃxes to change the unit. • “m” means milli or ×10−3, • “n” means nano or ×10−9 , and • “p” means pico or ×10−12 . Again, NS2 only reads the leading character of valid suﬃxes. Therefore, the suﬃxes “p” and “ps” are the same to NS2. Example 3.6. From Example 3.4, let timevar also be a time variable of “obj”. The following shows diﬀerent ways of setting timevar to be 2 micro seconds. $obj $obj $obj $obj set set set set timevar timevar timevar timevar 2m 2e-3 2e6n 2e9ps

3.3.4 Boolean Boolean is speciﬁed as either true (or a positive number) or false (or a zero). A boolean variable will be true if the ﬁrst letter of the value is greater than 0, is “t”, or is “T”. Otherwise, the variable will be false. Example 3.7. In Example 3.4, let boolvar be a boolean variable of “obj”. The following show diﬀerent ways of setting boolvar to be true and false. # set boolvar to $obj set boolvar $obj set boolvar $obj set boolvar $obj set boolvar $obj set boolvar $obj set boolvar $obj set boolvar # set boolvar to $obj set boolvar $obj set boolvar $obj set boolvar $obj set boolvar $obj set boolvar $obj set boolvar be TRUE 1 T true tasty 20 3.37 4xxx be FALSE 0 f false something 0.9 -5.29

3.4 Class TclObject

49

NS2 ignores all letters except for the ﬁrst one. As can be seen from Example 3.7, there are several strange ways for setting a boolean variable (e.g., tasty, something, -5.29). For better understanding, the readers are encouraged to experiment with boolean variable debug_ and real variable rate_ in the following codes4 : # Create a Simulator instance set ns [new Simulator] # Create an error model object set err [new ErrorModel] # Set values for class variables $err set debug_ something $err set rate_ 12e3 # Show the results puts "debug_(bool) is [$err set debug_]" puts "rate_(double) is [$err set rate_]" The results of execution of the above codes are as follows: >>debug_(bool) is 0 >>rate_(double) is 12000 After assigning a value to an OTcl variable, NS2 converts the string value to the corresponding type in C++. Except for boolean, NS2 converts the string to either double or int. During the conversion, valid suﬃxes are also converted (e.g., “M”is converted by multiplying 106 to the value). For boolean data type, NS2 retrieves the ﬁrst character in the string and throws away all other characters. If the retrieved character is an integer, NS2 will do nothing. If the retrieved character is a non-integer, NS2 will convert the character to one if it is “t” or “T” and to zero otherwise. After converting the string to a one-digit integer, NS2 casts the converted integer to boolean and updates the bound compiled variable.

3.4 Class TclObject
Class TclObject provides an instruction to create a compiled shadow object, when an interpreted object is created. The C++ class TclObject is mapped to the OTcl class SplitObject. These two classes are the base classes from which all classes (excluding the standalone classes) in their hierarchies develop. When an object is instantiated from the OTcl domain, the constructor of class SplitObject is invoked to initialize the object. One of the initialization is
4

Save the codes to a ﬁle (e.g., test.tcl) and run it (e.g., ns test.tcl).

50

3 Linkage Between OTcl and C++ in NS2

the shadow object construction whose instruction, which will be discuss in this section, Section 3.4.1 shows how a TclObject is referred to in both the hierarchies. Section 3.4.2 explains the shadow object creation and deletion procedure. The variable binding process performed during object construction is discussed in Section 3.4.3. Finally, Section 3.4.4 discusses a special function command(...), which provides an access to the compiled class from the OTcl domain. 3.4.1 Reference to a TclObject OTcl and C++ employ diﬀerent method to access their objects. As a compiler, C++ directly accesses the memory space allocated for a certain object (e.g., 0xd6f9c0). As an interpreter, OTcl does not directly access the memory. Rather, it uses a string (e.g., _o10) as a reference (or a handle)5 to the object. By convention, the name string of a SplitObject is of format _, where is a number uniquely generated for each SplitObject. Example 3.8. Let the variables c_obj and otcl_obj be C++ and OTcl objects, respectively. Table 3.2 shows examples of the reference value of C++ and OTcl objects.
Table 3.2. Examples of reference to (or handle of) TclObjects. Domain C++ OTcl Variable name c_object otcl_object Example 0xd6f9c0 _o10

We can see the format of a value stored in an OTcl object by running the following codes6 : set ns [new Simulator] set tcp [new Agent/TCP] puts "The value of tcp is $tcp" which show the following line on the screen. The value of tcp is _o10 3.4.2 Creating and Destroying a Shadow TclObject In most cases, objects are created and destroyed in the OTcl domain, or more precisely in a Tcl simulation script. Again, the OTcl commands create and
5 6

In NS2, the term “handle” means the object itself. Put the sample codes in a ﬁle (e.g, test.tcl) and run the ﬁle (e.g., ns test.tcl).

3.4 Class TclObject

51

destroy can be used to create and destroy, respectively, a standalone OTcl object. However, these commands are rarely used in NS2, since they do not create the shadow compiled object. In NS2, the global procedures new{...} and delete{...} are used to create and delete, respectively, an OTcl object as well as a shadow compiled object. Creating a TclObject A TclObject is created by using the global procedure new{...}, whose syntax is new [] The details of procedure “new{...}” are shown in Program 3.5. The procedure “new{className args}” takes two input arguments. The ﬁrst argument (mandatory) is the OTcl class name. The subsequent arguments (optional) is fed as input arguments to the OTcl constructor. The procedure “new{className args}” creates an object whose OTcl class is as well as its corresponding shadow compiled object. It will return the reference string (Line 11) if the construction process is successful. Otherwise, it will show an error message on the screen (Line 9). Program 3.5 Global instance procedure new.
//~tclcl/tcl-object.tcl proc new { className args } { set o [SplitObject getid] if [catch "$className create $o $args" msg] { if [string match "__FAILED_SHADOW_OBJECT_" $msg] { delete $o return "" } global errorInfo error "class $className: constructor failed: $msg" $errorInfo 10 } 11 return $o 12 } 1 2 3 4 5 6 7 8 9

The internal mechanism of the procedure “new{className args}” proceeds as follows. First, Line 2 retrieves a reference string for an object, and stores the string in variable “o”. Instproc getid{} of class SplitObject creates a reference string according to the naming format deﬁned in Section 3.4.2. Next, Line 3 creates an object whose OTcl class is className and associates the created object with the string stored in “o”. Finally, if the object

52

3 Linkage Between OTcl and C++ in NS2

is successfully created, Line 11 returns the reference string “o” to the caller7 . Otherwise, an error message (Line 9) will be shown on the screen. The OTcl command create in Line 3 invokes instproc alloc{...} to allocate a memory space for an object of class className, and instproc init{...} to initialize the object. In most cases, instproc init{...} is referred to an OTcl constructor. Each class overrides function init{...} and deﬁnes its own initialization in this function. Program 3.6 Samples of object constructor: Classes Agent/TCP and SimpleLink.
1 2 3 4 5 //~ns/tcl/lib/ns-agent.tcl Agent/TCP instproc init {} { eval $self next set ns [Simulator instance] $ns create-eventtrace Event $self }

//~tclcl/tcl-object.tcl 6 SplitObject instproc init args { 7 $self next 8 if [catch "$self create-shadow $args"] { 9 error "__FAILED_SHADOW_OBJECT_" "" 10 } 11 }

Program 3.6 shows an example of the OTcl constructor. Instproc next{...} in Line 2 invokes the instproc with the same name (i.e., init{...} in this case) of the parent class. This is a common concept in an Object Oriented Programming, where the constructor of the parent class needs to be called earlier. The construction therefore moves up the hierarchy until it reaches class SplitObject (see Lines 6–11 in Program 3.6). Here, Line 8 creates a shadow compiled object by invoking the command create-shadow, which will be discussed later in Section 3.5. We now conclude this section with an example of a creation of a Agent/TCP OTcl object as well as its shadow compiled object. Example 3.9. To create an OTcl Agent/TCP object, we can execute “new Agent/TCP” from a Tcl Simulation Script. In the interpreted hierarchy, class Agent/TCP derives from class Agent, which derives from class SplitObject. In the compiled hierarchy, these three classes correspond to class TcpAgent, Agent, and TclObject, respectively.
7

Note that Line 11 returns a reference string stored in o, not the variable o. Hence the procedure new returns a reference string stored in the variable o.

3.4 Class TclObject
Forward path

53

new Agent/Tcp Set o [SplitObject getid] Agent/Tcp create $o Agent::Agent Agent/Tcp alloc $o Agent/Tcp::init $self next … Other initialization Return the created object (o) Return TcpAgent::TcpAgent - Bind variables - Other Initialization Return Agent/Tcp::init $self next … TcpClass::create Other initialization Return new TcpAgent Return - Bind variables - Other Initialization Return
Return path

Return

SplitObject::init create-shadow TcpClass::create-shadow class_tcp.create Other initialization Return Return

OTcl

C++

Fig. 3.3. Object creation diagram: Class Agent/TCP derives from class Agent, which derives from class SplitObject.

Figure 3.3 shows the creation process of an object (o) of class Agent/TCP. Again, the ﬁrst step is to retrieve a reference string by invoking instproc getid{} of class SplitObject. The next step is to invoke instproc init{...} up the hierarchy. On the top level, class SplitObject invokes command create-shadow to create a shadow compiled object (on the right hand side of Fig. 3.3 which will be discussed in Section 3.5). After returning from instproc create-shadow, the process performs the rest of initialization and moves (or returns) down the interpreted hierarchy

54

3 Linkage Between OTcl and C++ in NS2

until it reaches class Agent/TCP. Then, it returns to procedures create{...} and new{...}, respectively, where the reference string corresponding to the created object “o” is returned to the caller (of procedure new{...}). Note that the above procedures are used to create an interpreted TclObject which is linked to the compiled hierarchy. Standalone C++ or OTcl objects do not need any shadow object, and do not have to go through the above procedures. They can be constructed in a normal way. Destroying a TclObject OTcl uses instproc delete{...} to destroy an interpreted object as well as its shadow compiled object (by invoking instproc delete-shadow). Program 3.7 shows a sample usage of instproc delete{...}. Instproc Simulator::use-sch eduler{...} removes the existing scheduler (if any; Line 3) by using instproc delete{...}, and creates an object of class Scheduler/$type using the global procedure new{...}. We will discuss the details of instproc Simulator::use-scheduler{...} in Chapter 4. Program 3.7 An example usage of global procedures new and delete.
1 2 3 4 5 //~ns/tcl/lib/ns-lib.tcl Simulator instproc use-scheduler type { $self instvar scheduler_ delete $scheduler_ set scheduler_ [new Scheduler/$type] }

3.4.3 Binding Variables in the Compiled and Interpreted Hierarchies In general, both interpreted and compiled objects have their own class variables, and they are not allowed to directly access one another’s class variables. NS2, therefore, provides a mechanism which binds class variables in both hierarchies together. After the binding, a change in a class variable in one hierarchy will result in an automatic change in the bound variable in another hierarchy. Binding Variables in Both Hierarchies NS2 binds an interpreted class variable to a compiled variable during shadow object construction. More speciﬁcally, class TclObject invokes the following functions in the constructor to bind variables in both hierarchies (see ﬁle ˜tclcl /tclcl.h).

3.4 Class TclObject

55

bind("iname",&cname) bind_bw("iname",&cname) bind_time("iname",&cname) bind_bool("iname",&cname) where iname and cname are the names of the class variables in the interpreted and compiled hierarchies, respectively. Essentially, the ﬁrst and second arguments of the above functions are the name string of the interpreted variable and the address of the compiled variable, respectively. Example 3.10. Let class Test in both hierarchies be bound together. Let icount_, idelay_, ispeed_, ivirtual_time_, iis_running_ be OTcl class variables whose types are integer, real, bandwidth, time, and boolean, respectively. The following codes show declaration and the constructor of C++ class Test. class Test { /* Declaration */ public: int count_; double delay_,virtual_time_,speed_; bool is_running_; Test() { /* Constructor */ bind("icount_",&count_); bind("idelay_",&delay_); bind_bw("ispeed_",&speed_); bind_time("ivirtual_time_",&virtual_time_); bind_bool("iis_running_",&is_running_); }; }; All class variables are bound in the compiled constructor (i.e., Test()). By convention, we use the same variable name for both hierarchies. Here, however, we would like to show that bound variables do not need to have the same names. Setting the Default Values NS2 sets the value of bound variables as speciﬁed in ﬁle ˜ns/tcl/lib/nsdefault.tcl. The syntax for setting a default value is similar to the value assignment syntax in OTcl. That is, set which sets the instvar of class to be . As an example, a part of ﬁle ˜ns/tcl/lib/ns-default.tcl is shown in Program 3.8. To set the default values for the variables, NS2 invokes instproc init-instvar{...} of class SplitObject (see ﬁle ˜tclcl /tcl-object.tcl). Instproc init-instvar{...} takes variables’ default values from ﬁle ˜ns/tcl/lib/

56

3 Linkage Between OTcl and C++ in NS2

Program 3.8 An example for specifying default values in NS2: A part of ﬁle ˜ns/tcl/lib/ns-default.tcl.
1 2 3 4 5 6 7 8 9 10 11 //~ns/tcl/lib/ns-default.tcl ErrorModel set enable_ 1 ErrorModel set markecn_ false ErrorModel set delay_pkt_ false ErrorModel set delay_ 0 ErrorModel set rate_ 0 ErrorModel set bandwidth_ 2Mb ErrorModel set debug_ false Classifier Classifier Classifier Classifier set set set set offset_ 0 shift_ 0 mask_ 0xffffffff debug_ false

ns-default.tcl, and assigns them to the bound variables. If we bind a variable but do not specify the default value, instproc SplitObject::warn-instvar {...} invoked from within SplitObject::init-instvar{...} will show a warning message on the screen. A warning message will not be shown, if a default value is assigned to an invalid variable (e.g., not-bound or does not exist). 3.4.4 OTcl Commands Section 3.2.2 showed an approach to access the interpreted hierarchy from the compiled hierarchy. This section discusses the reverse: a method to access the compiled hierarchy from the interpreted hierarchy called “command”. Review of Instance Procedure Invocation Mechanism Before we proceed further, let us review the OTcl instproc invocation mechanism. An instproc is invoked according to the following syntax: $obj [] where the instproc name and the input argument are mandatory and optional, respectively, for such an invocation. The internal mechanism of the above instproc invocation proceeds as follows: (i) Look for a matching instproc in the object class. If found, execute the matched instproc and return. If not, proceed to the next step. (ii) Look for instproc “unknown{...}”. If found, execute “unknown{...}” and return. If not, proceed to the next step. The instproc “unknown{...}” is the default instproc which will be invoked if no matching instproc is found.

3.4 Class TclObject

57

(iii) Repeat steps (i) and (ii) for the base class of the object. (iv) If the top class is reached but neither the input instproc nor the instproc unknown is found, report an error and exit the program. OTcl Command Invocation The syntax of a command is the same as that of an instproc, i.e., $obj [] The main diﬀerence is that is replaced with . Since the syntax for invoking a command is the same as that for invoking an instproc, OTcl executes the command as if it is an instproc. In the following, we will explain the command invocation mechanism of an OTcl Agent/TCP object (see Program 3.9). Figure 3.4 shows the internal mechanism of the command invocation process, which proceeds as follows: (i) Execute the statement “$tcp ”. (ii) Look for an instproc in the OTcl class Agent/TCP. If found, invoke the instproc and complete the process. Otherwise, proceed to the next step. (iii) Look for an instproc unknown{...} in the OTcl class Agent/TCP. If found, invoke the instproc unknown{...} and complete the process. Otherwise, proceed to the next step. (iv) Repeat steps (ii) and (iii) until reaching class SplitObject. If the instprocs unknown{...} is not found in any class in the inheritance tree, the following statement will be executed. SplitObject unknown The instproc unknown{...} of class SplitObject is deﬁned in ﬁle ˜tclcl /tcl-object.tcl. Here, the statement “$self cmd $args” is executed, where args are the input arguments of instproc unknown{...}. Based on the above invocation, this statement interpolates to SplitObject cmd where is “ ”. (v) Instproc cmd passes the entire statement (i.e., “cmd ”) as an input argument vector (argv) to function “command(argc,argv)” of the shadow object (TcpAgent in this case). As shown in Program 3.9, function “command(argc,argv)” always takes two input arguments. The second input argument (argv) is an argument vector, which is an array of strings containing arguments passed from the instproc “cmd”. The ﬁrst input argument (argc) is the total number of input arguments (i.e., the number of non-empty elements of argv). The ﬁrst and second elements of argv are “cmd” and the command name (), respectively. The subsequent elements contain the input arguments () of the original invocation, each of which is separated by one or more white spaces (see Table 3.3).

58

3 Linkage Between OTcl and C++ in NS2

Fig. 3.4. Command invocation process.

(vi) Function command(argc,argv) checks for the matching number of arguments (stored in argc) and command name (stored in argv[1]). If found, it takes the desired actions (e.g., Lines 6–7 in Program 3.9), and returns TCL_OK. If no criterion matches with (argc,argv), it will skip to the last line (Line 12). (vii) Line 12 in Program 3.9 invokes function command(argc,argv) of the base class (i.e., class Agent::command(argc,argv)), feeding (argc,argv) as input arguments.

3.4 Class TclObject

59

Program 3.9 Function TcpAgent::command.
//~ns/tcp/tcp.cc 1 int TcpAgent::command(int argc, const char*const* argv) 2 { 3 ... 4 if (argc == 3) { 5 if (strcmp(argv[1], "eventtrace") == 0) { 6 et_ = (EventTrace *)TclObject::lookup(argv[2]); 7 return (TCL_OK); 8 } 9 ... 10 } 11 ... 12 return (Agent::command(argc, argv)); 13 }

Table 3.3. Description of elements of array argv of function command. index (i) 1 2 3 4 . . . Element (argv[i]) cmd The command name () The ﬁrst input argument in The second input argument in . . .

(viii) Repeat steps (vi) and (vii) up the hierarchy until the criterion is matched. If the process reaches class TclObject and the criterion does not match, function command of class TclObject will report an error (e.g., no such method, requires additional args), and return TCL_ERROR (see ﬁle ˜tclcl /Tcl.cc). (xi) Return down the class hierarchy. When reaching C++ class TcpAgent, return to OTcl (instprocs cmd and unknown{...}, respectively) with a return value (e.g., TCL_OK or TCL_ERROR), and complete the command invocation. An Alternative for OTcl Command Invocation In the last subsection, we invoked an OTcl command by executing $tcp which starts from position (1) in Fig. 3.4. Alternatively, we can also invoke a command using the following syntax: $tcp cmd which starts from position (2) in Fig. 3.4.

60

3 Linkage Between OTcl and C++ in NS2

The latter (position (2)) invocation method avoids the ambiguity when OTcl deﬁnes an instproc whose name is the same as the OTcl command name. Suppose an OTcl command associated with an object “tcp” has an implementation in the C++ class TcpAgent. Suppose further that an instproc (same name) is also deﬁned in OTcl class Agent/TCP. When invoking “$tcp ”, NS2 will perform the actions speciﬁed in the OTcl instproc . To invoke the OTcl command whose implementation is in C++, we need to invoke “$tcp cmd ”. Since instproc cmd is deﬁned solely in class SplitObject, this invocation avoids the ambiguity of OTcl command and instproc names.

OTcl Command Returning Mechanism After performing the desired actions speciﬁed in C++, NS2 returns to OTcl with a certain return value. In ﬁle nsallinone-2.30/tcl8.4.13/generic/ tcl.h, NS2 deﬁnes ﬁve following return values (as 0–5), as speciﬁed in Program 3.10, which inform the interpreter of the command invocation result. Program 3.10 Return values in NS2.
1 2 3 4 5 //nsallinone-2.30/tcl8.4.13/generic/tcl.h #define TCL_OK 0 #define TCL_ERROR 1 #define TCL_RETURN 2 #define TCL_BREAK 3 #define TCL_CONTINUE 4

• TCL_OK: The command completes successfully. • TCL_ERROR: The command does not complete successfully. The interpreter will explain the reason for the error. • TCL_RETURN: After returning from C++, the interpreter exits (or returns from) the current instproc without performing the rest of instproc. • TCL_BREAK: After returning from C++, the interpreter breaks the current loop. This is similar to executing C++ keyword break, but the results prevail to the OTcl domain. • TCL_CONTINUE: After returning from C++, the interpreter continues to the next iteration. This is similar to executing C++ keyword continue, but the results prevail to the OTcl domain.. Among these ﬁve types, TCL_OK and TCL_ERROR are the most common ones. If C++ returns TCL_OK, the interperter may read the value passed from the C++ domain. Recalling from Section 3.2.3, the interpreter does not read

3.5 Class TclClass

61

the return value, but it reads the value speciﬁed in the statement. The return code TCL_OK only tells OTcl that the value stored by the statement tcl.result(...) is valid. If an OTcl command returns TCL_ERROR, on the other hand, the interpreter will invoke procedure tkerror (deﬁned in ﬁle ˜tclcl/tcl-object.tcl), which simply shows an error on the screen. Exercise 3.11. What are the diﬀerences among a C++ function, an OTcl instproc, and an OTcl command?

3.5 Class TclClass
When a TclObject is created, NS2 automatically constructs a shadow compiled object. In Section 3.4.2, we have explained the TclObject creation mechanism. We have mentioned that class TclClass is responsible for the shadow object creation process. We now explain the details of class TclClass as well as the shadow object creation process. 3.5.1 An Overview of Class TclClass Class TclClass is mainly responsible for creating a shadow object in the compiled hierarchy. It maps an OTcl class to a C++ static mapping variable, and provides a method to create a shadow object in the compiled hierarchy. As an example, Program 3.11 shows the details of class TcpClass, which maps class Agent/TCP in the interpreted hierarchy to the static mapping variable class_tcp in the compiled hierarchy. Program 3.11 Declaration and implementation of class TcpClass.
1 2 3 4 5 6 7 //~ns/tcp/tcp.cc static class TcpClass : public TclClass { public: TcpClass() : TclClass("Agent/TCP") {} TclObject* create(int , const char*const*) { return (new TcpAgent()); } } class_tcp;

Unlike other classes, a child class of class TclClass is declared, implemented, and instantiated (e.g., of variable class_tcp in Line 7) in the same place. From Program 3.11, a child class of class TclClass consists of only two functions: the constructor (TcpClass in Line 3) and function create(...) (Lines 4–6) which creates a shadow object. To construct a shadow object for an OTcl object of class Agent/TCP, we need to perform the following actions in the compiled hierarchy:

62

3 Linkage Between OTcl and C++ in NS2

(i) (ii) (iii) (iv)

Create a shadow compiled class (e.g., TcpAgent). Derive a mapping class (e.g., TcpClass) from class TclClass. Instantiate a static mapping variable (e.g., class_tcp). Deﬁne the constructor of the mapping class (Line 3 in Program 3.11). Feed the OTcl class name (e.g., Agent/TCP) as an input argument to the base constructor (i.e., class TclClass). (v) Deﬁne function “create(...)” to construct a shadow compiled object; Invoke “new” to create a shadow compiled object (e.g., new TcpAgent) and return the created object to the caller (Line 5 in Program 3.11).

3.5.2 TclObject Creation We now explain the entire TclObject creation process. Once again, consider Fig. 3.3. The TclObject creation process proceeds as follows: • Create an OTcl object as in Section 3.4.2. • Invoke instproc create-shadow of class TclClass (see ﬁle ˜tclcl /Tcl.cc). • From within function create_shadow(...), invoke function create(...) of class TcpClass. • In Program 3.11, function create(...) in Line 5 executes “new TcpAgent” and returns the created object to the caller. • Construct a TcpAgent object, by calling the constructor of its parent classes (Agent and TclObject). • Construct Agent object. This includes binding all variables to those in the interpreted hierarchy.8 • Return to class TcpAgent. Construct the TcpAgent object, and bind all variables to those in the interpreted hierarchy. • Return the created shadow object to instproc SplitObject::init{...}, and proceed as speciﬁed in Section 3.4.2. 3.5.3 Naming Convention for Class TclClass The convention to name a class derived from class TclClass and the corresponding static variable are described now. First, every class derives directly from class TclClass, irrespective of its class hierarchy. For example, class RenoTcpAgent derives from class TcpAgent. However, their mapping classes RenoTcpClass and TcpClass derive from class TclClass. Secondly, the naming convention is very similar to the C++ variable naming convention. In most cases, we simply name the mapping class by attaching the word Class to the C++ class name. The static mapping variable is named by attaching the word “class_” to the front. Table 3.4 shows few examples of the above naming convention.
8

Recall from Section 3.4.3 that NS2 binds variables of both hierarchies in the constructor.

3.6 Class TclCommand Table 3.4. Examples of naming convention for class TclClass. TclObject TcpAgent RenoTcpAgent DropTail SplitObject Agent/TCP Agent/TCP/Reno Queue/DropTail Mapping class TcpClass RenoTcpClass DropTailClass Mapping variable class_tcp class_reno class_drop_tail

63

3.5.4 Instantiation of Mapping Variables At the startup, NS2 instantiates all static mapping variables. Here, class TclClass stores the OTcl class names in its member variable classname_ and stores all mapping variables to its linked list “all_”. After all mapping variables are inserted into the linked list, function TclClass::bind(...) is invoked. Function bind(...) registers all mapping variables in “all_” into the system, and creates the interpreted class hierarchy. Function bind(...) also binds instprocs create-shadow and delete-shadow to functions create_sha dow(...) and delete_shadow(...) of the mapping classes (e.g., TcpClass9), respectively. After this point, NS2 recognizes all OTcl class names. Creation of an OTcl object will follow the procedures speciﬁed in Sections 3.4.2 and 3.5.2. Exercise 3.12. What are the major diﬀerences among classes TclObject, TclClass, and InstVar? Explain their roles during an object creating process.

3.6 Class TclCommand
As discussed in Section 3.4.4, OTcl command is a method to access the compiled hierarchy form the interpreted hierarchy. This section discusses another method called TclCommand to do the same. The main diﬀerence of OTcl command and TclCommands is as follows. Each OTcl command is associated with an OTcl/C++ class and cannot be invoked independently. Each TclCommand, on the other hand, is not bound to any class and is available globally. Since TclCommands violate the object oriented concept, it is not advisable to create this type of commands. 3.6.1 Invoking a TclCommand A TclCommand is invoked as if it is a global OTcl procedure. We will explain how to invoke a TclCommand through Example 3.13. Example 3.13. Consider the TclCommands ns-version and ns-random, speciﬁed in ﬁle ˜ns/common/misc.cc.
9

In fact, class TcpClass inherits functions create shadow and delete shadow from class TclClass.

64

3 Linkage Between OTcl and C++ in NS2

• TclCommand ns-version takes no argument and returns NS2 version. • TclCommand ns-random returns a random number uniformly distributed in [0, 231 − 1] when no argument is speciﬁed. If an input argument is given, it will be used to set the seed of the random value generator. These two TclCommands can be invoked globally. For example, >>ns-version 2.30 >>ns-random 729236 >>ns-random 1193744747 ### TERMINATE NS2 ### >>ns-random 729236 >>ns-random 1193744747 ### TERMINATE NS2 ### >>ns-random 101 101 >>ns-random 72520690 >>ns-random 308637100 By executing ns-version, the version (2.30) of NS2 is shown on the screen. TclCommand ns-random with no argument returns a random number (e.g., 729236, 1193744747, · · ·). In NS2, a random number is generated by picking a number from a sequence of pseudo-random numbers. A random seed speciﬁes the starting position in the sequence. By default, NS2 always sets random seed to be 0. The results from multiple simulations would be the same unless the seeds are set diﬀerently. In the above example, we do not feed the seed for the ﬁrst two runs. Therefore, the generated random numbers are the same for the ﬁrst two runs. In the third run, we set the seed to be 101, and obtain a diﬀerent set of random values (i.e., 72520690, 308637100, · · ·). An important note: you must set random seeds diﬀerently for diﬀerent runs. Otherwise, NS2 will generate the same result. 3.6.2 Creating a TclCommand A TclCommand creation process is similar to those of a TclClass and function command of a TclObject. A TclCommand is deﬁned in a class derived from class TclCommand. The name of a Tclcommand is provided as an input argument of class TclCommand, while the implementation is deﬁned in function “command(...)”. When NS2 starts, it binds all TclCommand names to function “command(...)” of the corresponding classes.

3.6 Class TclCommand

65

Program 3.12 Declaration and function command of class RandomCommand.
1 2 3 4 5 //~ns/common/misc.cc class RandomCommand : public TclCommand { public: RandomCommand() : TclCommand("ns-random") { } virtual int command(int argc, const char*const* argv); };

6 int RandomCommand::command(int argc, const char*const* argv) 7 { 8 Tcl& tcl = Tcl::instance(); 9 if (argc == 1) { 10 sprintf(tcl.buffer(), "%u", Random::random()); 11 tcl.result(tcl.buffer()); 12 } else if (argc == 2) { 13 int seed = atoi(argv[1]); 14 if (seed == 0) 15 seed = Random::seed_heuristically(); 16 else 17 Random::seed(seed); 18 tcl.resultf("%d", seed); 19 } 20 return (TCL_OK); 21 }

Program 3.12 shows the details of TclCommand ns-random, which is associated with class RandomCommand. Here ns-random is fed to the constructor of class TclCommand (Line 3). When invoking ns-random, NS2 invokes function command(...) of class RandomCommand, passing the command name as well as its input arguments to the function command(...). When invoking the command ns-random, Lines 10–11 generate a random number, and pass it to the interpreter. If the number of arguments is one, Lines 17-18 set the random seed to the input argument and pass the seed to the interpreter. TclCommands ns-version and ns-random in Example 3.13 are deﬁned in ﬁle ˜ns/common/misc.cc. At the startup time, NS2 invokes function init_misc(...) (see Program 3.13) in ﬁle ˜tclcl /TclAppInit.cc. This function simply instantiates all TclCommands by calling “new{...}” (e.g., Lines 3–4 in Program 3.13). After this point, every TclCommand invoked from the OTcl domain will refer to the corresponding instantiated TclCommand object. 3.6.3 Deﬁning Your Own TclCommand To create a TclCommand, you need to (i) Derive a TclCommand class directly from class TclCommand, (ii) Feed the TclCommand name to the constructor of class TclCommand,

66

3 Linkage Between OTcl and C++ in NS2

Program 3.13 Function misc init, which instantiates of TclCommands.
1 2 3 4 5 6 //~ns/common/misc.cc void init_misc(void) { (void)new VersionCommand; (void)new RandomCommand; ... }

(iii) Provide implementation (i.e., desired actions) in the function command (...), and (iv) Add an object instantiation statement in function init_misc(...). Example 3.14. Let the TclCommand print-all-args show all input arguments on the screen. We can implement this TclCommand by including the following codes to ﬁle ˜ns/common/misc.cc: class PrintAllArgsCommand : public TclCommand { public: PrintAllArgsCommand():TclCommand("print-all-args") {}; int command(int argc, const char*const* argv); } int PrintAllArgsCommand::command(int argc, const char*const* argv) { cout gen/ns_tcl.cc”, translates the expanded ﬁle in the ﬁrst part into an EmbeddedTcl object et_ns_lib (using a unix pipe “|”) and redirects the printed result into ﬁle ˜ns/gen/ns tcl.cc (using the unix redirect operator “>”). (ii) During NS2 startup, NS2 loads the translated EmbeddedTcl objects into NS2. To incorporate a new scripting ﬁle “file” into NS2, we need to source the ﬁle by inserting the statement “source file” into ﬁle ˜ns/tcl/lib/ns-lib.tcl. At the compilation, a new scripting ﬁle will be included into NS2, and will be ready to use thereafter.

3.8 Chapter Summary
NS2 is written in OTcl (interpreted class hierarchy) and C++ (compiled class hierarchy). Loosely speaking, OTcl sets up a network (e.g., creating and connecting nodes), while C++ runs actual simulation (e.g., passing packets from one node to another). When an object is created from the interpreted hierarchy, a so-called shadow object is also created in the compiled hierarchy. The connection between the interpreted and compiled hierarchies is established through TclCL which consists of following C++ classes.
10

Other EmbeddedTcl object (e.g., et tclobject) are created similarly.

68

3 Linkage Between OTcl and C++ in NS2

• Class TclObject is the main class where all classes in the compiled hierarchy derive. It corresponds to an OTcl class SplitObject, which is the base class for all classes in the interpreted hierarchy. Class TclObject has four main responsibilities. The ﬁrst two responsibilities are to provide methods to create and destroy a C++ shadow object, when an OTcl object is created and destroyed, respectively. The third responsibility is to bind class variables in both hierarchies together so that a change in the variable in one hierarchy will result an automatic update in the bound variable in another hierarchy. The last responsibility is to provide method–namely OTcl command–to access C++ from OTcl domain. • Class TclClass maps an OTcl class name to a C++ static mapping variable. While class TclObject initiates the shadow object creation process, the actual shadow object creation is performed by class TclClass. • Class InstVar deﬁnes NS2 variable data types which can be bound in both the hierarchies. • Class Tcl provides an access to the interpreted hierarchy from the compiled hierarchies. • Similar to OTcl command, class TclCommand provides a global access to the compiled hierarchy from the interpreted hierarchy. • Class EmbeddedTcl translates OTcl scripts into C++ codes.

4 Implementation of Discrete-Event Simulation in NS2

NS2 is a discrete-event simulator, where actions are associated with events rather than time. An event in a discrete-event simulator consists of execution time, a set of actions, and a reference to the next event (Fig. 4.1). These events connect to each other and form a chain of events on the simulation timeline (e.g., that in Fig. 4.1). Unlike a time-driven simulator, in an event-driven simulator, time between a pair of events does not need to be constant. When the simulation starts, events in the chain are executed from left to right (i.e., chronologically).1 In the next section, we will discuss the simulation concept of NS2. In Sections 4.2, 4.3, and 4.4, we will explain the details of classes Event and Handler, class Scheduler, and class Simulator, respectively. Finally, we summarize this chapter in Section 4.6.

create event

insert event

Event5 time = 3.7 Action5 Event3 time = 5 Action3 Event4 time = 6.8 Action4 Time (second)

Event1 time = 0.9 Action1

Event2 time = 2.2 Action2

1

2

3

4

5

6

7

Fig. 4.1. A sample chain of events in a discrete-event simulation. Each event contains execution time and a reference to the next event. In this ﬁgure, Event1 creates and inserts Event5 after Event2 (the execution time of Event 5 is at 3.7 second).

1

By execution, we mean taking actions associated with an event.

T. Issariyakul, E. Hossain, Introduction to Network Simulator NS2, DOI: 10.1007/978-0-387-71760-9 4, c Springer Science+Business Media, LLC 2009

70

4 Implementation of Discrete-Event Simulation in NS2

4.1 NS2 Simulation Concept
NS2 simulation consists of two major phases. Phase I: Network Conﬁguration Phase In this phase, NS2 constructs a network and sets up an initial chain of events. The initial chain of events consists of events which are scheduled to occur at certain times (e.g., start FTP (File Transfer Protocol) traﬃc at 1 second.). These events are called at-events (see Section 4.2). This phase corresponds to every line in a Tcl simulation script before executing instproc run{} of the Simulator object. Phase II: Simulation Phase This part corresponds to a single line, which invokes instproc Simulator::run {}. Ironically, this single line contributes to most (e.g., 99%) of the simulation. In this part, NS2 moves along the chain of events and executes each event chronologically. Here, the instproc Simulator::run{} starts the simulation by dispatching the ﬁrst event in the chain of events. In NS2, “dispatching an event” or “ﬁring an event” means “taking actions corresponding to that event”. An action is, for example, starting FTP traﬃc or creating another event and inserting the created event into the chain of events. In Fig. 4.1, at 0.9 s, Event1 creates Event5 which will be dispatched at 3.7 s, and inserts Event5 after Event2. After dispatching an event, NS2 moves down the chain and dispatches the next event. This process repeats until the last event corresponding to instproc halt{} of OTcl class Simulator is dispatched, signifying the end of simulation.

4.2 Events and Handlers
4.2.1 An Overview of Events and Handlers As shown in Fig. 4.1, an event speciﬁes an action to be taken at a certain time. In NS2, an event contains a handler which speciﬁes the action, and the ﬁring time or dispatching time. Program 4.1 shows declaration of classes Event and Handler. Class Event declares variables handler_ (whose class is Handler; Line 5) and time_ (Line 6) as its associated handler and ﬁring time, respectively. To maintain the chain of events, each Event object contains pointers next_ (Line 3) and prev_ (Line 4) to the next and previous Event objects, respectively. Variable uid_ (Line 7) is an ID unique to every event. Lines 10–14 in Program 4.1 show the declaration of an abstract class Handler. Class Handler speciﬁes the default action to be taken when an associated event is dispatched in its pure virtual function handle(e) (Line 13)2 .
2

We call actions speciﬁed in the function handle(e) default action, since they are taken by default when the associated event is dispatched.

4.2 Events and Handlers

71

Program 4.1 Declaration of classes Event and Handler.
1 2 3 4 5 6 7 8 9 //~/ns/common/scheduler.h class Event { public: Event* next_; /* event list */ Event* prev_; Handler* handler_; /* handler to call when event ready */ double time_; /* time at which event is ready */ scheduler_uid_t uid_; /* unique ID */ Event() : time_(0), uid_(0) {} };

10 class Handler { 11 public: 12 virtual ~Handler () {} 13 virtual void handle(Event* e) = 0; 14 };

This declaration forces all its instantiable derived classes to provide the action in function handle(e). In the following, we will discuss few classes which derive from classes Event and Handler. These classes are NsObject, Packet, AtEvent, and AtHandler.

4.2.2 Class NsObject: A Child Class of Class Handler Derived from class Handler, class NsObject is one of the main classes in NS2. It is a base class for most of the network components. We will discuss the details of this class in Chapter 5. Here, we only show the implementation of function NsObject::handle(e) in Program 4.2. Function NsObject::handle(e) casts an Event object associated with the input pointer (e) to a Packet object. Then it feeds the casted object to function recv(p) (Line 3). Usually, function recv(p), where p is a pointer to a packet, indicates that an object has received a packet p (see Chapter 5). Unless function handle(e) is overridden, function handle(e) (i.e., an action associated with an event *p) of an NsObject simply indicates packet reception. Program 4.2 Function NsObject::handle.
1 } //~/ns/common/object.cc void NsObject::handle(Event* e) 2 { 3 recv((Packet*)e); 4

72

4 Implementation of Discrete-Event Simulation in NS2

4.2.3 Classes Packet and AtEvent: Child Classes of Class Event Classes Packet and AtEvent are among key NS2 classes which derive from class Event. These two classes can be placed on the chain of events so that their associated handler will take actions at the ﬁring time. While the details of class AtEvent are discussed in this section, that of class Packet will be discussed later in Chapter 8. Program 4.3 Declaration of classes AtEvent and AtHandler, and function AtHandler::handle.
1 2 3 4 5 6 7 8 9 //~/ns/common/scheduler.cc class AtEvent : public Event { public: AtEvent() : proc_(0) { } ~AtEvent() { if (proc_) delete [] proc_; } char* proc_; };

10 class AtHandler : public Handler { 11 public: 12 void handle(Event* event); 13 } at_handler; 14 void AtHandler::handle(Event* e) 15 { 16 AtEvent* at = (AtEvent*)e; 17 Tcl::instance().eval(at->proc_); 18 delete at; 19 }

Declared in Program 4.3, class AtEvent represents events whose action is the execution of an OTcl statement. It contains one string variable proc_ (Line 8) which holds an OTcl statement string. At the ﬁring time, its associated handler, whose class is AtHandler, will retrieve and execute the OTcl string from this variable. Derived from class Handler, class AtHandler speciﬁes the actions to be taken at ﬁring time in its function handle(e) (Lines 14–19). Here, Line 16 casts the input event into an AtEvent object. Then Line 17 extracts and executes the OTcl statement from variable proc_ of the cast event. In the OTcl domain, an AtEvent object is placed in a chain of events at a certain ﬁring time by instproc “at{time statement}” of class Simulator. The syntax for the invocation is given below:

4.2 Events and Handlers

73

$ns at where ns is the Simulator object (see Section 4.4), is the ﬁring time, and is an OTcl statement string which will be executed when the simulation time is second. Program 4.4 Instance procedure at of class Simulator and command at of class Scheduler.
1 2 3 4 //~/ns/tcl/lib/ns-lib.tcl Simulator instproc at args { $self instvar scheduler_ return [eval $scheduler_ at $args] }

//~/ns/common/scheduler.cc 5 if (strcmp(argv[1], "at") == 0) { 6 /* t < 0 means relative time: delay = -t */ 7 double delay, t = atof(argv[2]); 8 const char* proc = argv[3]; 9 AtEvent* e = new AtEvent; 10 int n = strlen(proc); 11 e->proc_ = new char[n + 1]; 12 strcpy(e->proc_, proc); 13 delay = (t < 0) ? -t : t - clock(); 14 if (delay < 0) { 15 tcl.result("can’t schedule command in past"); 16 return (TCL_ERROR); 17 } 18 schedule(&at_handler, e, delay); 19 sprintf(tcl.buffer(), UID_PRINTF_FORMAT, e->uid_); 20 tcl.result(tcl.buffer()); 21 return (TCL_OK); 22 }

Program 4.4 shows the details of instproc at{...} of an OTcl class Simulator and an OTcl command at of class Scheduler. The instproc “at{...}” of class Simulator invokes an OTcl command “at” of the Scheduler object (See Lines 5–22). Command at of class Scheduler stores the ﬁring time in variable t (Line 7). Line 9 then creates an AtEvent object. Lines 8 and 10–12 store the input OTcl command in the variable proc_ of the created AtEvent object. Line 13 converts the ﬁring time to the delay time from the current time. Finally, Line 18 schedules the created AtEvent e at delay seconds in future, feeding the address of variable at_handler (see Program 4.3) as an input argument to function schedule(...).

74

4 Implementation of Discrete-Event Simulation in NS2

4.3 The Scheduler
The scheduler maintains the chain of events and simulation (virtual) time. At runtime, it moves along the chain, and dispatches one event after another. Since there is only one chain of events in a simulation, there is exactly one Scheduler object in a simulation. Hereafter, we will refer to the Scheduler object simply as the Scheduler. Also, NS2 supports the four following types of schedulers: List Scheduler, Heap Scheduler, Calendar Scheduler (default), and Real-time Scheduler. For brevity, we do not discuss the diﬀerences among all these schedulers here. The details of these schedulers can be found in [15]. Program 4.5 Declaration of class Scheduler.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 22 //~ns/common/scheduler.h class Scheduler : public TclObject { public: static Scheduler& instance() { return (*instance_); } void schedule(Handler*, Event*, double delay); virtual void run(); virtual void cancel(Event*) = 0; virtual void insert(Event*) = 0; virtual Event* lookup(scheduler_uid_t uid) = 0; virtual Event* deque() = 0; virtual const Event* head() = 0; double clock() const { return (); } virtual void reset(); protected: void dispatch(Event*); void dispatch(Event*, double); Scheduler(); virtual ~Scheduler(); int command(int argc, const char*const* argv); double clock_; static Scheduler* instance_; static scheduler_uid_t uid_; int halted_; };

4.3.1 Main Components of the Scheduler Declared in Program 4.5, class Scheduler consists of a few main variables and functions. Variable clock_ (Line 19) contains the current simulation time, and function clock() (Line 11) returns the value of the variable clock_. Variable halted_ (Line 22) is initialized to 0, and is set to 1 when the simulation is stopped or paused. Variable instance_ (Line 20) is the reference to the

4.3 The Scheduler

75

Scheduler, and function instance() (Line 3) returns the variable instance_. Variable uid_ is the event unique ID. In NS2, the Scheduler acts as a single point of unique ID management. When an event is inserted into the simulation timeline, the Scheduler creates a new unique ID, and assigns the ID to the event. Both the variables instance_ and uid_ are static, since there is only one scheduler and unique ID in a simulation. 4.3.2 Data Encapsulation and Polymorphism Concepts Program 4.5 implements the concepts of data encapsulation and polymorphism (see Appendix B). It hides the chain of events from the outside world, and declares pure virtual functions cancel(e), insert(e), lookup(uid), deque(), and head() in Lines 6–10 to manage the chain. Classes derived from class Scheduler provide implementation of the chain as well as all of the above functions. The beauty of this mechanism is the ease of modifying type of scheduler at runtime. NS2 implements most of the codes in relation to class Scheduler, not its derived classes (e.g., CalendarScheduler). At runtime (e.g., in a Tcl simulation script), we can select a scheduler to be of any derived class (e.g., CalendarScheduler) of class Scheduler without having to modify the codes for the base class (Scheduler). 4.3.3 Main Functions of the Scheduler Three main functions of class Scheduler are run() (Program 4.6), schedule( h,e,delay) (Program 4.7) and dispatch(p,t) (Program 4.8). In Program 4.6, function run() ﬁrst sets variable instance_ to the address of the scheduler (this) in Line 3. Then, it keeps dispatching events (Line 6) in the chain until halted_= 0 or untill all the events are executed (Line 5). Program 4.6 Function run of class Scheduler.
1 2 3 4 5 6 7 8 //~ns/common/scheduler.cc void scheduler::run() { instance_ = this; Event *p; while (!halted_ && (p = deque())) { dispatch(p, p->time_); } }

Function schedule(h,e,delay) in Program 4.7 takes three input arguments: A Handler pointer(h), an Event pointer(e), and the delay(delay), respectively. It inserts the input Event object(*e) into the chain of events.

76

4 Implementation of Discrete-Event Simulation in NS2

Lines 3–12 check for possible errors. Line 13 increments the unique ID of the Scheduler and assigns it to the input Event object. Line 14 associates the input Handler object with the input Event object. Line 15 converts input delay time (delay) to the ﬁring time (time_) of the Event object e. Line 17 inserts the conﬁgured Event object e in the chain of events via function insert(e). Since the scheduler increments its unique ID when invoking function schedule(...), every scheduled event will have diﬀerent unique ID. Finally, the errors in Lines 3–12 include 1. 2. 3. 4. Null handler (Line 3) Positive Event unique ID (Lines 4-7; See Section 4.3.4) Negative delay (Line 8) Negative Scheduler unique ID3

Program 4.7 Function schedule of class Scheduler.
//~ns/common/scheduler.cc 1 void Scheduler::schedule(Handler* h, Event* e, double delay) 2 { 3 if (!h) { /* error: Do not feed in NULL handler */ }; 4 if (e->uid_ > 0) { 5 printf("Scheduler: Event UID not valid!\n\n"); 6 abort(); 7 } 8 if (delay < 0) { /* error: negative delay */ }; 9 if (uid_ < 0) { 10 fprintf(stderr, "Scheduler: UID space exhausted!\n"); 11 abort(); 12 } 13 e->uid_ = uid_++; 14 e->handler_ = h; 15 double t = clock_ + delay; 16 e->time_ = t; 17 insert(e); 18 }

Function dispatch(p,t) in Program 4.8 is invoked by function run() at the ﬁring time. It takes a dispatching event (*p) and ﬁring time (t) as input arguments. Since the scheduler moves forward in the simulation time, the ﬁring time (t) cannot be less than the current simulation time (clock_). Line 3 will show an error, if t < clock_. Line 4 sets the current simulation virtual time to be the ﬁring time of the event. Line 5 inverts the sign of the
3

The unique ID of the Scheduler is always positive. Its negative value indicates possible abnormality such as memory overﬂow or inadvertent memory access violation.

4.3 The Scheduler

77

uid_ of the event, indicating that the is event is being dispatched. Line 6 invokes function handle(p) of the associated handler handler_, feeding the event (p) as an input argument. Program 4.8 Function dispatch of class Scheduler.
1 2 3 4 5 6 7 //~ns/common/scheduler.cc void Scheduler::dispatch(Event* p, double t) { if (t < clock_) { /* error */ }; clock_ = t; p->uid_ = -p->uid_; // being dispatched p->handler_->handle(p); // dispatch }

4.3.4 Dynamics of the Unique ID of an Event The dynamics of the event’s unique ID (uid_) is fairly subtle. In general, the scheduler maintains the unique ID, and assigns the unique ID to the event being scheduled. To make uid_ unique, the Scheduler increments uid_ and assigns the incremented uid_ to the scheduling event in its function schedule(...) (Line 13 in Program 4.7). When dispatching an event, the scheduler inverts the sign of uid_ of the dispatching event (Line 5 in Program 4.8). Figure 4.2 shows the dynamics of the unique ID caused by the above schedule(...) and dispatch(...) functions. The sign toggling mechanism of unique ID ensures that events will be scheduled and dispatched properly. If a scheduled event is not dispatched, or is dispatched twice, its unique ID will be positive, and an attempt to schedule this undispatched event will cause an error (Lines 5 and 6 in Program 4.7).

–2

Fig. 4.2. Dynamics of Event unique ID (uid) : Take a positive value from Scheduler::uid when being scheduled, and invert the sign when being dispatched. Increment upon schedule and inversion of sign upon dispatch

78

4 Implementation of Discrete-Event Simulation in NS2

4.3.5 Scheduling-Dispatching Mechanism We conclude this section through an example explaining the schedulingdispatching mechanism. Consider the following script set ns [new Simulator] $ns at 10 [puts "An event is dispatched"] $ns run which prints out the message “An event is dispatched” 10 seconds after the simulation has started. Figure 4.3 shows the functions (shown in rectangles) and objects (shown in rounded rectangles) related to the schedulingdispatching mechanism, whose names are shown in boldface font. Again, an AtEvent object is scheduled by the OTcl command “at” (in the upper-left rectangle), of class Scheduler. The Scheduler creates an AtEvent object e and stores input command (the fourth input argument str = puts "An event is dispatched") in e->proc_. Then, it schedules the event e with delay converted from time = 10 (the third input argument), feeding the address of AtHandler object (at_handler in the lower right rounded rectangle) as the corresponding handler. The lower-left rectangle in Fig. 4.3 shows the details of function schedule( h,e,delay) of class Scheduler. Before inserting event e into the chain of events, function schedule(...) conﬁgures event e as follows: Update uid_ to be the same as that of Scheduler, store at_handler in the handler of event e, and set ﬁring time to be clock_ (current time) + delay. At the ﬁring time, the scheduled AtEvent object is dispatched through function dispatch(p,t) (the upper-right rectangle in Fig. 4.3). When the scheduled Event object e4 is dispatched, function dispatch(...) inverts the sign of its variable uid_, and invokes function handle(e) of the corresponding handler feeding Event object e as an input argument. Since the handler is at_handler (see the upper-left rectangle), the OTcl command puts "An event is dispatched" stored in e is executed. 4.3.6 Null Event and Dummy Event Scheduling When being dispatched, an event p is fed to function handle(p) of the associated handler for a certain purpose. For example, the function handle(p) of class NsObject executes “recv(p), where “p” is a packet reception event. Here, the event *p must have been created and fed to function schedule(...) prior to the ongoing dispatching process. In some cases, an event only indicates the time where the default action is taken but takes no part in such the action. For example, a queue unblocking event, informs the associated Queue object of the completion of the ongoing
4

In Program 4.8, the ﬁrst argument of function dispatch is p. Here, we use e as the ﬁrst argument for the sake of explanation.

4.3 The Scheduler

79

dispatch( p , t ) command (argv = [ cmd, at, time, str] ) clock_=t p->uid_ = -p->uid_ p->handler_->handle(p)

AtEvent e e->proc_ = str schedule( &at_handler , e , delay )

AtHandler at_handler handle(e)

schedule( h , e , delay ) uid_++ Event uid_ invoke OTcl command stored in e->proc_

handler_

Clock_+( ) time_ insert(e) Scheduling Dispatching

Fig. 4.3. Scheduling and dispatching mechanism of an AtEvent.

transmission (see Section 7.3). Function handle(p) of the associated handler in this case simply invokes function resume() which take no input argument of the associated Queue object. Clearly the queue unblocking event takes no role in the dispatching process. In this case, we do not need to explicity create an event. Instead, we can use a null event or a dummy event as an input to function schedule(...). Scheduling of a Null Event Function schedule(h,e,delay) takes a pointer to an event as its second input argument. A null event refers to a null pointer which is fed as the second input argument to the function schedule(...) (e.g., schedule(handler,0,delay)). Although simple to use, a null event could lead to runtime error which is diﬃcult to be located. A null event is not an actual event. Its unique ID does not follow semantic in Fig. 4.3. The Scheduler ignores the unique ID when scheduling and dispatching a null event, and allows an undispatched event to be rescheduled. This breaks the scheduling-dispatching protection mechanism.

80

4 Implementation of Discrete-Event Simulation in NS2

Using null events, the users are responsible for ensuring the proper sequence of scheduling-dispatching by themselves. Scheduling of a Dummy Event This is another approach to schedule and dispatch events which do not take part in default actions. A dummy event is usually declared as a member variable of a C++ class, and is used repeatedly in a scheduling-dispatching process. Consider a packet departure event which is modeled by class LinkDelay (see Section 7.2) for example. During simulation, an NsObject informs a LinkDelay object to schedule packet departure events. At the ﬁring time, the packet completely departs the NsObject, and the NsObject is allowed to fetch another packet for transmission. The packet departure event takes no part in the default action, since a new packet is fetched or created by another object. As we shall see, a packet departure event is represented by a dummy event variable intr_ of class LinkDelay, and the packet departure is scheduled through the variable intr_ only. Since variable intr_ is a dummy Event, its unique ID follows the semantic in Fig. 4.3. An attempt to schedule an undispatched event would immediately cause runtime error. Note that intr_ is a variable of class LinkDelay. It is used over and over again to indicate packet departure from a LinkDelay object. As a ﬁnal note, under a simple conﬁguration, it is recommended to use the null event scheduling approach. For a complicated conﬁguration, on the other hand, the dummy event scheduling is preferable, since it provides a protection against scheduling of undispatched events.

4.4 The Simulator
OTcl and C++ classes Simulator are the main classes which supervise the entire simulation. Like the Scheduler object, there can be only one Simulator object throughout a simulation. This object will be referred to as the Simulator hereafter. The Simulator contains two types of key components: simulation objects and information-storing objects. While simulation object (e.g., the Scheduler) are the key components which derive the simulation, as well as the simulator are created during the Network Conﬁguration Phase, and will be used in the Simulation Phase. Information-storing objects (e.g., the reference to created nodes) contain information which is shared among several objects. For example, NS2 needs to know all created nodes and links in order to construct a routing table. These information-storing objects are created via various instprocs (e.g., Simulator::node{}) during the Network Conﬁguration Phase. In the Simulation Phase, most objects access these information-storing objects via its instvar ns_ (set by executing set ns_ [Simulator instance]), which is the reference to the Simulator.

4.4 The Simulator

81

4.4.1 Main Components of a Simulation Interperted Hierarchy Created by various instprocs, the main OTcl simulation components are as follows: • The Scheduler (scheduler_ created by instproc Simulator::init) maintains the chain of events and executes the events chronologically. • The null agent (nullAgent_ created by instproc Simulator::init) provides the common packet dropping point.5 • Node reference (Node_ created by instproc Simulator::node) is an associative array whose elements are the created nodes and indices are node IDs. • Link reference (link_ created by instprocs simplex-link{...} or duplexlink{...}) is an associative array. Associated with an index with format “sid:did”, each element of link_ is the created uni-directional link which carries packet from node “sid” to node “did”. Compiled Hierarchy In the compiled hierarchy, class Simulator also contains variables and functions as shown in Program 4.9. Variable instance_ (Line 18) is a pointer to the Simulator. It is a static variable, which means that there is only one variable instance_ of class Simulator for the entire simulation. Variable nodelist_ (Line 14) is the linked list containing the created nodes. The linked list can contain upto “size_” elements (Line 17), while the total number of nodes is “nn_” (Line 16). Variable rtobject_ (Line 15) is a pointer to a RouteLogic object, which is responsible for the routing mechanism (see Chapter 6). Function populate_flat_classifiers{...} (Line 7) pulls out the routing information stored in variable *rtobject_ and installs the routing table in the created nodes and links (see Section 6.6). Function add_node(...) (Line 8) puts the input argument node into the linked list of nodes (nodelist_). Function get_link_head(...) returns the link head object (see Chapter 7) of the link with ID “nh” which connects to a ParentNode object *node. Function node_id_by_addr(addr) (Line 10) converts node address "addr" to node ID. Function alloc(n) (Line 11) allocates spaces in nodelist_ which can accommodate up to “n” nodes, and clears all components of nodelist_ to NULL. Function check(n) immediately returns if n is less than size_. Otherwise, it will create more space in nodelist_, which can accommodate upto "n" nodes. Static function instance() in Line 3 returns the variable instance_ which is the pointer to the simulator.
5

By “dropping a packet”, we mean “removing a packet” from the simulation. We will discuss the dropping mechanism in Chapter 5. For the moment, it is suﬃcient to know that nullAgent drops or removes all received packets from the simulation.

82

4 Implementation of Discrete-Event Simulation in NS2

Program 4.9 Declaration of class Simulator.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 //~ns/common/simulator.h class Simulator : public TclObject { public: static Simulator& instance() { return (*instance_); } Simulator() : nodelist_(NULL), rtobject_(NULL), nn_(0), size_(0) {} ~Simulator() { delete []nodelist_;} int command(int argc, const char*const* argv); void populate_flat_classifiers(); void add_node(ParentNode *node, int id); NsObject* get_link_head(ParentNode *node, int nh); int node_id_by_addr(int address); void alloc(int n); void check(int n); private: ParentNode **nodelist_; RouteLogic *rtobject_; int nn_; int size_; static Simulator* instance_; };

4.4.2 Retrieving the Instance of the Simulator Program 4.10 Retrieving the instance of the Simulator using instproc instance of class Simulator.
//~ns/tcl/lib/ns-lib.tcl 1 Simulator proc instance {} { 2 set ns [Simulator info instances] 3 if { $ns != "" } { 4 return $ns 5 } 6 ... 7 }

From the interpreted hierarchy, we can also retrieve the simulator instance by invoking instproc instance{} of class Simulator. This instproc executes the OTcl built-in command “info” with an option “instances”. This execution returns all the instances of a certain class. Since there is only one Simulator instance, the statement “Simulator info instances” returns the Simulator object as required.

4.4 The Simulator

83

4.4.3 Simulator Initialization Simulator initialization refers to the process in the Network Conﬁguration Phase, which creates the Simulator as well as its components. The Simulator is created by executing new Simulator. This command invokes the constructor (i.e., instproc init{...} of class Simulator) shown in Program 4.11. Program 4.11 Instance procedures init and use-scheduler of class Simulator.
1 2 3 4 5 6 7 //~ns/tcl/lib/ns-lib.tcl Simulator instproc init args { $self create_packetformat $self use-scheduler Calendar $self set nullAgent_ [new Agent/Null] $self set-address-format def eval $self next $args }

8 Simulator instproc use-scheduler type { 9 $self instvar scheduler_ 10 if [info exists scheduler_] { 11 if { [$scheduler_ info class] == "Scheduler/$type" } { 12 return 13 } else { 14 delete $scheduler_ 15 } 16 } 17 set scheduler_ [new Scheduler/$type] 18 }

The constructor ﬁrst initializes the packet format in Line 2, and invokes instproc use-scheduler{type} in Line 3 to specify the type of the Scheduler. By default, the type of the Scheduler is Calendar. Line 4 creates a null agent (nullAgent). Line 5 sets the address format to the default format in Line 5. Instproc use-scheduler{type} (Lines 8–18) will delete the existing scheduler if it is diﬀerent from that speciﬁed in the input argument type. Then it will create a scheduler with type = type, and store the created Scheduler object in instvar scheduler_. 4.4.4 Running Simulation The Simulation Phase starts at the invocation of instproc Simulator::run{}. As shown in Program 4.12, the instproc Simulator::run{} ﬁrst invokes instproc “configure{}” of class RouteLogic (Line 2). This instproc computes the optimal routes and creates the routing table (see Chapter 6). Lines

84

4 Implementation of Discrete-Event Simulation in NS2

5–10 reset nodes and queues. Finally, Line 11 starts the Scheduler by invoking the OTcl command run{} of class Scheduler, which in turn invokes the C++ function run{} of class Scheduler shown in Program 4.6. Again, this function executes events in the chain of events one after another until the Simulator is halted (i.e., varaible halted_ of class Scheduler is 1), or untill all the events are executed. Program 4.12 Instance procedure Simulator::run.
//~/ns/tcl/lib/ns-lib.tcl 1 Simulator instproc run { 2 [$self get-routelogic] configure 3 $self instvar scheduler_ Node_ link_ started_ 4 set started_ 1 5 for each nn [array names Node_] { 6 $Node_($nn) reset 7 for each qn [array names link_] { 8 set q [$link_($qn) queue] 9 $q reset 10 } 11 return [$scheduler_ run] 12 }

4.5 Instprocs of OTcl Class Simulator
The list of useful instprocs of class Simulator is shown below. now{} nullagent{} use-scheduler{type} at{time stm} run{} halt{} cancel{e} Retrieve the current simulation time. Retrieve the shared null agent. Set the scheduler to be . Execute the statement at second. Start the simulation. Terminate the simulation. Cancel the scheduled event .

4.6 Chapter Summary
This chapter explains the details of discrete-event simulation in NS2. The simulation is carried out by running a Tcl simulation script, which consists of two parts. First, the Network Conﬁguration Phase establishes a network, and conﬁgures all simulation components. This phase also creates a chain of events by connecting the created events chronologically. Secondly, the Simulation

4.6 Chapter Summary

85

Phase chronologically executes (or dispatches) the created events until the Simulator is halted, or untill all the events are executed. There are four main classes involved in an NS2 simulation: • Class Simulator supervises the simulation. It contains simulation components such as the Scheduler, the null agent, etc. It also contains information storing objects which are share by other (simulation) components. • Class Scheduler maintains the chain of events and chronologically dispatches the events. • Class Event consists of the ﬁring time and the associated handler. Events are put together to form a chain of events, which are dispatched one by one by the Scheduler. Classes Packet and AtEvent are among the classes derived from class Event, which can be placed on the simulation timeline (i.e., in the chain of event). They are associated with diﬀerent handlers, and take diﬀerent actions at the ﬁring time. • Class Handler: Associated with an event, a handler speciﬁes default actions to be taken when the associated event is dispatched. Classes NsObject and AtHandler are among classes derived from class Handler. They are always associated with Packet and AtEvent events, respectively. Their actions are to receive the Packet object and to execute an OTcl statement speciﬁed in the AtEvent object, respectively.

5 Network Objects: Creation, Conﬁguration, and Packet Forwarding

NS2 is a simulation tool designed speciﬁcally for communication networks. The main functionalities of NS2 are to set up a network of connecting nodes and to pass packets from one node (which is a network object) to another. A network object is one of the main NS2 components, which is responsible for packet forwarding. NS2 implements network objects by using the polymorphism concept in Object-Oriented Programming (OOP). Polymorphism allows network objects to take diﬀerent actions ways under diﬀerent contexts. For example, a Connector object immediately passes the received packet to the next network object, while a Queue1 object enques the received packets and forwards only the head of the line packet. This chapter ﬁrst introduces the NS2 components by showing four major classes of NS2 components, namely, network objects, packet-related objects, simulation-related objects, and helper objects in Section 5.1. A part of the C++ class hierarchy, which is related to network objects, is also shown here. Section 5.2 presents class NsObject which acts as a template for all network objects. An example of network objects as well as packet forwarding mechanism are illustrated through class Connector in Section 5.3. Finally, the chapter summary is given in Section 5.4. Note that the readers who are not familiar with object-oriented programming are recommended to go through a review of the OOP polymorphism concept in Appendix B before proceeding further.

5.1 Overview of NS2 Components
5.1.1 Functionality-Based Classiﬁcation of NS2 Modules Based on the functionality, NS2 modules (or objects) can be classiﬁed into four following types:
1

Class Queue is a child class of class Connector.

T. Issariyakul, E. Hossain, Introduction to Network Simulator NS2, DOI: 10.1007/978-0-387-71760-9 5, c Springer Science+Business Media, LLC 2009

88

5 Network Objects

• Network objects are responsible for sending, receiving, creating, and destroying packet-related objects. Since these objects are those derived from class NsObject, they will be referred to hereafter as NsObjects. • Packet-related objects are various types of packets which are passed around a network. • Simulation-related objects control simulation timing, and supervise the entire simulation. As discussed in Chapter 4, examples of simulation-related objects are events, handlers, the Scheduler, and the Simulator. • Helper objects do not explicitly participate in packet forwarding. However, they implicitly help to complete the simulation. For example, a routing module calculates routes from a source to a destination, while network address identiﬁes each of the network objects. In this chapter, we focus only on network objects. Note that, the simulationrelated objects were discussed in Chapter 4. The packet-related objects will be discussed in Chapter 8. The main helper objects will be discussed in Chapter 12. 5.1.2 C++ Class Hierarchy This section gives an overview of C++ class hierarchies. The entire hierarchy consists of over 100 C++ classes and struct data types. Here, we only show a part of the hierarchy (in Fig. 5.1). The readers are referred to [16] for the complete class hierarchy.
OTcl Interface TclObject Handler Default Action

Simulator

PacketQueue

NsObject

AtHandler

QueueHandler

RoutingModule

Network Component

Classifier

Connector

LanRouter

Uni-directional Point-topoint Object Connector

Queue

Agent

ErrorModel

LinkDelay

Trace

Fig. 5.1. A part of NS2 C++ class hierarchy (this chapter emphasizes on classes in boxes with thick solid lines).

5.1 Overview of NS2 Components

89

As discussed in Chapter 3, all classes deriving from class TclObject form the compiled hierarchy. Classes in this hierarchy can be accessed from the OTcl domain. For example, they can be created by the global OTcl procedure “new{...}”. Classes derived directly from class TclObject include network classes (e.g., NsObject), packet-related classes (e.g., PacketQueue), simulation-related classes (e.g., Scheduler), and helper classes (e.g., RoutingModule). Again, classes which do not need OTcl counterparts (e.g., classes derived from class Handler) form their own standalone hierarchies. These hierarchies are not a part of the compiled hierarchy nor the interpreted hierarchy. As discussed in Chapter 4, class Handler speciﬁes an action associated with an event. Again, class Handler contains a pure virtual function handle(e) (see Program 4.1). Therefore, its derived classes are responsible for providing the implementation of function handle(e). For example, function handle(e) of class NsObject tells the NsObject to receive an incoming packet (Program 4.2), while that of class QueueHandler invokes function resume() of the associated Queue object (Lines 1–4 in Program 5.1; also see Section 7.3.2).

Program 5.1 Function handle of class QueueHandler.
//~/ns/queue/queue.cc 1 void QueueHandler::handle(Event*) 2 { 3 queue_.resume(); 4 }

Derived directly from class TclObject and Handler (see Program 5.2), class NsObject is the template class for all NS2 network objects. It inherits OTcl interfaces from class TclObject and the default action (i.e., function handle(e)) from class Handler. In addition, it deﬁnes a packet reception template, and forces all its derived classes to provide packet reception implementation. We will discuss the details of class NsObject in Section 5.2. There are three main classes deriving from class NsObject: Connector, Classifier, and LanRouter. Connecting two NsObjects, a Connector object immediately forwards a received packet to the connecting NsObject (see Section 5.3). Connecting an NsObject to several NsObjects, a Classifier object classiﬁes packets based on packet header (e.g., destination address, ﬂow ID), and forwards the packets with the same classiﬁcation to the same connecting NsObject (see Section 6.4). Class LanRouter also has multiple connecting NsObjects. However, it forwards every received packet to all connecting NsObjects.

90

5 Network Objects

5.2 NsObjects: A Network Object Template
5.2.1 Class NsObject Representing NsObjects, class NsObject is the base class for all network objects in NS2 (see the declaration in Program 5.2). Again, the main responsibility of an NsObject is to forward packets. Therefore, class NsObject deﬁnes a pure virtual function recv(p,h) (see Line 5 in Program 5.2) as a uniform packet reception interface to force all its derived classes to implement this function. Program 5.2 Declaration of class NsObject.
1 2 3 4 5 6 7 8 9 10 11 //~/ns/common/object.h class NsObject : public TclObject, public Handler { public: NsObject(); virtual ~NsObject(); virtual void recv(Packet*, Handler* callback = 0) = 0; virtual int command(int argc, const char*const* argv); protected: virtual void reset(); void handle(Event*); int debug_; };

Function recv(p,h) is in fact the very essence of packet forwarding mechanism in NS2. In NS2, an upstream object maintains a reference to the connecting downstream object. It passes a packet to the downstream object by invoking the function recv(p,h) of the downstream object and feeding the packet as an input argument. Since NS2 focuses mainly on forwarding packets in a downstream direction, NsObjects do not need to have a reference to its upstream objects. In most cases, NsObject conﬁguration involves downstream (not upstream) objects only. Function recv(p,h) takes two input arguments: a packet p to be received and a handler h. Most invocation of function recv(p,h) involves only packet “p”, not the handler.2 For example, a Queue object (see Section 7.3.3) puts the received packet in the buﬀer and transmits the packet at the head of the buﬀer. An ErrorModel object (see Section 12.3) imposes error probability on the received packet, and forwards the packet to the connecting object if the transmission is not in error.

2

We will discuss the callback mechanism which involves a handler in Section 7.3.3.

5.3 Connectors

91

Class NsObject derives from classes TclObject and Handler. Again, the functionality of class TclObject creates and binds the compiled shadow NsObject when an NsObject is created from the interpreted hierarchy. As a handler, an NsObject overrides function handle(e) which speciﬁes the default action taken at the ﬁring time of an associated event. Again, since the main responsibility of an NsObject is the packet forwarding, its function handle(e) (i.e., default action) is to receive a packet (cast from an event) through function recv(p,h) (see Program 4.2). 5.2.2 Packet Forwarding Mechanism of NsObjects An NsObject forwards packets in two following ways: • Immediate packet forwarding: To forward a packet to a downstream object, an upstream object needs to obtain a reference (e.g., a pointer) to the downstream object and invokes function recv(p,h) of the downstream object through the obtained reference. For example, a Connector (see Section 5.3) has a private pointer target_ to its downstream object. Therefore, it forwards a packet to its downstream object by invoking target_->recv(p,h). • Delayed packet forwarding: To delay packet forwarding, a Packet object is cast to be an Event object, associated with a packet receiving NsObject, and placed on the simulation timeline at a given simulation time. At the ﬁring time, function handle(e) of the NsObject will be invoked, and the packet will be received through function recv(p,h) (see an example of delayed packet forwarding in Section 5.3).

5.3 Connectors
As shown in Fig. 5.2, a Connector is a NsObject which connects three NsObjects in a uni-directional manner. It receives a from an upstream NsObject. By default, a Connector immediately forwards the received packet to its downstream NsObject. Alternatively, it can drop the packet by forwarding the packet to a packet dropping object.3 In NS2, each NsObject acts as a packet forwarder. Since it has no knowledge about its upstream objects, it does not have any interface to conﬁgure an upstream object. From Fig. 5.2, a Connector is interested in conﬁguring its downstream NsObject and packet dropping NsObject only. The connection from an upstream object to a Connector, on the other hand, must be conﬁgured from within the scope of the upstream object.

3

A packet dropping network object (e.g., a null agent) is responsible for destroying packets.

92

5 Network Objects
Connector NsObject Upstream NsObject target_
Packet forwarding path

NsObject Downstream NsObject

drop_
Packet dropping path

NsObject Packet Dropping NsObject

Fig. 5.2. Diagram of a connector. The solid arrows represent pointers, while the dotted arrows show packet forwarding and dropping paths.

Program 5.3 Declaration and function recv of class Connector.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 //~/ns/common/connector.h class Connector : public NsObject { public: Connector(); inline NsObject* target() { return target_; } void target (NsObject *target) { target_ = target; } virtual void drop(Packet* p); void setDropTarget(NsObject *dt) {drop_ = dt; } protected: virtual void drop(Packet* p, const char *s); int command(int argc, const char*const* argv); void recv(Packet*, Handler* callback = 0); inline void send(Packet* p, Handler* h){target_->recv(p, h);} NsObject* target_; NsObject* drop_; };

// drop target for this connector

//~/ns/common/connector.cc 17 void Connector::recv(Packet* p, Handler* h){send(p, h);}

5.3.1 Class Declaration Program 5.3 shows the declaration of class Connector. Class Connector contains two pointers (Lines 14–15 in Program 5.3) to NsObjects:4 target_ and
4

Since class Connector contains two pointers to abstract object (i.e., class NsObject), it can be regarded as an abstract user class for class composition discussed in Section B.8. We will discuss the details of how the class composition concept applies to a Connector in the next section.

5.3 Connectors

93

drop_. From Fig. 5.2, target_ is the pointer to the connecting downstream object, while drop_ is the pointer to the packet dropping object. Class Connector derives from the abstract class NsObject. It overrides the pure virtual function recv(p,h), by simply invoking function send(p,h) (see Line 12 in program 5.3). Function send(p,h) simply forwards the received packet to its downstream object by invoking function recv(p,h) of the downstream object (i.e., target_->recv(p,h) in Line 12). Program 5.4 Function Connector::drop.
1 2 3 4 5 6 7 //~/ns/common/connector.cc void Connector::drop(Packet* p) { if (drop_ != 0) drop_->recv(p); else Packet::free(p); }

Program 5.4 shows the implementation of function drop(p), which drops or destroys a packet. Function drop(p) takes one input argument, which is a packet to be dropped. If the dropping NsObject exists (i.e., drop_= 0), this function will forward the packet to the dropping NsObject by invoking drop_->recv(p,h). Otherwise, it will destroy the packet by invoking function Packet::free(p) (see Chapter 8). Note that function drop(p) is declared as virtual (Line 9). Hence, classes derived from class Connector may override this function without any function ambiguity5 . 5.3.2 OTcl Conﬁguration Commands As discussed in Section 4.1, NS2 simulation consists of two steps: Network Conﬁguration Phase and Simulation Phase. In the Network Conﬁguration Phase, a Connector is set up as shown in Fig. 5.2. Again, a Connector conﬁgures its downstream and packet dropping NsObjects only. Suppose OTcl has instantiated three following objects: a Connector object (conn_obj), a downstream object (down_obj), and a dropping object (drop_obj). Then, the Connector is conﬁgured using the following two OTcl commands (see Program 5.5): • OTcl command target with one input argument conforms to the following syntax: $conn_obj target $down_obj
5

Function ambiguity is discussed in Appendix B.2

94

5 Network Objects

Program 5.5 OTcl commands target and drop-target of class Connector.
//~/ns/common/connector.cc 1 int Connector::command(int argc, const char*const* argv) 2 { 3 Tcl& tcl = Tcl::instance(); 4 if (argc == 2) { 5 if (strcmp(argv[1], "target") == 0) { 6 if (target_ != 0) 7 tcl.result(target_->name()); 8 return (TCL_OK); 9 } 10 if (strcmp(argv[1], "drop-target") == 0) { 11 if (drop_ != 0) 12 tcl.resultf("%s", drop_->name()); 13 return (TCL_OK); 14 } 15 } 16 else if (argc == 3) { 17 if (strcmp(argv[1], "target") == 0) { 18 if (*argv[2] == ’0’) { 19 target_ = 0; 20 return (TCL_OK); 21 } 22 target_ = (NsObject*)TclObject::lookup(argv[2]); 23 if (target_ == 0) { 24 tcl.resultf("no such object %s", argv[2]); 25 return (TCL_ERROR); 26 } 27 return (TCL_OK); 28 } 29 if (strcmp(argv[1], "drop-target") == 0) { 30 drop_ = (NsObject*)TclObject::lookup(argv[2]); 31 if (drop_ == 0) { 32 tcl.resultf("no object %s", argv[2]); 33 return (TCL_ERROR); 34 } 35 return (TCL_OK); 36 } 37 } 38 return (NsObject::command(argc, argv)); 39 }

5.3 Connectors

95

This command casts the input argument down_obj to be of type NsObject* and stores it in variable target_ (Line 22). • OTcl command target with no input argument (e.g., $conn_obj target) returns OTcl instance corresponding to the C++ variable target_ (Line 5–9). Note that function name() of class TclObject returns the OTcl reference string associated with the input argument. • OTcl command drop-target with one input argument is very similar to that of OTcl command target but the input argument is cast and stored in the variable drop_ instead of the variable target_. • OTcl command drop-target with no input argument is very similar to that of OTcl command target but it returns the OTcl instance corresponding to the variable drop_ instead of the variable target_. Example 5.1. Consider the connector conﬁguration in Fig. 5.3. Let the downstream object be of class TcpAgent, which corresponds to class Agent/Tcp in the OTcl domain. Also, let a Agent/Null object be a packet dropping NsObject. The following code shows how the network is set up from the OTcl domain: set conn_obj [new Connector] set tcp [new Agent/TCP] set null [new Agent/Null] $conn_obj target $tcp $conn_obj drop-target $null The ﬁrst three lines create a Connector (conn), a TCP object (tcp), and a packet dropping object (null). The last two lines use the OTcl commands target and drop-target to set tcp and null as the downstream object and the dropping object of the Connector, respectively.
NsObject Upstream network component Connector
By declaration

NsObject recv(p,h)=0;

target_ recv(p,h) { … };

Implementation by polymorphism

TcpAgent recv(p,h){...};

Fig. 5.3. A polymorphism implementation of a Connector. A Connector declares target as an NsObject pointer. In the Network Conﬁguration Phase, the OTcl command target is invoked to setup a downstream object of the Connector, and the NsObject *target is cast to a TcpAgent object.

casting

96

5 Network Objects

Connector conﬁguration complies with the class composition programming concept discussed in Appendix B.5. Table 5.1 shows the components in Example 5.1 and the corresponding class composition. Classes Agent/TCP and Agent/Null are OTcl classes whose corresponding C++ classes derive from class NsObject. Class Connector stores pointers (i.e., target_ and drop_) to NsObjects, and is therefore considered to be an abstract user class. Finally, as a user class, the Tcl Simulation Script instantiates NsObjects conn, tcp, and null from classes Connector, Agent/Tcp, and Agent/Null, respectively, and binds tcp and null to variables target_ and drop_, respectively.
Table 5.1. Class composition of network components in Example 5.1. Abstract class Derived class Abstract user class User class NsObject Agent/Tcp and Agent/Null Connector A Tcl Simulation Script

When invoking “target” and “drop-target”, tcp and null are ﬁrst type-cast to NsObject pointers. Then they are assigned to target_ and to drop_, respectively. Since a virtual function is unaﬀected by type casting, function recv(p,h) of both tcp and null are associated to class Agent/TCP and Agent/Null, respectively. 5.3.3 Packet Forwarding Mechanism From Section 5.2.2, an NsObject forwards a packet in two ways: immediate and delayed packet forwarding. This section demonstrates both the packet forwarding mechanisms through a Connector. Immediate Packet Forwarding Immediate packet forwarding is carried out by invoking function recv(p,h) of a downstream object. In Example 5.1, the Connector forwards a packet to the TCP object by invoking function recv(p,h) of the TCP object (i.e., target_->recv(p,h), where target_ is conﬁgured to be a TCP object). C++ polymorphism is responsible for associating function recv(p,h) to class Agent/TCP (i.e., the construction type), not NsObject (i.e., the declaration type). Delayed Packet Forwarding Delayed packet forwarding is implemented with the aid of the Scheduler. Here, a packet is cast to an event, associated with a receiving NsObject, and placed

5.3 Connectors

97

on the simulation timeline. For example, to delay packet forwarding in Example 5.1 for “d” seconds, we may invoke the following statement instead of target_->recv(p,h). Scheduler& s = Scheduler::instance(); s.schedule(target_, p, d); Consider Fig. 5.4 and Program 5.6 altogether. Figure 5.4 shows the diagram of delayed packet forwarding, while Program 5.6 shows the details of functions schedule(h,e,delay) as well as dispatch(p,t) of class Scheduler. When “schedule(target_, p, d)” is invoked, function schedule (...) casts packet *p and the NsObject *target_ into Event and Handler objects, respectively (Line 1 of Program 5.6). Line 5 of Program 5.6 associates packet *p with the NsObject *target_. Lines 6-7 insert packet *p into the simulation timeline at the appropriate time. At the ﬁring time, the event (*p) is dispatched (Lines 9-14). The Scheduler invokes function handle(p) of the handler associated with event *p. In this case, the associated handler is the NsObject *target_. Therefore, in Line 13, the default action handle(p) of target_, invokes function recv(p,h) to receive the scheduled packet.

Fig. 5.4. Delayed packet forwarding.

98

5 Network Objects

Program 5.6 Functions schedule and dispatch of class Scheduler.
1 2 3 4 5 6 7 8 //~/ns/common/scheduler.cc void Scheduler::schedule(Handler* h, Event* e, double delay) { ... e->uid_ = uid_++; e->handler_ = h; e->time_ = clock_ + delay; insert(e); }

9 void Scheduler::dispatch(Event* p, double t) 10 ... 11 clock_ = t; 12 p->uid_ = -p->uid_; // being dispatched 13 p->handler_->handle(p); // dispatch 14 }

5.4 Chapter Summary
Referred to as an NsObject, a network object is responsible for sending, receiving, creating, and destroying packets. As an object of class NsObject, it derives OTcl interfaces from class TclObject and the default action (i.e., function handle(e)) from class Handler. It deﬁnes a pure virtual function recv(p,h) as a uniform packet reception interface for all its derived classes. Based on the polymorphism concept, all its derived classes must provide their own implementation of how to receive a packet. In NS2, an NsObject needs to create a connection to its downstream object only. Normally, an NsObject forwards a packet to a downstream object by invoking function recv(p,h) of its downstream object. In addition, an NsObject can defer packet forwarding by associating a packet to the downstream object and inserting the packet on the simulation timeline. At the ﬁring time, the scheduler dispatches the packet, and the default action of the downstream object is invoked to receive the packet. As an example, we show the details of class Connector, one of the main NsObject classes in NS2. Class Connector contains two pointers to NsObjects: target_ pointing to a downstream object and drop_ pointing to a packet dropping object. To conﬁgure a Connector, an object whose class derives from class NsObject can be set as downstream and dropping objects via OTcl command target and drop-target, respectively. These two OTcl commands cast the downstream and dropping objects to NsObjects, and assign them to C++ variables *target_ and *drop_, respectively.

6 Nodes as Routers or Computer Hosts

This chapter focuses on a basic network component, Node. In NS2, a Node acts as a computer host (e.g., a source or a destination) and a router (e.g., an intermediate node). It receives packets from an attached application or an upstream object, and forwards them to the attached links speciﬁed in the routing table (as a router) or delivers them to the ports speciﬁed in the packet header (as a host). In the following, we ﬁrst give an overview of Nodes and routing mechanism in NS2 in Sections 6.1 and 6.2, respectively. Sections 6.3, 6.4, and 6.5 discuss three main routing components: Route logic, classiﬁers, and routing modules, respectively. In Section 6.6 we show how the aforementioned Node components are assembled to compose a Node. Finally, the chapter summary is provided in Section 6.7.

6.1 An Overview of Nodes in NS2
A Node plays two important roles in NS2. As a router, it forwards packets to the connecting link based on a routing table. As a host, it delivers packets to the transport layer agent attached to the port speciﬁed in the packet header. NS2 conﬁgures the connection to its downstream NsObjects only. A Node does not need to have a connection to its upstream NsObject (e.g., a sending transport agent or an upstream link). Instead, its upstream NsObject will create a connection to the Node. 6.1.1 Architecture of a Node In the OTcl domain, a Node is deﬁned in a C++ class Node which is bound to an OTcl class with the same name. Unless speciﬁed otherwise, this chapter deals with the OTcl class only. A Node is a composite object whose architecture is shown in Fig. 6.1. It provides a single point of packet entrance,

T. Issariyakul, E. Hossain, Introduction to Network Simulator NS2, DOI: 10.1007/978-0-387-71760-9 6, c Springer Science+Business Media, LLC 2009

100

6 Nodes as Routers or Computer Hosts
NODE PortClassifier (dmux_) Node Entry (entry_) AddrClassifier (classifier_) Agent Agent Agent

Link

Link

Fig. 6.1. Node architecture.

entry_ (which is a Connector object). After entering the Node entry, the packet enters an address classiﬁer (an instvar classifier_). If the Node is not the ﬁnal destination, the address classiﬁer will forward the packet to the link speciﬁed in the routing table. Otherwise, it will forward the packet to the demultiplexer or port classiﬁer (an instvar dmux_), which forwards the packet to the agent attached to the port speciﬁed in the packet header. Apart from the above packet forwarding components, a Node also has other components. The list of major Node OTcl components is given below. id_ Node ID agents_ List of attached transport layer agents nn_ Total number of Nodes (a static class instvar belonging to an OTcl class Node) neighbor_ List of neighboring nodes nodetype_ Node type (e.g., regular node or mobile node) ns_ Simulator dmux_ Demultiplexer or port classiﬁer module_list_ List of enabled routing modules reg_module_ List of registered routing modules rtnotif_ List of routing modules which will be notiﬁed of route updates ptnotif_ List of routing modules which will be notiﬁed of port attachment/detachment hook_assoc_ Sequence of the chain of classiﬁers mod_assoc_ Association of classiﬁers and routing modules, whose indexes and values are classiﬁers and the associated routing modules, respectively.

6.1 An Overview of Nodes in NS2

101

6.1.2 Related Instproc of Class Node An OTcl class Node deﬁnes the following main instprocs, which can be classiﬁed into three categories. Initialization Instprocs enable-module{mod_name} Appends “mod_name” to the module list “module_list_”. disable-module{mod_name} Removes “mod_name” from the module list “module_list_”. register-module{mod} Inserts an input routing module “mod” into an entry of the instance associative array “reg_module_” whose index in the module name. unregister-module{mod} Removes an entry of an instance associative array “reg_module_” whose index matches with the name of the input routing module “mod”. route-notify{module} Inserts an input routing module “module” into the route notiﬁcation list “rtnotif_”. unreg-route-notify{... Removes a routing module “module” from module} the route notiﬁcation list “rtnotif ” port-notify{module} Inserts an input routing module “module” into the agent attachment list “ptnotif_”. unreg-port-notify{... Removes an input routing module “module” module} from the agent attachment list “ptnotif ”. Route Adding/Deleting and Agent Attachment/Detachment Instprocs add-route{... Recursively adds a routing entry (dst,target), dst target} where “dst” and “target” are a destination node and a forwarding NsObject, respectively, for all routing modules in the link list “rtnotif ”. delete-route{args} Recursively removes a routing entry speciﬁes in the input arguments from all routing modules in the linked list “rtnotif_”. alloc-port{... Returns a free port of the demultiplexer “dmux ” of nullagent} the Node. agent{port} Returns the agent whose port is “port”. add-target{... Recursively attaches the input agent “agent” to agent port} the port “port” of the demultiplexer “dmux ” associated with all routing modules in the instvar “ptnotif ”

102

6 Nodes as Routers or Computer Hosts

attach{agent port} Attaches an input agent “agent” to the port “port” of the Node; Sets up necessary instvars and invoke instprocs “add-target” to install the input agent “agent” in slot “port” of the demultiplexer “dmux_” associated with all routing modules in the instvar “ptnotif_”. detach{agent ... Recursively detaches an input agent “agent” from nullagent} the demultiplexer “dmux ” associated with all routing modules in “ptnotif ”. Replaces the “agent” installed in the demultiplexer with the input null agent “nullagent”. Classiﬁer Manipulation Instprocs insert-entry{module... Inserts an input classiﬁer “clsfr” as the clsfr hook} head (i.e., the ﬁrst) classiﬁer connecting from the Node entry, and installs the existing (if any) head classiﬁer in the slot “hook” of the classiﬁer “clsfr”. Also, updates the instvars “hook assoc ” and “mod assoc ” accordingly. install-entry{module... Does what the instproc “insert-entry” clsfr hook} does. Also destroy the existing head classiﬁer if any. install-demux{demux... Replaces the existing demultiplexer “dmux ” port} with the input demultiplexer “demux”. If “port” is an integer, installs the existing demultiplexer “dmux ” in the slot “port” of “demux”. mk-default-classifier{} Creates classiﬁers and routing modules speciﬁed in the instvar “module_list_”, and associates them to the Node. 6.1.3 Default Nodes and Node Conﬁguration Interface A default NS2 Node is based on ﬂat-addressing and static routing. With ﬂataddressing, an address of every new node is incremented by one from that of the previously created node. Static routing assumes no change in topology. The routing table is computed once at the beginning of the Simulation phase and does not change thereafter. By default, NS2 employs the Dijkstra shortest path algorithm [17] to compute optimal routes for all pairs of Nodes. The details about other routing protocols as well as hierarchical addressing can be found in the NS manual [15]. To provide a default Node with more functionalities such as link layer or Medium Access Control (MAC) protocol functionalities, we may use instproc node-config of class Simulator whose syntax is as follows:

6.2 Routing Mechanism in NS2

103

$ns node-config - [] where $ns is the Simulator object. This instproc does not immediately conﬁgure the Nodes as speciﬁed in the . Instead, it stores in the instvars of the Simulator corresponding to . This stored conﬁguration will be used during a Node construction process. Therefore, this instproc must be executed prior to the Node construction. An example use of the instproc node-config{args} for the default setting is shown below: $ns_ node-config -addressType flat -adhocRouting -llType -macType -propType -ifqType -ifqLen -phyType -antType -channel -channelType -topologyInstance By default, almost every option is speciﬁed as NULL with the exception of addressType, which is set to be ﬂat addressing. Another important option reset is used to restore default parameter setting: $ns node-config -reset The details of instproc node-config (e.g., other options) can be found in the ﬁle ˜ns/tcl/lib/ns-lib.tcl and [15].

6.2 Routing Mechanism in NS2
In general, a Node may connect to several downstream NsObjects (i.e., targets). As a router, it needs to select one of the downstream NsObjects as a forwarding NsObject for each incoming packet. In most cases, this process is carried out using a so-called routing table each row of which is called a routing entry. A routing entry speciﬁes a forwarding NsObject for a packet which matches a predeﬁned criterion. For example, (dst,target) speciﬁes that, a packet whose destination address is dst, must be forwarded to a forwarding NsObject target. The routing mechanism in NS2 consists of four main components: • Routing agent: collects information (e.g., the network topology) needed to compute a routing table.

104

6 Nodes as Routers or Computer Hosts

• Route logic: uses the information collected by the routing agent, and compute the routing table. • Classiﬁer: employs the computed routing table for packet forwarding. • Routing module: acts as a single point of management of a group of classiﬁers in a Node. It takes conﬁguration commands from a routing agent, a route logic, and a Node, and propagates them to relevant classiﬁers. In this book we focus on static routing, where routing agents are not involved in the routing process. Therefore, we omit the details of routing agents hereafter (the details of which can be found in [15]). Figure 6.2 shows the routing components in NS2. Each box in this ﬁgure represents an object whose type is indicated on the top, while each word in a box represents an instproc of the corresponding object. The arrow shows the sequence of instproc invocation (details of the instprocs will be shown later in this chapter). For example, the instproc new{...} of the Node invokes the insproc register{proto args} of the routing module. Depending on their functionality, the above four routing components are stored in diﬀerent simulation objects. A route logic computes the routing table for every node. It is shared by several simulation objects, and is therefore stored in the Simulator. Acting as a routing table, an address classiﬁer is speciﬁc to and is hence stored in a Node. A routing module is an interface to all the routing components of a Node. Hence, it is stored as an instvar of a Node. Next, we will discuss the details of route logic, classiﬁers, and routing module in Sections 6.3, 6.4, and 6.5, respectively. Then, in Section 6.6, we will revisit NS2 routing mechanism, and discuss how the above routing components are conﬁgured in a Node.
SIMULATOR NODE ROUTING MODULE new attach add-route register attach add-route CLASSIFIER install

attach-agent

ROUTE LOGIC compute configure

Fig. 6.2. Conﬁguration of routing components in NS2.

6.3 Route Logic

105

6.3 Route Logic
The main responsibility of a route logic object is to compute the routing table. Route logic is implemented in a C++ class RouteLogic which is bound to the OTcl class with the same name (see Program 6.1). Class RouteLogic has two key variables: “adj_”, which is the adjacency matrix used to compute the routing table, and “route_”, which is the routing table. It has the following three main functions: Program 6.1 Declaration of class RouteLogic and the corresponding OTcl mapping class.
1 2 3 4 5 7 8 9 10 11 12 13 14 15 16 //~/ns/routing/route.h class RouteLogic : public TclObject { public: RouteLogic(); ~RouteLogic(); int command(int argc, const char*const* argv); virtual int lookup_flat(int sid, int did); protected: void reset(int src, int dst); void reset_all(); void compute_routes(); void insert(int src, int dst, double cost); void insert(int src, int dst, double cost, void* entry); adj_entry *adj_; route_entry *route_; };

//~/ns/routing/route.cc 17 class RouteLogicClass : public TclClass { 18 public: 19 RouteLogicClass() : TclClass("RouteLogic") {} 20 TclObject* create(int, const char*const*) { 21 return (new RouteLogic()); 22 } 23 } routelogic_class;

106

6 Nodes as Routers or Computer Hosts

insert(src,... Inserts a new entry including a source ID (src), a dst,cost) destination ID (dst), and the corresponding routing cost (cost) into the adjacency matrix. compute_route() Uses the adjacency matrix adj_ to compute the optimal routes for all source-destination pairs and store the computed routes in the variable route_. lookup_flat(... Searches within variable route for an entry with sid,did) matching source ID (sid) and destination ID (did), and returns the index of the forwarding object (e.g., connecting link).

Program 6.2 Instprocs register, configure and lookup of class RouteLogic.
1 2 3 4 5 6 7 8 //~/ns/tcl/lib/ns-route.tcl RouteLogic instproc register {proto args} { $self instvar rtprotos_ node_rtprotos_ default_node_rtprotos_ if [info exists rtprotos_($proto)] { eval lappend rtprotos_($proto) $args } else { set rtprotos_($proto) $args } }

9 RouteLogic instproc configure {} { 10 $self instvar rtprotos_ 11 if [info exists rtprotos_] { 12 foreach proto [array names rtprotos_] { 13 eval Agent/rtProto/$proto init-all $rtprotos_($proto) 14 } 15 } else { 16 Agent/rtProto/Static init-all 17 } 18 } 19 RouteLogic instproc lookup { nodeid destid } { 20 if { $nodeid == $destid } { 21 return $nodeid 22 } 23 set ns [Simulator instance] 24 set node [$ns get-node-by-id $nodeid] 25 $self cmd lookup $nodeid $destid 26 }

6.4 Classiﬁers: Multi-target Packet Forwarders

107

In the interpreted hierarchy, the OTcl class RouteLogic has two major instprocs to conﬁgure the route logic and one major instproc to query the routing information (see Program 6.2). register{... Stores a routing agent as an element of the proto,args} instance associative array rtprotos whose index is . configure Reads instvar rtprotos_ and invokes instproc init-all of all registered routing agents to create routing tables. lookup{... Looks in the routing table for the forwarding object nodeid destid} corresponding to the input source and destination pair (nodeid,destid). Returns nodeid (Line 11) if nodeid=destid. Otherwise, returns the forwarding object returned from the function lookup flat of the C++ class RouteLogic.

6.4 Classiﬁers: Multi-target Packet Forwarders
A classiﬁer is a packet forwarding object with multiple connecting target. It forwards incoming packets whose header matches with a certain criterion (e.g., same destination host) to the same forwarding NsObject. Similar to a Connector, a classiﬁer identiﬁes each target using a pointer. It installs each of these pointers so-called slots. Based on a predeﬁned criterion, a classiﬁer selects a slot for each incoming packet, and forwards the packet to the NsObject whose pointer is installed in that slot. In this section, we will explain the packet forwarding mechanism, the internal variables and functions, and the conﬁguration interface of the classiﬁers. The process of assembling classiﬁers and composing a Node will be discussed in Section 6.6. 6.4.1 Class Classifier and Its Main Components NS2 implements classiﬁers in a C++ class Classifier (see the declaration in Program 6.3), which is bound to an OTcl class with the same name. The main components of a classiﬁer include the following. C++ Variables The C++ class Classifier has two key variables: slot_ and default_ target_. The variable slot_ (Line 13 in Program 6.3) is a linked list of pointers whose entries are a pointer a to downstream NsObjects. Each of these NsObjects corresponds to a predeﬁned criterion. Packets matched with a predeﬁned criterion are forwarded to the corresponding NsObject. Class Classifier also deﬁne another pointer to an NsObject, default_target_. The variable default_target_ points to a receiving NsObject for packets which do not match with any predeﬁned criterion.

108

6 Nodes as Routers or Computer Hosts

Program 6.3 Declaration of class Classifier.
//~/ns/classifier/classifier.h 1 class Classifier : public NsObject { 2 public: 3 Classifier(); 4 virtual ~Classifier(); 5 virtual void recv(Packet* p, Handler* h); 6 virtual NsObject* find(Packet*); 7 virtual int classify(Packet *); 8 virtual void clear(int slot); 9 virtual void install(int slot, NsObject*); 10 inline int mshift(int val) {return((val >> shift_) & mask_);} 11 protected: 12 virtual int command(int argc, const char*const* argv); 13 NsObject** slot_; 14 NsObject *default_target_; 15 int shift_; 16 int mask_; 17 };

The class Classifier also have two supplementary variables: shift_ (Line 15) and mask_ (Line 16). These two variables are used in function mshift(val) (Line 10) to reformat the address (see also Section 12.4). C++ Functions The main C++ functions of class Classifier can be classiﬁed into packet forwarding functions (i.e., recv(p,h), find(p), and classify(p)) and conﬁguration functions (i.e., install(slot,p), install_next(node), do_install( dst,target), and clear(slot)). recv(p,h) Receives a packet *p and handler *h. find(p) Returns a forwarding NsObject pointer for an incoming packet *p. classify(p) Returns a slot number of an entry which match with the header of an incoming packet *p. install(slot,p) Stores the input NsObject pointer “p” in the slot number “slot” of the variable slot_. install_next(node) Installs the NsObject pointer “node” in the next available slot. do_install(... Installs an input NsObject pointer target in the dst,target) slot number dst. clear(slot) Removes the NsObject pointer installed in the slot number “slot”.

6.4 Classiﬁers: Multi-target Packet Forwarders

109

mshift(val) Shifts val to the left by “shift_” bits. Masks the shifted value by using a logical AND (&) operation with “mask_”. As an NsObject, a classiﬁer receives a packet by having its upstream object invoke its function recv(p,h), passing the packet “*p” and a handler “*h” as input arguments. In Program 6.4, Line 3 retrieves for an NsObject pointer “node” for an incoming packet “*p” by invoking function find(*p). Then, Line 8 passes the packet“*p” and the handler “*h” to its forwarding NsObject *node by executing node->recv(p,h).

Program 6.4 Functions recv and find of class Classifier.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 //~/ns/classifier/classifier.cc void Classifier::recv(Packet* p, Handler* h) { NsObject* node = find(p); if (node == NULL) { Packet::free(p); return; } node->recv(p,h); } NsObject* Classifier::find(Packet* p) { NsObject* node = NULL; int cl = classify(p); if (cl < 0 || cl >= nslot_ || (node = slot_[cl]) == 0) { /*There is no potential target in the slot;*/ } return (node); }

Function find(p) (Lines 10–18 in Program 6.4) examines the incoming packet *p, and retrieves the matched NsObject pointer installed in the variable slot_. Line 13 invokes function classify(p) to retrieve the slot number (cl) corresponding to the packet *p. Then, Lines 14 and 17 return the NsObject pointer (i.e., node) stored in slot cl of variable slot_. Function classify(p) is perhaps the most important function of a classiﬁer. This is the place where the classiﬁcation criterion is deﬁned. The function classify(p) returns an NsObject pointer installed in the slot whose criterion matches with the input packet *p. Since classiﬁcation criteria could be diﬀerent for diﬀerent types of classiﬁers, the function classify(p) is usually overridden in the derived classes of class Classifier. In Sections 6.4.2 and

110

6 Nodes as Routers or Computer Hosts

6.4.3, we will show two example implementations of function classify(p) in classes HashClassifier and PortClassifier, respectively.

Program 6.5 Functions clear, install, and install next of class Classifier.
1 2 3 4 5 6 7 8 //~ns/classifier/classifier.cc void Classifier::install(int slot, NsObject* p) { if (slot >= nslot_) alloc(slot); slot_[slot] = p; if (slot >= maxslot_) maxslot_ = slot; }

9 int Classifier::install_next(NsObject *node) { 10 int slot = maxslot_ + 1; 11 install(slot, node); 14 12 return (slot); 13 } 14 void Classifier::clear(int slot) 15 { 16 slot_[slot] = 0; 17 if (slot == maxslot_) 18 while (--maxslot_ >= 0 && slot_[maxslot_] == 0); 19 } //~ns/classifier/classifier.h 20 virtual void do_install(char* dst, NsObject *target) { 21 int slot = atoi(dst); 22 install(slot, target); 23 }

Consider Program 6.5. Function install(slot,p) stores the input NsObject pointer “p” in the slot number “slot” of the variable “slot_” (Line 5), and updates the variable maxslot_ (the total number of slots) if necessary. Function install_next(node) installs the input NsObject pointer “node” in the next available slot (Lines 10–11). Function do_install(dst,target) converts dst to be an integer variable (Line 21), and installs the NsObject pointer target in the slot corresponding to dst (Line 22). Finally, function clear(slot) removes the installed NsObject pointer from the slot number “slot” of the variable slot_ (Line 6).

6.4 Classiﬁers: Multi-target Packet Forwarders

111

Deﬁned in Line 10 of Program 6.3, function mshift(val) simply returns val. The constructor of class Classifier sets the default values of shift_ and mask_ to be zero and 0xffffffff. The function mshift(val) shifts the input argument val by zero bit. Also, the logical AND with 0xffffffff leaves the input argument unchanged. Hence, function mshift(val) of class Classifier has no eﬀect on the input argument val. OTcl Commands Class Classifier also deﬁnes the following key OTcl commands in a C++ function command of class Classifier. These OTcl command can be invoked from the OTcl domain. slot{index} Returns the NsObject stored in the slot number index clear{slot} Clears the NsObject pointer installed in the slot number slot. install{index object} Installs object in the slot number index. installNext{object} Installs object in the next available slot. 6.4.2 Hash Classiﬁers An Overview of Hash Classiﬁers Hash table is a data structure which facilitate a key-value lookup process1 . It eliminates the need to sequentially search for a matched key and retrieve the corresponding value. A hash table uses a hash function to transform a hash key into a hash index, and stores the corresponding hash value in an array entry (i.e., a record of the hash table) whose index corresponding to the hash index. Given a hash key, the search process transforms the hash key into a hash index using a hash function, and directly accesses the array entry corresponding to the hash index. Since a hash function has low complexity, the search time when using hash table is usually much smaller than that when using a sequential search. In NS2, a hash classiﬁer classiﬁes packets based on a hash table. Table 6.1 shows an example of hash tables used for a hash classiﬁer. Here, each row of the hash table is called a hash record. A hash value is the slot number. A hash key has three components: Flow ID, source address, and destination address. A hash classiﬁer examines the header of an incoming packet, searches in the hash table for a hash entry whose key matches with information provided in the packet header, and returns the hash value (i.e., slot number) of the matched entry. From Table 6.1, the hash classiﬁer returns the slot number 1 for a packet
1

Suppose we have a table which associates keys and values. The objective of a key-value lookup process is as follows. Given a key, search in the table for the matched key and return the corresponding value.

112

6 Nodes as Routers or Computer Hosts Table 6.1. An example of hash table.

Slot number 1 2 . . .

Flow ID 1 1 . . .

Source address 1 1 . . .

Destination address 1 2 . . .

with (ﬂow ID, source address, destination address) = (1,1,1), and returns 2 for a packet with (ﬂow ID, source address, destination address) = (1,1,2). Implementation of Hash Classiﬁer in NS2 Hash classiﬁer is declared in a C++ class HashClassifier in the compiled hierarchy (Program 6.6), and mapped to an OTcl class Classifier/Hash in the interpreted hierarchy. Program 6.6 Declaration of class HashClassifier.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 //~ns/classifier/classifier-hash.h class HashClassifier : public Classifier { public: HashClassifier(int keylen): default_(-1), keylen_(keylen); ~HashClassifier(); virtual int classify(Packet *p); virtual long lookup(Packet* p) ; void set_default(int slot) { default_ = slot; } protected: long lookup(nsaddr_t src, nsaddr_t dst, int fid); void reset(); int set_hash(nsaddr_t src, nsaddr_t dst, int fid, long slot); long get_hash(nsaddr_t src, nsaddr_t dst, int fid); virtual int command(int argc, const char*const* argv); virtual const char* hashkey(nsaddr_t, nsaddr_t, int)=0; int default_; Tcl_HashTable ht_; int keylen_; };

Declared in Program 6.6, the class HashClassifier has three main variables. First, variable default_ (Line 15) contains the default slot for a packet which does not match with any entry in the table. Secondly, variable ht_ (Line 16) is the hash table. Finally, variable keylen_ (Line 17) is the total number of hash keys. By default, the hash keys include ﬂow ID, source address, and destination address, and the variable keylen_ is 3.

6.4 Classiﬁers: Multi-target Packet Forwarders

113

Apart from function classify(p) derived from class Classifier, class HashClassifier deﬁnes the following functions (see the function declaration in Program 6.6): lookup(p) Returns the slot number of the entry which matches with the incoming packet “p” (Line 6). lookup(src,... Returns the slot number of the entry whose source dst,fid) address, destination address, and ﬂow ID are “src”, “dst”, and “fid”, respectively. (Line 9). set_hash(src,... Inserts an entry with source address “src”, destination dst,fid,slot) address “dst”, and ﬂow ID “fid” to the hash table, and associates the entry to slot number “slot” (Line 11). get_hash(src,... Returns the slot number which matches with the dst,fid) values returned from function hashkey (. . .) (Line 12). hashkey(src,...) Returns an identiﬁer for a hash entry corresponding to dst,fid) the input hash key (src,dst,fid). This function is pure virtual and should be overridden by child classes of HashClassifier. Program 6.7 Functions lookup and get hash of class HashClassifier.
1 2 3 4 5 //~ns/classifier/classifier-hash.cc long HashClassifier::lookup(Packet* p) { hdr_ip* h = hdr_ip::access(p); return get_hash(mshift(h->saddr()),mshift(h->daddr()), h->flowid()); }

long HashClassifier::get_hash(nsaddr_t src, nsaddr_t dst, int fid) { 6 Tcl_HashEntry *ep= Tcl_FindHashEntry(&ht_, hashkey(src, dst, fid)); 7 if (ep) 8 return (long)Tcl_GetHashValue(ep); 9 return -1; 10 }

Program 6.7 shows the details of functions lookup(p) and get_hash(src, dst,fid) of class HashClassifier. Function lookup(p) returns the slot number of an entry whose source address, destination address, and ﬂow ID match with those indicated in the header of an incoming packet *p2 (by invoking function get_hash(...) in Line 3). To retrieve an entry, the function get_hash(...) invokes function Tcl_FindHashEntry (. . .) to get the input
2

See the details of IP packet header in Section 8.3.3.

114

6 Nodes as Routers or Computer Hosts

entry from the hash table ht_ in Line 6. If the entry exists, Line 8 will retrieve the slot number by invoking function Tcl_GetHashValue(ep). Declared as pure virtual in class HashClassifier, function hashkey(...) (invoked in Line 6), which computes a hash index from a hash key, should be overridden by the child classes of class HashClassifier. Program 6.8 Declaration of class DestHashClassifier.
1 2 3 4 5 6 7 8 9 10 11 12 //~ns/classifier/classifier-hash.h class DestHashClassifier : public HashClassifier { public: DestHashClassifier() : HashClassifier(TCL_ONE_WORD_KEYS) {} virtual int command(int argc, const char*const* argv); int classify(Packet *p); virtual void do_install(char *dst, NsObject *target); protected: const char* hashkey(nsaddr_t, nsaddr_t dst, int) { long key = mshift(dst); return (const char*) key; } };

As an example, consider class DestHashClassifier (Program 6.8), a child class of class HashClassifier, which classiﬁes incoming packets by the destination address only. Class DestHashClassifier overrides functions classify(p), do_install(dst,target), and hashkey(...), and uses other functions (e.g., lookup(p)) of class HashClassifier (i.e., its parent class). Program 6.9 shows the implementation of function classify(p) of class DestHashClassifier. This function obtains a matching slot number “slot” by invoking lookup(p) (Line 2; See also Fig. 6.3), and returns “slot” if it is valid (Line 4). Otherwise, Line 6 will return variable “default_” if “slot” is invalid. If neither slot nor default_ is valid, Line 7 will return –1, indicating no matching entry in the hash table. Function do_install(dst,target) installs (Line 12) an NsObject pointer target in the next available slot, and registers this installation in the hash table (Line 13). Deﬁned in class Classifier, function getnxt(target) in Line 11 returns the slot where target is installed or the next available slot if target is not found. Again, the statement set_hash(0,d,0,slot) inserts an entry with source address “0”, destination address “d”, and ﬂow ID “0” into the hash table, and associates the entry with a slot number “slot”. Figure 6.3 shows a process when a DestHashClassifier object invokes function lookup(p). In this ﬁgure, the function name is indicated at the top of each box, while the corresponding class is shown in the right of a block arrow. The process follows what we discussed earlier. The important point here is the function hashkey(...). From Lines 8–11 in Program 6.8,

6.4 Classiﬁers: Multi-target Packet Forwarders

115

Program 6.9 Functions DestHashClassifier.
1 2 3 4 5 6 7 8

classify

and

do install

of

class

//~ns/classifier/classifier-hash.cc int DestHashClassifier::classify(Packet * p) { int slot = lookup(p); if (slot >= 0 && slot = 0) return (default_); else return (-1); }

9 void DestHashClassifier::do_install(char* dst, NsObject *target) { 10 nsaddr_t d = atoi(dst); 11 int slot = getnxt(target); 12 install(slot, target); 13 if (set_hash(0, d, 0, slot) < 0) 14 /* show error */ 15 }

class DestHashClassifier overrides function hashkey(...) by returning the destination address (see the detail of function mshift(val) in Line 10 of Program 6.3). In Fig. 6.3, functions lookup(p) and get_hash(...) belong to class HashClassifier, while function hashkey(...) is attributed to class lookup(p) p

HashClassifier

Retrieving source address (src), destination address (dst), and flow ID (fid) from the header of packet p src dst fid

get_hash(src, dst, fid) dst HashClassifier

hashkey(src, dst, fid) (const char*) mshift(dst) ep=Tcl_FindHashEntry(&ht_, ep DestHashClassifier

)

(long)Tcl_GetHashValue(ep)

Return slot number

Fig. 6.3. Flowchart of function lookup(p) invoked from class DestHashClassifier.

116

6 Nodes as Routers or Computer Hosts

DestHashClassifier. This is a beauty of OOP, since we only need to override one function for a derived class (e.g., class DestHashClassifier), and are able reuse the rest of the code from the parent class (e.g., class HashClassifier). Apart from class DestHashClassifier, class HashClassifier has three other major child classes (class names on the left and right are compiled and interpreted classes, respectively): • SrcDestHashClassifier ⇔ Classifier/Hash/SrcDest: classiﬁes packets based on source and destination addresses. • FidHashClassifier ⇔ Classifier/Hash/Fid: classiﬁes packets based on a ﬂow ID. • SrcDestFidHashClassifier ⇔ Classifier/Hash/SrcDestFid: classiﬁes packets based on source address, destination address, and ﬂow ID. 6.4.3 Port Classiﬁers A port classiﬁer classiﬁes packets based on the destination port. From Line 5 in Program 6.10, function classify(p) returns the destination port number of the IP header of the incoming packet p. Program 6.10 Function classify of class PortClassifier.
//~ns/classifier/classifier-port.cc 1 int PortClassifier::classify(Packet *p) 2 { 3 hdr_ip* iph = hdr_ip::access(p); 4 return iph->dport(); 5 }

A port classiﬁer is used as a demultiplexer (e.g., dmux_ in Fig. 6.1) which bridges a node to a receiving transport agent. When function recv(p,h) of dmux_ (i.e., a PortClassifier object) is invoked, the packet is forwarded to an NsObject associated with slot_[cl], where cl is the destination port number speciﬁed in the packet header. By installing a receiving agent in slot_[cl], the classiﬁer forwards packets whose destination port is “cl” to the receiving agent. 6.4.4 Installing Classiﬁers in a Node This section discusses the how classiﬁers are installed in a Node. As shown in Fig. 6.1, a Node can have more than one classiﬁer. These classiﬁers are inter-connected and form a so-called chain of classiﬁers. Class Node has three instvars related to classiﬁer installation: classifier_, hook_assoc_, and mod_assoc_. Instvar classifier_ is the head of the chain

6.4 Classiﬁers: Multi-target Packet Forwarders Table 6.2. An example of hook assoc for a chain of classiﬁers index _o2 _o3 _o4 hook_assoc_(index) _o1 _o2 _o3

117

of classiﬁers, which connects from the node entry. Instvar hook_assoc_ is an associative array whose index is a classiﬁer and its value is the downstream classiﬁer in the chain. For example, let us install classiﬁers _o1, _o2, _o3, and _o4 in sequence into a Node. Then, the instvar classifier_ would be _o4. The value of hook_assoc_ in this case is shown in Table 6.2. Finally, instvar mod_assoc_ is an associative array whose index is a classiﬁer and its value is the associated routing module. As discussed in Section 6.1, class Node provides three instprocs to conﬁgure classiﬁers. First, as shown in Program 6.11, instproc insert-entry{module clsfr hook} takes three input arguments: a routing module module, a classiﬁer clsfr, and an optional argument hook. Line 4 updates the instvar hook_assoc_. Line 8 installs the current head classiﬁer in the slot number “hook” of the input classiﬁer clsfr. Line 11 associates clsfr with the input routing module module. Line 12 replaces the head classiﬁer classifier_ with the input classiﬁer clsfr. Note that clsfr does not need to be a classiﬁer. If clsfr is an NsObject, it can be inserted into the head of the chain. In this case, hook must be speciﬁed as “target” so that Line 6 will set the target of clsfr to be the head classiﬁer. Program 6.11 Instproc insert-entry of class Node
//~ns/tcl/lib/ns-node.tcl 1 Node instproc insert-entry { module clsfr {hook ""} } { 2 $self instvar classifier_ mod_assoc_ hook_assoc_ 3 if { $hook != "" } { 4 set hook_assoc_($clsfr) $classifier_ 5 if { $hook == "target" } { 6 $clsfr target $classifier_ 7 } elseif { $hook != "" } { 8 $clsfr install $hook $classifier_ 9 } 10 } 11 set mod_assoc_($clsfr) $module 12 set classifier_ $clsfr 13 }

The second classiﬁer conﬁguration instproc install-entry{module clsfr hook} is shown in Program 6.12. It is very similar to instproc insert-entry. The only diﬀerence is, it also destroys the existing head classiﬁer, if any.

118

6 Nodes as Routers or Computer Hosts

Program 6.12 Instproc install-entry of class Node.
//~ns/tcl/lib/ns-node.tcl 1 Node instproc install-entry { module clsfr {hook ""} } { 2 $self instvar classifier_ mod_assoc_ hook_assoc_ 3 if [info exists classifier_] { 4 if [info exists mod_assoc_($classifier_)] { 5 $self unregister-module $mod_assoc_($classifier_) 6 unset mod_assoc_($classifier_) 7 } 8 if [info exists hook_assoc_($classifier_)] { 9 if { $hook == "target" } { 10 $clsfr target $hook_assoc($classifier_) 11 } elseif { $hook != "" } { 12 $clsfr install $hook $hook_assoc_($classifier_) 13 } 14 set hook_assoc_($clsfr) $hook_assoc_($classifier_) 15 unset hook_assoc_($classifier_) 16 } 17 } 18 set mod_assoc_($clsfr) $module 19 set classifier_ $clsfr 20 }

Finally, Program 6.13 shows the details of instproc install-demux{demux port}. This instproc takes two input arguments: demux (mandatory) and port (optional). It replaces the existing demultiplexer3 dmux_ with the input demultiplexer demux (Line 2, 9 and 10). If port exists, the current demultiplexer dmux_ will be installed in the slot number “port” of the input demultiplexer demux (Lines 5–7).

6.5 Routing Modules
6.5.1 An Overview of Routing Modules The main functionality of a routing module is to facilitate classiﬁer management. Since a Node maintains only the head of the chain of classiﬁers, access to a classiﬁer in a long chain could be diﬃcult. In addition, it is fairly inconvenient to (possibly selectively) propagate a conﬁguration command to several classiﬁers. Such the diﬃculty is shown in Figure 6.4, where 10 address classiﬁers are connected from the head classiﬁer. As the network topology changes, all the address classiﬁers need to be reconﬁgured. NS2 employs routing modules to facilitate the classiﬁer conﬁguration process.
3

A demultiplexer classiﬁes packets based on port number speciﬁed in the packet header (see Section 6.4.3 for more details).

6.5 Routing Modules

119

Program 6.13 Instproc Node::install-demux.
//~ns/tcl/lib/ns-node.tcl 1 Node instproc install-demux {demux {port ""} } { 2 $self instvar dmux_ address_ 3 if { $dmux_ != "" } { 4 $self delete-route $dmux_ 5 if { $port != "" } { 6 $demux install $port $dmux_ 7 } 8 } 9 set dmux_ $demux 10 $self add-route $address_ $dmux_ 11 }

Routing Modules

RM1 next_rtm_ RM2 next_rtm_ RM11 next_rtm_ ... classifier_ classifier_

0 classifier_ PortClassifier

Agent slot 10

slot 1 Flow classifer (classifier_)

Fig. 6.4. The relationship among routing modules and classiﬁers in a Node.

...

Address classifier 10

Agent

Address classifier 1

Link

Link

120

6 Nodes as Routers or Computer Hosts

Routing modules provide a single point of management for a group of classiﬁers. Here, each routing module is associated with a classiﬁer, and has a pointer to another routing module (see Fig. 6.4). Together, they form a linked list of routing modules for a group of classiﬁers. The head of the linked list acts as an interface to propagate conﬁguration commands to classiﬁers in the linked list. For example, to add a route, we only need to keep the reference of the head routing module (as opposed to keeping the references of 10 address classiﬁers). Then, the new routing information is entered through this head routing module which will propagate the information to all routing modules in the linked list. Each routing module determines whether the information is relevant to the associated classiﬁer. If so, it will (re)conﬁgure the classiﬁer according to the received information. From this point of view, the routing agents and the route logic interact only to the head routing module to deliver classiﬁer conﬁguration commands (e.g., adding or deleting routes) to the relevant classiﬁers. Note that a classiﬁer can also be conﬁgured directly, if the reference is available. Routing modules only facilitate the conﬁguration process of a group of classiﬁers. Routing modules are implemented in a C++ class RoutingModule, which are bound to an OTcl class RtModule (see Program 6.14). Again, these two classes are the base classes from which more speciﬁc classes derive (see the built-in routing module classes in Table 6.3). In the following, we will discuss the base class routing module (classes RoutingModule and RtModule) and the base routing modules (classes BaseRoutingModule and RtModule/Base) only.
Table 6.3. Built-in routing modules in NS2. Routing module Routing Module Base Routing Module Multicast Routing Module Hierarchical Routing Module Manual Routing Module Source Routing Module Quick Start for TCP/IP Routing Module (Determine initial congestion window) Virtual Classiﬁer Routing Module Pragmatic General Multicast Routing Module (Reliable multicast) Light-Weight Multicast Services Routing Module (Reliable multicast) C++ class RoutingModule BaseRoutingModule McastRoutingModule HierRoutingModule ManualRoutingModule SourceRoutingModule QSRoutingModule OTcl class RtModule RtModule/Base RtModule/Mcast RtModule/Hier RtModule/Manual RtModule/Source RtModule/QS

VCRoutingModule PgmRoutingModule LmsRoutingModule

RtModule/VC RtModule/PGM RtModule/LMS

Hereafter, we deﬁne a term name of a routing module as the suﬃx (which follows RtModule/) of the OTcl class name (see Table 6.3). For example,

6.5 Routing Modules

121

Program 6.14 Declaration and the constructor of a C++ class RoutingModule which is bound to an OTcl class RtModule.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 //~ns/routing/rtmodule.h class RoutingModule : public TclObject { public: RoutingModule(); inline Node* node() { return n_; } virtual int attach(Node *n) { n_ = n; return TCL_OK; } virtual int command(int argc, const char*const* argv); virtual const char* module_name() const { return NULL; } void route_notify(RoutingModule *rtm); void unreg_route_notify(RoutingModule *rtm); virtual void add_route(char *dst, NsObject *target); virtual void delete_route(char *dst, NsObject *nullagent); RoutingModule *next_rtm_; protected: Node *n_; Classifier *classifier_; };

17 static class RoutingModuleClass : public TclClass { 18 public: 19 RoutingModuleClass() : TclClass("RtModule") {} 20 TclObject* create(int, const char*const*) { 21 return (new RoutingModule); 22 } 23 } class_routing_module; 24 RoutingModule::RoutingModule() : 25 next_rtm_(NULL), n_(NULL), classifier_(NULL) { 26 bind("classifier_", (TclObject**)&classifier_); 27 }

the name of classes RtModule/Base and RtModule/Hier are Base and Hier, respectively. 6.5.2 C++ Class RoutingModule Program 6.14 shows the declaration of class RoutingModule, which has three main variables. Variable classifier_ in Line 15 is a pointer to a Classifier object. To provide a single pointer of management for a group of classiﬁers, routing modules form a linked list using their pointers next_rtm_ (Line 12) to another RoutingModule object. Another important variable is n_ (Line 14), which is a pointer to the associated Node object. These three variables are initialized to NULL in the constructor of class RoutingModule (Line 25).

122

6 Nodes as Routers or Computer Hosts

Also, variable classifier_ is bound to an OTcl instvar with the same name (Line 26). The key functions of class RoutingModule include (see Program 6.15). node() Returns the attached Node object n_. attach(n) Stores an input Node object “n” in the variable n_. module_name() Returns the name of the routing module. route_notify(rtm) Adds an input RoutingModule *rtm to the end of the linked list. unreg_route_notify(rtm) Removes an input RoutingModule pointer *rtm from the linked list. add_route(dst,target) Informs every classiﬁer in the link list to add a routing entry (dst,target). delete_route(... Informs every classiﬁer in the linked list to dst,nullagent) delete a routing entry with destination dst.

Class RoutingModule is usually not instantiated from the OTcl domain. Therefore, its name is deﬁned as NULL in function module_name() (Line 7 in Program 6.14). Its derived classes override this function by returning their own name to the caller (for class BaseRoutingModule see Line 4 in Program 6.6). Program 6.15 shows the details of functions route_notify(rtm) and unreg_route_notify(rtm). Function route_notify(rtm) recursively invokes itself until it reaches the last routing module in the linked list, where next_rtm_ is NULL. Then, it attaches the input routing module *rtm as the last component of the linked list (Line 5). Function unreg_route_notify recursively searches down the linked list (Line 13) until it ﬁnds the input routing module pointer rtm (Line 9), and removes it from the linked list (Line 10). Lines 17–30 in Program 6.15 show the details of functions add_route(dst, target) and delete_route(dst,nullagent). Function add_route(dst, target) takes a destination node dst and a forwarding NsObject pointer target as input arguments. It installs the pointer target in all the associated classiﬁers (Line 20). Again, this entry is propagated down the linked list (Line 22), until reaching the last element of the linked list (Line 14). Function delete_route(dst,nullagent) does the opposite of the function add_route(dst,target) does. It recursively installs a null agent “nullagent” (i.e., a packet dropping point) as a target for packets destined for a destination node dst in all the classiﬁers, essentially removing the entry with the destination dst from all the classiﬁers. Class RoutingModule also deﬁnes three OTcl commands – namely node, attach-node, and module-name – which simply invoke the functions node(), attach(n), and module_name(), respectively.

6.5 Routing Modules

123

Program 6.15 Functions route notify, unreg route notify, add route, and delete route of class RoutingModule.
1 2 3 4 5 6 //~ns/routing/rtmodule.cc void RoutingModule::route_notify(RoutingModule *rtm) { if (next_rtm_ != NULL) next_rtm_->route_notify(rtm); else next_rtm_ = rtm; }

7 void RoutingModule::unreg_route_notify(RoutingModule *rtm) { 8 if (next_rtm_) { 9 if (next_rtm_ == rtm) { 10 next_rtm_ = next_rtm_->next_rtm_; 11 } 12 else { 13 next_rtm_->unreg_route_notify(rtm); 14 } 15 } 16 } 17 void RoutingModule::add_route(char *dst, NsObject *target) 18 { 19 if (classifier_) 20 classifier_->do_install(dst,target); 21 if (next_rtm_ != NULL) 22 next_rtm_->add_route(dst, target); 23 } 24 void RoutingModule::delete_route(char *dst, NsObject *nullagent) 25 { 26 if (classifier_) 27 classifier_->do_install(dst, nullagent); 28 if (next_rtm_) 29 next_rtm_->add_route(dst, nullagent); 30 }

124

6 Nodes as Routers or Computer Hosts

6.5.3 OTcl Class RtModule In the OTcl domain, the routing module is deﬁned in class RtModule. Class RtModule has two instvars: classifier_ and next_rtm_. Bound to the compiled variable with the same name, instvar classifier_ stores a reference to the associated classiﬁer. Instvar next_rtm_ provides a support to create a linked list of routing module. This instvar has no relationship with variable next_rtm_ of the compiled class, since the bond is not created in the constructor of the C++ class RoutingModule (see Lines 24–27 of Program 6.14). The OTcl class RtModule also deﬁnes the following instprocs which can be classiﬁed into two categories. For brevity, we do not show the details of these instprocs here. The readers may ﬁnd the details of these instprocs in ﬁle ˜ns/tcl/lib/ns-rtmodule.tcl. Initialization Instprocs register{node} Associates the input Node node with the routing module, and updates instvars rtnotif_ and ptnotif_ of the input Node node. unregister{} Removes the classiﬁer of the routing module. Also removes the routing module from instvars rtnotif_ and ptnotif_ of the associated Node. route-notify{... Moves down the linked list in the OTcl module} domain (via instvar next rtm ) and stores the input routing module “module” as the last element of the link-list. unreg-route-notify{... Looks for the input routing module “module” module} and removes it from the linked list of routing modules in the OTcl domain. Instprocs for Route Addition/Deletion and Agent Attachment/Detachment add-route{dst target} Adds a routing entry with a destination “dst” and a forwarding NsObject “target” in all the classiﬁers in the linked list of routing modules. delete-route{... Removes the routing entry with destination dst dst nullagent} from all the classiﬁers in the linked list of routing module. Replaces the target of the classiﬁers with the null agent “nullagent”. attach{agent port} Attaches the input agent “agent” to the associated Node. Set the target of the input (sending) agent “agent” to be the entry of the Node. Also, installs the input (receiving) agent “agent” in the slot number “port” of the demultiplexer “dmux_” of the Node.

6.6 Node Object Conﬁguration

125

6.5.4 C++ Class BaseRoutingModule and OTcl class RtModule/Base Derived from the C++ class RoutingModule, class BaseRoutingModule is declared in Program 6.16, and is bound to an OTcl class RtModule/Base. It overrides function module_name(), by setting its name to be “Base” (Line 4). A base routing module classiﬁes packets based on its destination address only. Therefore, the type of the variable classifier_ is deﬁned as a DestHashClassifier pointer. Program 6.16 Declaration of class BaseRoutingModule which is bound to the OTcl class RtModule/Base.
1 2 3 4 5 6 7 8 //~ns/routing/rtmodule.h class BaseRoutingModule : public RoutingModule { public: BaseRoutingModule() : RoutingModule() {} virtual const char* module_name() const { return "Base"; } virtual int command(int argc, const char*const* argv); protected: DestHashClassifier *classifier_; };

//~ns/routing/rtmodule.cc 9 static class BaseRoutingModuleClass : public TclClass { 10 public: 11 BaseRoutingModuleClass() : TclClass("RtModule/Base") {} 12 TclObject* create(int, const char*const*) { 13 return (new BaseRoutingModule); 14 } 15 } class_base_routing_module;

In the OTcl domain, class RtModule/Base also overrides instproc register { node} of class RtModule. We will discuss the details of this instproc later in Section 6.6.4.

6.6 Node Object Conﬁguration
Having discussed the key Node components, we now show how these components are assembled to compose a Node. In Section 6.6.1 we ﬁrst show the relationship among few closely related Node components. We show the instprocs to add/delete routes in Section 6.6.2, and the instprocs to attach/detach agents in Section 6.6.3. We show the Node construction process (via procedure new{...}) in Section 6.6.4. As we will see, the main Node component (e.g., routing module, classiﬁers, demultiplexer) are assembled during this process.

126

6 Nodes as Routers or Computer Hosts

Finally, the route conﬁguration process (i.e., conﬁguring classiﬁers) is shown in Section 6.6.5. 6.6.1 Relationship Among Instvars module list , reg module , rtnotif , and ptnotif As shown in Fig. 6.5, the following ﬁve instvars of an OTcl Node are closely related: module_list_, reg_module_, rtnotif_, ptnotif_, and mod_assoc_. Instvar module_list_ is a list of strings, each of which represents the name of enabled routing module. Instvar reg_module_ is an associative array whose index and value are the name of the routing module and the routing module instance. Instvars rtnotif_ and ptnotif_ are the objects which should be notiﬁed of a route change and an agent attachment/detachment, respectively. While rtnotif_ is the head of the linked list of the routing modules, ptnotif_ is simply an OTcl list whose elements contain the routing modules. Finally, instvar mod_assoc_ is an associative array whose indexes and values are classiﬁers and the associated routing modules, respectively. The relationship among module_list_, reg_module_, rtnotif_, and ptnotif_ is shown in Fig. 6.5. The instvars are shown in boxes, while the instprocs of class Node are encircled with ellipses. The arrow from an instproc to an instvar indicates that the instvar is conﬁgured from within the instproc. Here, instprocs enable-module{mod_name} and disable-module{mod_name}

route-notify

add-route delete-route rtnotif_ next_rtm_ next_rtm_

module_list_ Hier Mcast new

reg_module_ Hier Mcast RtModule/Hier set RtModule/Mcast next_rtm_ next_rtm_

ptnotif_ enable-module disable-module attach detach

mk-default-classifier

port-notify

Fig. 6.5. Relationship among instvars module list, reg module , rtnotif , and ptnotif of class Node.

6.6 Node Object Conﬁguration

127

place and remove the name of a routing module mod_name in and from instvar module_list_, respectively. When instproc mk-default-classifier is invoked, the names in module_list_ are used to instantiate routing module instances. The instantiated objects are stored in the associative array reg_module_ whose indexes are the corresponding names. Instproc mk-default-classifier also invokes the instprocs route-notify{module} and port-notify{module} to add all the instantiated objects into the list of routing modules rtnotif_ and ptnotif_, respectively. Note that instvar ptnotif_ is an OTcl list, and its pointer next_rtm_ is not used. In Fig. 6.5, instprocs port-notify{...}, attach{agent port}, and detach{agent nullag ent} (see ﬁle ˜ns/tcl/lib/ns-node.tcl) can directly access any component of ptnotif_. However, instprocs route-notify{...}, add-route{dst target}, and delete-route{dst nullagent} must access a routing module through the head of the linked list (i.e., rtnotif_) only. 6.6.2 Adding/Deleting a Routing Entry A routing entry consists of a destination node address dst and a forwarding NsObject target. It can be added to a Node object by using instproc add-route{dst target} of class Node. In Program 6.17, instproc add-route{dst target} of class Node invokes the same instproc of the routing module rtnotif_ which is of class RtModule (Line 4). Line 10 installs the routing entry in the classifier_ of the routing module. Lines 11–13 recursively invoke instproc add-route{dst target} of all the routing modules in the linked list to install the routing entry in the classifier_ associated with each routing module. The mechanism for deleting a route entry is similar to that for adding a route entry, and is omitted for brevity. The readers may ﬁnd the details of route entry deletion in instproc delete-route{dst nullagent} of classes Node and RtModule (see ﬁle ˜ns/tcl/lib/ns-node.tcl and ﬁle ˜ns/tcl/lib/nsrtmodule.tcl). 6.6.3 Agent Attachment/Detachment To attach an agent to a Node, we use instproc attach-agent{node agent} of class Simulator whose syntax is $ns attach-agent $node $agent Here, $ns, $node, and $agent are Simulator, Node, and Agent objects, respectively. Program 6.18 shows the instprocs related to an agent attachment process. The process proceeds as follows: • Simulator::attach-agent{node agent}: Invoke “$node attach $agent” (Line 2).

128

6 Nodes as Routers or Computer Hosts

Program 6.17 Instprocs add-route of classes Node and RtModule.
1 2 3 4 5 6 7 //~ns/tcl/lib/ns-node.tcl Node instproc add-route { dst target } { $self instvar rtnotif_ if {$rtnotif_ != ""} { $rtnotif_ add-route $dst $target } $self incr-rtgtable-size }

//~ns/tcl/lib/ns-rtmodule.tcl 8 RtModule instproc add-route { dst target } { 9 $self instvar next_rtm_ 10 [$self set classifier_] install $dst $target 11 if {$next_rtm_ != ""} { 12 $next_rtm_ add-route $dst $target 13 } 14 }

• Node::attach{agent port}: Update instvar “agent” (Lines 6-8 and Line 16), create “dmux_” if necessary (Lines 9-15), and invoke “$self addtarget $agent $port” (Line 17). • Node::add-target{agent port}: For each routing module “m” stored in the instvar ptnotif_, execute “$m attach $agent $port” (Lines 21-23). • RtModule::attach{agent port}: As a sending agent, set the node entry to be the target of “agent” (Line 26). As a receiving agent, install “agent” in the slot number “port” of demultiplexer “dmux_” (Line 27). Note that although an agent can be either a sending agent or a receiving agent, this instproc assigns both roles to an agent. This does not cause any problem at runtime due to the following reasons. A sending agent is attached to a source node, and always transmits packets destined to a destination node. It takes no action when receiving a packet from a demultiplexer. A receiving agent, on the other hand, does not generate a packet. Therefore, it can never send a packet to the node entry.

6.6.4 Node Construction As has already been mentioned before, a Node object is created in the OTcl domain by executing “$ns node”, where $ns is the Simulator instance. Instproc “node” of class Simulator (see Line 4 in Program 6.19) employs instproc “new{...}” to create a Node object (Line 4 where node_factory_ is set to Node in Line 1). It also updates instvars of the Simulator so that they can be later used by other simulation objects throughout the simulation. The main steps in the node construction process are shown in Table 6.4.

6.6 Node Object Conﬁguration

129

Program 6.18 Instprocs attach and add-target of classes Node, and instproc attach of class RtModule.
1 2 3 //~ns/tcl/lib/ns-lib.tcl Simulator instproc attach-agent { node agent } { $node attach $agent }

//~ns/tcl/lib/ns-node.tcl 4 Node instproc attach { agent { port "" } } { 5 $self instvar agents_ address_ dmux_ 6 lappend agents_ $agent 7 $agent set node_ $self 8 $agent set agent_addr_ [AddrParams addr2id $address_] 9 if { $dmux_ == "" } { 10 set dmux_ [new Classifier/Port] 11 $self add-route $address_ $dmux_ 12 } 13 if { $port == "" } { 14 set port [$dmux_ alloc-port [[Simulator instance] nullagent]] 15 } 16 $agent set agent_port_ $port 17 $self add-target $agent $port 18 } 19 Node instproc add-target { agent port } { 20 $self instvar ptnotif_ 21 foreach m [$self set ptnotif_] { 22 $m attach $agent $port 23 } 24 } //~ns/tcl/lib/ns-rtmodule.tcl 25 RtModule instproc attach { agent port } { 26 $agent target [[$self node] entry] 27 [[$self node] demux] install $port $agent 28 }

130

6 Nodes as Routers or Computer Hosts

Program 6.19 Default value of instvar node factory and instproc node of class Simulator.
1 //~ns/tcl/lib/ns-node.tcl Simulator set node_factory_ Node

//~ns/tcl/lib/ns-node.tcl 2 Simulator instproc node args { 3 $self instvar Node_ routingAgent_ 4 set node [eval new [Simulator set node_factory_] $args] 5 set Node_([$node id]) $node 6 $self add-node $node [$node id] 7 $node nodeid [$node id] 8 $node set ns_ $self 9 return $node 10 }

Table 6.4. Main steps in the Node construction process. Step Class 1 Node 2 Node 3 Instproc Key statement(s) init $self mk-default-classifier mk-default-classifier $self register-module [... new RtModule/Base] Node register-module{mod} $mod register $self set reg_module([$mod ... module-name]) $mod RtModule/Base register{node} $self next $node $self set classifier_ [... new Classifier/Hash/Dest] 32 $node install-entry $classifier_ RtModule register{node} $self attach-node $node $node route-notify $self $node port-notify $self

4

5

6.6 Node Object Conﬁguration

131

Step 1: Constructor of the OTcl class Node Instproc init{...} sets up instvars of class Node, and invokes instproc mk-default-classifier{} of the created Node object (Line 22 in Program 6.20). Program 6.20 Constructor of class Node.
1 //~/ns/tcl/lib/ns-node.tcl Node set module_list_ { Base }

2 Node instproc init args { 3 eval $self next $args 4 $self instvar id_ agents_ dmux_ neighbor_ rtsize_ address_ \ 5 nodetype_ multiPath_ ns_ rtnotif_ ptnotif_ 6 set ns_ [Simulator instance] 7 set id_ [Node getid] 8 $self nodeid $id_ ;# Propagate id_ into c++ space 9 if {[llength $args] != 0} { 10 set address_ [lindex $args 0] 11 } else { 12 set address_ $id_ 13 } 14 $self cmd addr $address_; # Propagate address_ into C++ space 15 set neighbor_ "" 16 set agents_ "" 17 set dmux_ "" 18 set rtsize_ 0 19 set ptnotif_ {} 20 set rtnotif_ {} 21 set nodetype_ [$ns_ get-nodetype] 22 $self mk-default-classifier 23 set multiPath_ [$class set multiPath_] 24 } 25 Node instproc mk-default-classifier {} { 26 Node instvar module_list_ 27 foreach modname [Node set module_list_] { 28 $self register-module [new RtModule/$modname] 29 } 30 } 31 Node instproc register-module { mod } { 32 $self instvar reg_module_ 33 $mod register $self 34 set reg_module_([$mod module-name]) $mod 35 }

132

6 Nodes as Routers or Computer Hosts

Step 2: Instproc mk-default-classifier{} Instproc mk-default-classifier{} creates (using new{...}) and registers (using register-module{mod}) routing modules whose names are stored in the instvar module_list_ (Lines 27–29 in Program 6.20). By default, only “Base” routing module is stored in instvar module_list_ (Line 1 in Program 6.20). To enable/disable other routing module, the following two instprocs of class RtModule must be invoked prior to the execution of “$ns node”: enable-module{name} disable-module{name} where is the name of the routing module, which is to be enabled/ disabled. Step 3: Instproc register-module{mod} of class Node This instproc invokes instproc register{node} of the input routing module mod and stores the registered module in the instvar reg_module_. Step 4: Instproc register{node} of class RtModule/Base This instproc ﬁrst invokes instproc register{node} of its parent class (by the statement $self next $node in Line 7 of Program 6.21). Then, Lines 9–12 create (using new{...}) and conﬁgure (using install-entry{...}) the head classiﬁer (i.e., classifier_) of the Node. Step 5: Instproc register{node} of class RtModule This instproc attaches input Node object “node”to the routing module. It also invokes instproc route-notify{module} and port-notify{module} of the associated Node to include the routing module into the route notiﬁcation list rtnotif_ and port notiﬁcation list ptnotif_ of the associated Node (see Program 6.22). The details of instprocs route-notify{module} and port-notify {module} are shown in Program 6.22. The instproc route-notify{module} takes one input routing module. It stores the module in the last instvar next_rtm_ down the linked list of routing modules (see Lines 6 and 10-17). It also invokes the OTcl command route-notify of the input routing module (Line 8). The OTcl command route-notify invokes the C++ function route_notify(rtm) associated with the attached Node (see Lines 18-24) to store the routing module as the last routing module in the linked list (see Lines 25-30). As shown in Lines 31-34 of Program 6.22, the instproc port-notify{ module} takes a routing module as an input argument, and appends the input argument module to the end of the link-list.

6.6 Node Object Conﬁguration

133

Program 6.21 Instprocs register of classes RtModule and RtModule/Base.
1 2 3 4 5 //~/ns/tcl/lib/ns-rtmodule.tcl RtModule instproc register { node } { $self attach-node $node $node route-notify $self $node port-notify $self }

6 RtModule/Base instproc register { node } { 7 $self next $node 8 $self instvar classifier_ 9 set classifier_ [new Classifier/Hash/Dest 32] 10 $classifier_ set mask_ [AddrParams NodeMask 1] 11 $classifier_ set shift_ [AddrParams NodeShift 1] 12 $node install-entry $self $classifier_ 13 }

6.6.5 Route Conﬁguration At the beginning of the Simulation Phase, NS2 computes the optimal routes for all source-destination nodes, using the Dijkstra shortest path algorithm [17]. It installs the computed routing information in all the Nodes. This phase commences by the execution of instproc run{} of the Simulator. Table 6.5 shows the main steps in the instproc run{} which are related to the route conﬁguration process.
Table 6.5. Main steps in the route conﬁguration process. Step 1 2 3 4 5 Class Simulator RouteLogic Agent/... rtProto/Static Simulator Simulator Instproc run configure init-all Invocation [$self get-routelogic] configure Agent/rtProto/Static init-all [Simulator instance] ... compute-routes compute-routes $self compute-flat-routes compute-flat- set r [$self get-routelogic] routes $r compute set n [Node set nn_] $self ... populate-flat-classifiers $n

134

6 Nodes as Routers or Computer Hosts

Program 6.22 Instprocs and functions which are related to instprocs route-notify and port-notify of the OTcl class Node.
1 2 3 4 5 6 7 8 9 //~/ns/tcl/lib/ns-node.tcl Node instproc route-notify { module } { $self instvar rtnotif_ if {$rtnotif_ == ""} { set rtnotif_ $module } else { $rtnotif_ route-notify $module } $module cmd route-notify $self }

//~/ns/tcl/lib/ns-rtmodule.tcl 10 RtModule instproc route-notify { module } { 11 $self instvar next_rtm_ 12 if {$next_rtm_ == ""} { 13 set next_rtm_ $module 14 } else { 15 $next_rtm_ route-notify $module 16 } 17 } //~ns/routing/rtmodule.cc 18 int BaseRoutingModule::command(int argc, const char*const* argv) { 19 Tcl& tcl = Tcl::instance(); 20 if (argc == 3) { 21 if (strcmp(argv[1] , "route-notify") == 0) { 22 n_->route_notify(this); 23 } 24 } //~ns/common/node.cc 25 void Node::route_notify(RoutingModule *rtm) { 26 if (rtnotif_ == NULL) 27 rtnotif_ = rtm; 28 else 29 rtnotif_->route_notify(rtm); 30 } //~/ns/tcl/lib/ns-node.tcl 31 Node instproc port-notify { module } { 32 $self instvar ptnotif_ 33 lappend ptnotif_ $module 34 }

6.6 Node Object Conﬁguration

135

Step 1: Instproc run{} of class Simulator Shown in Line 2 of Program 4.12, instproc run{} of class Simulator retrieves the RouteLogic object using its instproc get-routelogic{} and invokes instproc configure{} associated with the retrieved RouteLogic object. Step 2: Instproc configure{} of class RouteLogic Deﬁned in ﬁle ˜ns/tcl/lib/ns-route.tcl, instproc configure{} of class Route Logic conﬁgures the routing table for all the Nodes by invoking instproc init-all{} of class Agent/rtProto/Static. Step 3: Instproc init-all{} of class Agent/rtProto/Static Deﬁned in ﬁle ˜ns/tcl/rtglib/route-proto.tcl, instproc init-all{} of class Agent /rtProto/Static invokes the instproc compute-routes{} of the Simulator. Step 4: Instproc compute-routes{} of class Simulator By default, NS2 uses ﬂat addressing. Therefore, instproc compute-routes{} of class Simulator invokes instproc compute-flat-routes{} to compute and setup the routing table (see ﬁle ˜ns/tcl/lib/ns-route.tcl). Step 5: Instproc compute-flat-routes{} of class Simulator Deﬁned in ﬁle ˜ns/tcl/lib/ns-route.tcl, instproc compute-flat-routes{} of class Simulator retrieves the associated route logic object (using instproc get-routelogic{}), computes the optimal route using the retrieved object (using instproc compute{}), and conﬁgures the classiﬁers in all the Nodes according to the computed route (using the command populate-flatclassifiers{n}). Program 6.23 shows the details of OTcl command populate-flat-cla ssifiers{n}. This OTcl command stores the input number of nodes “n” in the variable nn_ (Line 4), and invokes function populate_flat_classifiers() (Line 5) to install the computed route in all the classiﬁers. As shown in Lines 10–25 of Program 6.23, function populate_flat_class ifiers() is run for all pairs (i,j) of nn_ nodes. For each pair, Line 16 retrieves the next hop (i.e., forwarding) referencing point nh of a forwarding object for a packet traveling from Node “i” to Node “j”, and Line 18 retrieves the link entry point l_head corresponding to the variable nh. Lines 19–20 add a new routing entry for the node i (i.e., nodelist_[i]). The entry speciﬁes the link entry l_head as a forwarding target for packet destined for a destination node j. The entry is included to the Node “i” via its function add_route(dst,target).

136

6 Nodes as Routers or Computer Hosts

Program 6.23 An OTcl command populate-flat-classifiers, a function populate flat classifiers of class Simulator, and a function add route of class Node.
1 2 3 4 5 6 7 8 9 //~ns/common/simulator.cc int Simulator::command(int argc, const char*const* argv) { ... if (strcmp(argv[1], "populate-flat-classifiers") == 0) { nn_ = atoi(argv[2]); populate_flat_classifiers(); return TCL_OK; } ... }

10 void Simulator::populate_flat_classifiers() { 11 ... 12 for (int i=0; i= 0) { 18 NsObject *l_head=get_link_head(nodelist_[i],nh); 19 sprintf(tmp, "%d", j); 20 nodelist_[i]->add_route(tmp, l_head); 21 } 22 } 23 } 23 } 25 } //~ns/common/node.cc 26 void Node::add_route(char *dst, NsObject *target) { 27 if (rtnotif_) 28 rtnotif_->add_route(dst, target); 29 }

6.7 Chapter Summary

137

In Lines 26–29 of Program 6.23, function add_route(dst,target) simply invokes function add_route(dst,target) of the associated RoutingModule object rtnotif_. Deﬁned in Program 6.15, function add_route(dst,target) of class RoutingModule recursively installs the input routing entry down the linked list of routing modules, by executing do_install(dst,target) of the variable classifier_ associated with each routing module. The function do_install(...) installs NsObject target in slot dst of the classiﬁer such that packets destined for the destination dst are forwarded to NsObject target.

6.7 Chapter Summary
A Node is a basic component which acts as a router and a computer host. Its main responsibilities are to forward packets according to a routing table and to bridge the high-layer protocols to a low-level network. A Node consists of two key components: classiﬁers and routing modules. A classiﬁer is a multitarget packet forwarder. It is used in a Node to forward packets, which are destined to diﬀerent destinations, to diﬀerent forwarding NsObjects. It is also used as a demultiplexer, which forwards packets with diﬀerent destination ports to diﬀerent attached transport-layer agents. As another main component, a routing module acts as a single point of management for a group of classiﬁers in a Node. When receiving a conﬁguration command, it propagates the command to the related classiﬁers. It acts as an interface to other routing components such as route logic (which is responsible for computing the optimal routes), and to the agent attachment/detachment instprocs of class Node. Routing modules alleviate the need to conﬁgure every classiﬁer separately, and therefore, greatly facilitate the classiﬁer conﬁguration process especially for a highly-complicated node conﬁguration with numerous classiﬁers. During the Network Conﬁguration Phase, a Node is created by executing $ns node where $ns is the object. At the construction, address classiﬁers and routing modules are installed in the Node. However, the routing mechanism of the address classiﬁers are not conﬁgured here. The transport layer connections, on the other hand, are created in this phase using instproc attach-agent of class Simulator. At the beginning of the Simulation Phase, NS2 computes the optimal routes for all pairs of nodes, and installs the computed routing information in relevant classiﬁers.

7 Link and Buﬀer Management

A Link is an OTcl object which connects two nodes and carries packets from the beginning node to the terminating node. This chapter focuses on a class of most widely-used Link object, namely, SimpleLink objects. Conveying packets from one node to another, a SimpleLink object models packet transmission time, link propagation delay, and packet buﬀering. Here, packet transmission time refers to the time required by a transmitter to send out a packet. It is determined by the link bandwidth and packet size. Link propagation delay is the time needed to convey each bit from the beginning to the end of a link. In presence of bursty traﬃc, a transmitter may receive packets while transmitting a packet. The packets entering a busy transmitter could be placed in a buﬀer for future transmission. Unlike the real implementation, NS2 implements packet buﬀering in a Link, not a Node. In the following, we ﬁrst give an introduction to classes Link and SimpleLink in Section 7.1. Then, we show how NS2 models packet transmission time and propagation delay in Section 7.2. Next, the packet buﬀering, queue blocking, and callback mechanisms are discussed in Section 7.3. Section 7.4 shows a network construction and packet ﬂow example. Finally, the chapter summary is provided in Section 7.5.

7.1 Introduction to SimpleLink Objects
NS2 models a link using classes derived from OTcl class Link object, among which OTcl class SimpleLink is the simplest one which can be used to connect two Nodes. 7.1.1 Main Components of a SimpleLink Figure 7.1 shows the composition of class SimpleLink, which consists of the following basic objects and tracing objects in the interpreted hierarchy:

T. Issariyakul, E. Hossain, Introduction to Network Simulator NS2, DOI: 10.1007/978-0-387-71760-9 7, c Springer Science+Business Media, LLC 2009

140

7 Link and Buﬀer Management

Fig. 7.1. Architecture of a SimpleLink object.

Basic Objects head_ The entry point of a SimpleLink object. queue_ As a Queue object, queue_ models packet buﬀering of a “real” router (see Section 7.3). link_ A DelayLink object, which models packet transmission time and link propagation delay (see Section 7.2). ttl_ A time to live checker object whose class is TTLChecker. It decrements the time to live ﬁeld of an incoming packet. After the decrement, if the time to live ﬁeld is still positive, the packet will be forwarded to the next element in the link. Otherwise, it will be removed from the simulation (see ﬁle ˜ns/common/ttl.h,cc). drophead_ The common packet dropping point for the link. The dropped packets are forwarded to this object. It is usually connected to the null agent of the Simulator so that all SimpleLink objects share the same dropping point. Tracing Objects These objects will be inserted only if instvar $traceAllFile_ of the is deﬁned. We will describe the details of tracing objects in detail in Chapter 13. These objects are enqT_ deqT_ drpT_ rcvT_ Trace packets entering queue_. Trace packets leaving queue_. Trace packets dropped from queue_. Trace packets leaving the link or equivalently received by the next node.

7.1.2 Instprocs for Conﬁguring a SimpleLink Object In the OTcl domain, a SimpleLink object is created using the instprocs simplex-link{..} and duplex-link{...} of class Simulator whose syntax is as follows:

7.1 Introduction to SimpleLink Objects

141

$ns simplex-link $n1 $n2 $ns duplex-link $n1 $n2 where $ns is the Simulator object, and $n1 and $n2 are Node objects. Instproc simplex-link{...} above creates a uni-directional SimpleLink object connecting Node $n1 to Node $n2 (Program 7.1). The speed and the propagation delay of the link are given as (in bps) and (in seconds), respectively. Again, as opposed to a “real” router, NS2 incorporates a queue in a SimpleLink object, not in a Node object. The type of the queue in the link is speciﬁed by . Program 7.1 Instproc simplex-link of class Simulator.
//~ns/tcl/lib/ns-lib.tcl 1 Simulator instproc simplex-link { n1 n2 bw delay qtype args } { 2 $self instvar link_ queueMap_ nullAgent_ useasim_ 3 switch -exact $qtype { 4 /* See the detail in ~ns/tcl/lib/ns-lib.tcl */ 5 default { 6 set q [new Queue/$qtype $args] 7 } 8 } 9 switch -exact $qtypeOrig { 10 /* See the detail in ~ns/tcl/lib/ns-lib.tcl */ 11 default { 12 set link_($sid:$did) [new SimpleLink \ $n1 $n2 $bw $delay $q] 13 } 14 } 15 }

Program 7.1 shows details of instproc Simulator::simplex-link{...}. Line 6 creates an object of class Queue/qtype. Line 12 constructs a SimpleLink object, connecting node n1 to n2. It speciﬁes delay, bandwidth, and Queue object of the link to be bw, delay, and q, respectively. The Simulator stores the created SimpleLink object in its instance associative array link_ ($sid:$did), where sid is the source node ID, and $did is the destination node ID (see Chapter 4). Instproc duplex-link{...} creates two SimpleLink objects: one connecting Node $n1 to Node $n2 and another connecting Node $n2 to Node n1. For brevity, we do not show the detail here. The readers are encouraged to ﬁnd the details of instproc duplex-link{...} in ﬁle ˜ns/tcl/lib/ns-lib.tcl.

142

7 Link and Buﬀer Management

7.1.3 The Constructor of Class SimpleLink Program 7.2 shows the details of instproc init{...} (i.e., the constructor) of class SimpleLink, which constructs and connects objects according to Fig. 7.1. Lines 3, 5, 11, 12, and 18 create instvars drophead_, head_, queue_, link_, and ttl_, whose OTcl classes are Connector, Connector, Queue, DelayLink, and TTLChecker, respectively. Note that the bandwidth and delay of instvar link_ are conﬁgured in Lines 13–14. Program 7.2 The constructor of the OTcl class SimpleLink.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 //~ns/tcl/lib/ns-link.tcl SimpleLink instproc init { src dst bw delay q { lltype "DelayLink"} } { set ns [Simulator instance] set drophead_ [new Connector] $drophead_ target [$ns set nullAgent_] set head_ [new Connector] if { [[$q info class] info heritage ErrModule] == "ErrorModule" } { $head_ target [$q classifier] } else { $head_ target $q } set queue_ $q set link_ [new $lltype] $link_ set bandwidth_ $bw $link_ set delay_ $delay $queue_ target $link_ $link_ target [$dst entry] $queue_ drop-target $drophead_ set ttl_ [new TTLChecker] $ttl_ target [$link_ target] $self ttl-drop-trace $link_ target $ttl_ }

Apart from creating the above objects, the constructor also connects the created objects as in Fig. 7.1. Derived from class Connector, each of the created objects uses command target and drop-target to specify the next downstream object and the dropping point, respectively (see Chapter 5). Line 9 sets the target of head_ to be q. Line 15 sets the target of queue_ (which is set to “q” in Lines 11) to be link_. Line 16 sets the target of link_ to be the entry of the next node. Lines 19 and 21 insert ttl_ between link_ and the entry of the next node. Line 17 sets the dropping point of queue_ to be

7.2 Modeling Packet Departure

143

drophead_. Finally, Line 4 sets the target of drophead_ to be the null agent of the Simulator.

7.2 Modeling Packet Departure
7.2.1 Packet Departure Mechanism NS2 models packet departure by using a C++ class Linkdelay (see Program 7.3), which is bound to an OTcl class DelayLink object. Again, the OTcl class DelayLink is used to instantiate the instvar SimpleLink::link_ which models the packet departure process. Program 7.3 Declaration of class LinkDelay.
//~ns/link/delay.h 1 class LinkDelay : public Connector { 2 public: 3 LinkDelay(): dynamic_(0), latest_time_(0), itq_(0){ 4 bind_bw("bandwidth_", &bandwidth_); 5 bind_time("delay_", &delay_); 6 } 7 void recv(Packet* p, Handler*); 8 void send(Packet* p, Handler*); 9 void handle(Event* e); 10 inline double txtime(Packet* p) { /* Packet TXT Time */ 11 return (8. * hdr_cmn::access(p)->size() / bandwidth_); 12 } 13 protected: 14 int command(int argc, const char*const* argv); 15 double bandwidth_; 16 double delay_; 17 PacketQueue* itq_; 18 Event intr_; /* In transit */ 19 }; //~ns/link/delay.cc 20 static class LinkDelayClass : public TclClass { 21 public: 22 LinkDelayClass() : TclClass("DelayLink") {} 23 TclObject* create(int argc , const char*const* 24 return (new LinkDelay); 25 } 26 } class_delay_link;

argv ) {

A packet departure process consists of packet transmission time and link propagation delay. While the former deﬁnes the time a packet stays in an

144

7 Link and Buﬀer Management

upstream node, the summation of the former and the latter determines the time needed to deliver an entire packet to the connecting downstream node. Conceptually, when a LinkDelay object receives a packet, it places these two events on the simulation timeline: (i) Packet departure from an upstream object: Deﬁne packet transmission packet size as time needed to transmit a packet over a link. After time = bandwidth a period of packet transmission time, the packet completely leaves (or departs) the transmitter, and the transmitter is allowed to transmit another packet. Upon a packet reception, a LinkDelay object waits for a period of packet transmission time, and informs its upstream object that it is ready to receive another packet. (ii) Packet arrival at a downstream node: Deﬁne propagation delay as the time needed to deliver a data bit from the beginning to the end of the link. Again, an entire packet needs a period of “packet transmission time + propagation delay” to reach the destination. A LinkDelay object, therefore, schedules a packet reception event at the downstream node after this period. 7.2.2 C++ Class LinkDelay Program 7.3 shows the declaration of C++ class LinkDelay, which is mapped to the OTcl class DelayLink. Class LinkDelay has the following four main variables. Variables bandwidth_ (Line 15) and delay_ (Line 16) store the link bandwidth and propagation delay, respectively. In Lines 4–5, these two variables are bound to OTcl instvars with the same name. In a link with large bandwidth-delay product, a transmitter can send a new packet before the previous packet reaches the destination. Class LinkDelay stores all packets in-transit in its buﬀer itq_ (Line 17), which is a pointer to a PacketQueue object (See Section 7.3.1). Finally, variable intr_ is a dummy Event object, which represent a packet departure (from the transmitting node) event. As discussed in Section 4.3.6, the packet departure is scheduled using variable intr_ which does not take part in event dispatching1 . The main functions of class LinkDelay are recv(p,h), send(p,h), handle (e), and txttime(p). Function txttime(p) calculates the packet transmission time of packet *p (Lines 10–12 in Program 7.3). Function send(p,h) sends packet *p to the connecting downstream object (see Line 12 in Program 5.3). Function handle(e) is invoked when the Scheduler dispatches an event corresponding to the LinkDelay object (see Chapter 4). Function recv(p,h) (Program 7.4) takes a packet *p and a handler *h as input arguments, and schedules packet departure and packet arrival events.
1

As a dummy Event object, variable intr ensures that an error message will be shown on the screen, if an undispatched event is rescheduled.

7.2 Modeling Packet Departure

145

Program 7.4 Function recv of class LinkDelay.
//~ns/link/delay.cc 1 void LinkDelay::recv(Packet* p, Handler* h) 2 { 3 double txt = txtime(p); 4 Scheduler& s = Scheduler::instance(); 5 if (dynamic_) { /* See ~ns/link/delay.cc */ } 6 else if (avoidReordering_) { /* See ~ns/link/delay.cc */ } 7 else { 8 s.schedule(target_, p, txt + delay_); 9 } 10 s.schedule(h, &intr_, txt); 11 }

(i) Packet departure event: Since a packet spends “packet transmission time” (txt in Line 3) at the upstream object, function recv(p,h) schedules a packet departure event at txt seconds after the LinkDelay object receives the packet. To do so, Line 10 invokes function schedule(h,&intr,txt) of class Scheduler, where the ﬁrst, second, and third input arguments are handler pointer, dummy event pointer, and delay, respectively (see Chapter 4). After txt seconds, the Scheduler dispatches this event by invoking function handle(e) associated with the handler *h to inform the upstream object of a packet departure. In most cases, the upstream object responds by transmitting another packet, if available (see Section 7.3.3 for the callback mechanism). (ii) Packet arrival: Class LinkDelay also passes the packet to its downstream object (*target_). Line 8 schedules an event cast from the input packet *p with delay txt+delay_ seconds, where txt is the packet transmission time and delay_ is the link propagation delay. Here, *target_ is passed to function schedule(...) as a handler pointer. After “txt+delay_” seconds, h.handle(p) will invoke function recv(p) (see Program 4.2), and packet *p will be passed to *target_ after txt+delay_ seconds. The major diﬀerence between scheduling packet departure and arrival events is as follows. While a node can hold only one (head of the line) packet, a link can contain more than one packet. Correspondingly, at an instance, a link can schedule only one packet departure event (using intr_), and more than one packet arrival event (using *p which represents a packet). Every time a LinkDelay object receives a packet, it schedules the packet departure event using the same variable intr_. If variable intr_ has not been dispatched, such a scheduling will cause runtime error, because it attempts to place a packet in the head of the buﬀer which is currently occupied by another packet. A packet arrival event, on the other hand, is tied to incoming packet. A LinkDelay object schedules a new packet arrival event for every received packet (see Line 8 in Program 7.4). Therefore, a link can schedule a packet

146

7 Link and Buﬀer Management

arrival event, even if the previous arrival event has not been dispatched. This is essentially the case in a link with large bandwidth-delay product which can contain several packets.

7.3 Buﬀer Management
Another major component of a SimpleLink object is class Queue. Class Queue models the buﬀering mechanism in a network router. It stores a received packet in the buﬀer, and forwards a (in most case the head of the line) packet in the buﬀer to its downstream object when the ongoing transmission is complete. As shown in Program 7.5, class Queue derives from class Connector, and can be used to connect two NsObjects. It employs a PacketQueue object (see Section 7.3.1), *pq_ in Line 20, for packet buﬀering. The buﬀer size is speciﬁed in variable qlim_ (Line 16). The variables blocked_ (Line 16), unblock_on_resume_ (Line 17), and qh_ (Line 18) are related to the so-called callback mechanism, and shall be discussed later in Section 7.3.3. Program 7.5 Declaration of class Queue.
//~ns/queue/queue.h 1 class Queue : public Connector { 2 public: 3 virtual void enque(Packet*) = 0; 4 virtual Packet* deque() = 0; 5 virtual void recv(Packet*, Handler*); 6 void resume(); 7 int blocked() const { return (blocked_ == 1); } 8 void unblock() { blocked_ = 0; } 9 void block() { blocked_ = 1; } 10 int limit() { return qlim_; } 11 int length() { return pq_->length(); } 12 virtual ~Queue(); 13 protected: 14 Queue(); 15 void reset(); 16 int qlim_; 17 int blocked_; 18 int unblock_on_resume_; 19 QueueHandler qh_; 20 PacketQueue *pq_; 21 };

There are a number of important functions of class Queue. Function enque(p) and deque() (Lines 3–4) place and take, respectively, a packet from the PacketQueue object *pq_. They are declared as pure virtual, and must

7.3 Buﬀer Management

147

be implemented by instantiable derived classes of class Queue. Derived from class NsObject, function recv(p,h) (Line 5) is the main packet reception function. Function blocked() in Line 7 indicates whether the Queue object is in a blocked state. Functions resume() (Line 6), unblock() (Line 8), and block() (Line 9) are used in the callback mechanism which will be discussed in Section 7.3.3. Finally, functions limit() and length() return the buﬀer size and current buﬀer occupancy, respectively. 7.3.1 Class PacketQueue: A Model for Packet Buﬀering Declared in Program 7.6, class PacketQueue models low-level operations of the buﬀer including storing, enqueuing, and dequeuing packet. Class PacketQueue is a linked list of Packets, whose member variables are as follows. Variable head_ in Line 11 is the pointer to the beginning of the linked list. Variable tail_ in Line 12 is the pointer to the end of the linked list. The variable len_ in Line 13 is the number of packets in the buﬀer. Function enque(p) in Line 5 puts the input packet *p to the end of the buﬀer. Function deque() in Line 6 returns the head of the line packet pointer or returns NULL when the buﬀer is non-empty or empty, respectively. Function remove(p) in Line 7 searches for a matching packet *p and removes it from the buﬀer (if found). Note that packet admitting/dropping is the functionality of class Queue, not of class PacketQueue. We will show an example of packet admitting/dropping of class DropTail in Section 7.3.4. Program 7.6 Declaration of class PacketQueue.
//~ns/queue/queue.h 1 class PacketQueue : public TclObject { 2 public: 3 PacketQueue() : head_(0), tail_(0), len_(0), bytes_(0) {} 4 virtual int length() const { return (len_); } 5 virtual Packet* enque(Packet* p); 6 virtual Packet* deque(); 7 virtual void remove(Packet*); 8 Packet* head() { return head_; } 9 Packet* tail() { return tail_; } 10 protected: 11 Packet* head_; 12 Packet* tail_; 13 int len_; 14 };

148

7 Link and Buﬀer Management

7.3.2 Queue Handler Derived from class Handler (see Line 1 in Program 7.7), class QueueHandler is closely related to the (event) Scheduler. Again, a QueueHandler object deﬁnes its default actions in its function handle(e). These default actions will be taken when an associated event is dispatched. As shown in Lines 8–11 of Program 7.7, the default action of a QueueHandler object is to execute function resume() of the associated Queue object queue_. We will discuss the details of function resume() in Section 7.3.3. In the rest of this section, we will demonstrate how a connection between QueueHandler and Queue objects is created. Program 7.7 Declaration and function handle of class QueueHandler, and the constructor of class Queue.
1 2 3 4 5 6 7 //~ns/queue/queue.h class QueueHandler : public Handler { public: inline QueueHandler(Queue& q) : queue_(q) {} void handle(Event*); private: Queue& queue_; };

//~ns/queue/queue.cc 8 void QueueHandler::handle(Event*) 9 { 10 queue_.resume(); 11 } 12 Queue::Queue() : Connector(), blocked_(0), unblock_on_resume_(1), qh_(*this),pq_(0) 13 { /* See the detail in ~ns/queue/queue.cc */ }

To associate a Queue object with a QueueHandler object, classes Queue and QueueHandler declare their member variables qh_ (Line 19 in Program 7.5) and queue_ (Line 6 in Program 7.7), as a QueueHandler pointer and a Queue reference, respectively. These two variables are initialized when a Queue object is instantiated (Line 12 in Program 7.7). The constructor of class Queue invokes the constructor of class QueueHandler, feeding itself as an input argument (i.e., qh_(*this) in Line 12 of Program 7.7). The constructor of qh_ then sets its member variable queue_ to share the same address as the input Queue object (i.e. queue_(q) in Line 3 of Program 7.7), hence creating a two-way connection between the Queue and QueueHandler objects. After this point, the Queue and the QueueHandler objects refer to each other by the variables qh_ and queue_, respectively.

7.3 Buﬀer Management

149

Fig. 7.2. State diagram of the queue blocking mechanism.

7.3.3 Queue Blocking and Callback Mechanism Queue Blocking NS2 uses the concept of queue blocking2 to indicate whether a queue is currently transmitting a packet. By default, a queue can transmit one packet at a time. It is not allowed (i.e., blocked) to transmit another packet until the ongoing transmission is complete. A queue is said to be blocked or unblocked (i.e., blocked_ = 1 or blocked_ = 0), when it is transmitting a packet or is not transmitting a packet, respectively. Figure 7.2 shows the state diagram of the queue blocking mechanism. When in the “Not Blocked” state, a queue is allowed to transmit a packet by executing “target_->recv(p,&qh_)”, after which it enters the “Blocked” state. Here, a queue waits until the ongoing transmission is complete where the function resume() is invoked. After this point, the queue enters the “Not Blocked” state and the process repeats.

Callback Mechanism As discussed in Chapter 5, a node in NS2 passes packets to a downstream node by executing function recv(p,h), where *p denotes a packet and *h denotes a handler. A callback mechanism refers to a process where a downstream object invokes an upstream object along the downstream path for a certain purpose. In a queue blocking process, a callback mechanism is used to unblock a Queue object by invoking function resume() of the upstream Queue object. We now explain the callback mechanism process for queue unblocking via an example network in Fig. 7.3. Here, we assume that the following objects are sequentially connected: an upstream NsObject, a Queue object, a LinkDelay object, and a downstream object. Again, an NsObject passes a packet *p by invoking function recv(p,h) of its downstream object, where *h is a handler. In most cases, the input handler *h is passed along with the packet *p as input argument of function recv(p,h). However, this mechanism is diﬀerent for the Queue object.
2

Queue blocking has no relation to packet blocking when the buﬀer is full.

150

7 Link and Buﬀer Management

Fig. 7.3. Diagram of callback mechanism for a queue unblocking process.

Consider function recv(p,h) of class Queue in Program 7.8. Instead of immediately passing the incoming packet *p to its downstream object, Line 3 places the packet in the buﬀer (i.e., pq_). Again, a Queue object is allowed to transmit a packet only when it is not blocked (Line 4). In this case, Line 5 retrieves a packet from the buﬀer. If the packet is valid (Line 6), Line 7 will set the state of the Queue object to be “blocked”, and Line 8 will forward the packet to its downstream object (i.e., *target_). The Queue object passes its QueueHandler pointer qh_ (instead of the incoming handler pointer) to its downstream object. This QueueHandler pointer acts as a reference point for a queue blocking callback mechanism. Program 7.8 Function recv of class Queue.
//~ns/queue/queue.cc 1 void Queue::recv(Packet* p, Handler*) 2 { 3 enque(p); 4 if (!blocked_) { 5 p = deque(); 6 if (p != 0) { 7 blocked_ = 1; 8 target_->recv(p, &qh_); 9 } 10 } 11 }

7.3 Buﬀer Management

151

From Fig. 7.3, the downstream object of the Queue object is a LinkDelay object. Upon receiving a packet, it schedules two events: packet departure and arrival events (see Lines 10 and 8 in Program 7.4). A packet arrival event is associated with the downstream object (i.e., *target_). At the ﬁring time, the function handle(p) of the downstream object will invoke function recv(p) to receive packet *p (see Program 4.2). Function recv(p) of class LinkDelay also schedules a packet departure event. Since the input handler pointer is a QueueHandler pointer, the departure event is associated with the QueueHandler object qh_. At the ﬁring time, the Scheduler invokes function handle(p) of the associated QueueHandler object. In Program 7.7, this function in turn invokes function resume() to unblock the associated Queue object. Literally the LinkDelay object schedules an event which calls back to unblock the upstream Queue object. Program 7.9 Function resume of class Queue.
//~ns/queue/queue.cc 1 void Queue::resume() 2 { 3 Packet* p = deque(); 4 if (p != 0) 5 target_->recv(p, &qh_); 6 else 7 if (unblock_on_resume_) 8 blocked_ = 0; 9 else 10 blocked_ = 1; 11 }

Program 7.9 shows the details of function resume(). Function resume() ﬁrst retrieves the head of the line packet from the buﬀer (Line 3). If the buﬀer is non-empty (Line 4), Line 5 will send the packet to the downstream object of the queue regardless of the blocked status. In this case, the variable blocked_ would remain unchanged. If the Queue object is in a “Blocked” state, it will remained blocked after packet transmission, hence complying with the state diagram in Fig. 7.2. If the queue is idle (i.e., the buﬀer is empty), variable blocked_ will be set to zero and one in case that the ﬂag unblock_on_resume_ is one and zero, respectively. 7.3.4 Class DropTail: A Child Class of Class Queue Consider class DropTail, a child class of class Queue, which is bound to the OTcl class Queue/DropTail in Program 7.10. The constructor of class DropTail creates a pointer q_ (Line 13) to a PacketQueue object, and sets pq_ derived from class Queue to be the same as q_ (Line 5). Throughout

152

7 Link and Buﬀer Management

the implementation, class DropTail refers to its buﬀer by q_ instead of pq_. Class DropTail overrides function enque(p) (Line 11 and Program 7.11) and deque() (Line 12) of class Queue. It also allows packet dropping at the front of the buﬀer, if the ﬂag drop_front_ (Line 14) is set to 1. Class DropTail does not override function recv(p,h). Therefore, it receives a packet through the function recv(p,h) of class Queue. Program 7.10 Declaration of class DropTail.
//~ns/queue/drop-tail.h 1 class DropTail : public Queue { 2 public: 3 DropTail() { 4 q_ = new PacketQueue; 5 pq_ = q_; 6 bind_bool("drop_front_", &drop_front_); 7 }; 8 ~DropTail() { delete q_; }; 9 protected: 10 int command(int argc, const char*const* argv); 11 void enque(Packet*); 12 Packet* deque(); 13 PacketQueue *q_; 14 int drop_front_; 15 }; //~ns/queue/drop-tail.cc 16 static class DropTailClass : public TclClass { 17 public: 18 DropTailClass() : TclClass("Queue/DropTail") {} 19 TclObject* create(int, const char*const*) { 20 return (new DropTail); 21 } 22 } class_drop_tail;

In Program 7.11, function enque(p) ﬁrst checks whether the incoming packet will cause buﬀer overﬂow (Line 3). If so, it will drop the packet either from the front (Lines 5–7) or from the tail (Line 9), where function drop(p) (Lines 7 and 9) belongs to class Connector (see Program 5.4). If the buﬀer has enough space, Line 10 will enqueue packet (p) to its buﬀer (q_).

7.4 A Sample Two-Node Network
We have introduced two basic NS2 components: nodes and links. Based on these two components, we now create a two-node network with a uni-

7.4 A Sample Two-Node Network

153

Program 7.11 Function enque of class DropTail.
//~ns/queue/drop-tail.cc 1 void DropTail::enque(Packet* p) 2 { 3 if ((q_->length() + 1) >= qlim_) 4 if (drop_front_) { 5 q_->enque(p); 6 Packet *pp = q_->deque(); 7 drop(pp); 8 } else 9 drop(p); 10 else 11 q_->enque(p); 12 }

udp attach-agent

n1 run

SimpleLink simplex-link

n2 attach-agent

null

Fig. 7.4. A two-node network with a uni-directional link and the instprocs of class Simulator.

directional link and show the packet ﬂow mechanism within this network in Fig. 7.4. 7.4.1 Network Construction The network in Fig. 7.4 consists of a beginning node (n1), a termination node (n2), a SimpleLink connecting n1–n2, a source transport layer agent (udp), and a sink transport layer agent (null). This network can be created using the following Tcl simulation script: set set set $ns set set $ns $ns ns [new Simulator] n1 [$ns node] n2 [$ns node] simplex-link $n1 $n2 DropTail udp [new Agent/UDP] null [new Agent/Null] attach-agent $n1 $udp attach-agent $n2 $null

Here, command $ns node creates a Node object. The internal mechanism of the node construction process was described in Section 6.6. The statement $ns simplex-link $n1 $n2 DropTail creates a unidirectional SimpleLink object, which connects node n1 to node n2. The link

154

7 Link and Buﬀer Management

bandwidth and delay are bps and seconds, respectively. The buﬀer in the link is of class DropTail. From Section 6.6.3, the commands $ns attach-agent $n1 $udp and $ns attach-agent $n2 $null set the target of agent udp to be the entry of Node n1, and installs agent null in the demultiplexer of Node n2. 7.4.2 Packet Flow Mechanism To deliver a packet “*p” from agent udp to null, (i) Agent udp sends packet *p to the entry of Node n1.3 (ii) Packet *p is sent to the head classiﬁer classifier_ (which is of class DestHashClassifier) of Node n1. (iii) The DestHashClassifier object classifier_ examines the header of packet *p. In this case, the packet is destined to the Node n2. Therefore, it forwards the packet to the link head of the connecting SimpleLink object. (iv) The link head forwards the packet to the connecting Queue object. (v) The Queue object enqueues the packet. If not blocked, it will forward the head of the line packet to the connecting LinkDelay object and set its status to blocked. (vi) Upon receiving a packet, the LinkDelay object schedules the two following events: (a) Packet departure event, which indicates that packet transmission is complete. This event unblocks the associated Queue object. (b) Packet arrival event, which indicates the packet arrival at the connecting TTLChecker object. (vii) The TTLChecker object receives the packet, and decrements the TTL ﬁeld of the packet header. If the TTL ﬁeld of the packet is non-positive, the TTLChecker object will drop the packet. Otherwise, it will forward the packet to the entry of Node n2 (see ﬁle ˜ns/common/ttl.cc). (viii) Node n2 forwards the packet to the head classiﬁer (classifier_). Since the packet is destined to itself, the packet is forwarded to the demultiplexer (dmux_). (ix) The demultiplexer forwards the packet to the agent null installed in the demultiplexer.

7.5 Chapter Summary
This chapter focuses on class SimpleLink, a basic link class which can be used to connect two nodes. The connection between two nodes n1 and n2 can be created by the following instprocs:
3

Note that, each object sends a packet *p to its downstream object by invoking target -> recv(p,h), where target is a pointer to the downstream object.

7.5 Chapter Summary

155

$ns simplex-link $n1 $n2 $ns duplex-link $n1 $n2 where the bandwidth and delay of the SimpleLink object are bps and seconds, respectively. Also the type of queue implemented in the SimpleLink object is . A SimpleLink object models packet transmission time, link propagation delay, and packet buﬀering. Here, packet transmission time is the time repacket size , while the link quired to transmit a packet, and is computed by bandwidth propagation time is the time required to deliver a data bit from the beginning to the end of the SimpleLink object. As shown in Fig. 7.1, an OTcl SimpleLink object consists of instvars head_, drophead_, queue_, link_, and ttl_, whose classes are Connector, Connector, Queue, DelayLink, and TTLChecker, respectively. • Instvars head_ and drophead_ act as an entry point and a dropping point, respectively, of a SimpleLink object. • Instvar link_ models packet transmission time and link propagation delay of a link. When receiving a packet, it schedules two events: packet departure from the beginning node and packet arrival at the terminating node. • Instvar queue_ models packet buﬀering mechanism in a SimpleLink object. It operates very closely with the instvar link_. Upon receiving a packet, the instvar queue_ enques the packet. If not blocked, it will block itself and forward the packet as well as the associated queue handler to the instvar link_. When the packet departure event (scheduled by the instvar link_) is dispatched, instvar queue_ is unblocked (i.e., being called back) and allowed to transmit another packet. • Instvar ttl_ is a packet time to live checker, which drops packets which stay in the network for longer than a speciﬁed period of time.

8 Packets, Packet Headers, and Header Format

Generally, a packet consists of packet header and data payload. Packet header stores packet attributes (e.g., source and destination IP addresses) necessary for packet delivery, while data payload contains user information. Although this concept is typical in practice, NS2 models packets diﬀerently. In most cases, NS2 extracts information from data payload and stores the information into packet header. This idea removes the need to process data payload at runtime. For example, instead of counting the number of bits in a packet, NS2 stores packet size in variable hdr_cmn::size_ (see Section 8.3.5), and accesses this variable at runtime.1 This chapter discusses how NS2 models packets. Section 8.1 gives an overview on NS2 packet modeling. Section 8.2 discusses the packet allocation and deallocation processes. Sections 8.3 and 8.4 show the details of packet header and data payload, respectively. We give a guideline of how to customize packets (i.e., to deﬁne a new packet type and activate/deactivate new and existing protocols) in Section 8.5. Finally, the chapter summary is given in Section 8.6.

8.1 An Overview of Packet Modeling Principle
8.1.1 Packet Architecture Figure 8.1 shows the architecture of an NS2 packet model. From Fig. 8.1, a packet model consists of four main parts: actual packet, class Packet, protocol speciﬁc headers, and packet header manager. • Actual Packet: An actual packet refers to the portion of memory which stores packet header and data payload. NS2 does not directly access either the packet header or the data payload. Rather, it uses pointers
1

For example, class LinkDelay determines packet size from a variable hdr cmn:: size when computing packet transmission time (see Line 11 of Program 7.3).

T. Issariyakul, E. Hossain, Introduction to Network Simulator NS2, DOI: 10.1007/978-0-387-71760-9 8, c Springer Science+Business Media, LLC 2009

158

8 Packets, Packet Headers, and Header Format
Packet
bits_ hdrlen_ data_

Actual Packet Packet Header Data Payload

AppData

type_

PROTOCOL SPECIFIC HEADERS COMMON HEADER

class_cmnhdr
CommonHeaderClass (mapping class) sizeof()

hdr_cmn (C++ Class) offset_ hdr_ip (C++ Class)

IP HEADER

class_iphdr
IPHeaderClass (mapping class)

offset_

...
PacketHeader/IP (OTcl Class) hdrlen_

hdrlen_

PacketHeader/Common (OTcl Class)

classname_

...
C++ OTcl
PacketHeaderManager

hdrlen_

“a” points to “b” a b a b

Bind “a” to “b” a b

Store “a” in “b”

Arrow legend

Fig. 8.1. Packet modeling in NS2.

bits_ and data_ of class Packet to access packet header and data payload, respectively. The details of packet header and data payload will be given in Sections 8.3 and 8.4, respectively. • Class Packet: Declared in Program 8.1, class Packet is the C++ main class which represents packets. It contains the following variables and functions:

C++ variables of class Packet bits_ A string which contains packet header data_ Pointer to an AppData object which contains data payload fflag_ Set to true if the packet is currently referred to by other objects and false otherwise free_ Pointer to the head of the packet free list ref_count_ Number of objects which currently refer to the packet next_ Pointer to the next packet in the linked list of packets hdr_len_ Length of packet header

8.1 An Overview of Packet Modeling Principle

159

Program 8.1 Declaration of class Packet.
1 2 3 4 5 6 7 8 9 10 11 12 //~/ns/common/packet.h class Packet : public Event { private: unsigned char* bits_; AppData* data_; static void init(Packet*) {bzero(p->bits_, hdrlen_);} bool fflag_; protected: static Packet* free_; int ref_count_; public: Packet* next_; static int hdrlen_; //Packet Allocation and Deallocation Packet() : bits_(0), data_(0), ref_count_(0), next_(0) { } inline unsigned char* const bits() { return (bits_); } inline Packet* copy() const; inline Packet* refcopy() { ++ref_count_; return this; } inline int& ref_count() { return (ref_count_); } static inline Packet* alloc(); static inline Packet* alloc(int); inline void allocdata(int); static inline void free(Packet*); //Packet Access inline unsigned char* access(int off){return &bits_[off]);};

13 14 15 16 17 18 19 20 21

22 23 }

C++ functions of class Packet init(p) Clears the packet header bits_ of the input packet p. copy() Returns a pointer to a duplicated packet. refcopy() Increases the number of objects, which refer to the packet, by one. alloc() Creates a new packet and returns a pointer to the created packet. alloc(n) Creates a new packet with “n” bytes of data payload and returns a pointer to the created packet. allocdata(n) Allocates “n” bytes of data payload to the variable data_. free(p) Deallocates packet p. access(off) Retrieves a reference to a certain point (speciﬁed by the oﬀset “off”) of the variable bits_ (i.e., packet header). • Protocol Speciﬁc Header: From Fig. 8.1, packet header consists of several protocol speciﬁc headers. Each protocol speciﬁc header uses a

160

8 Packets, Packet Headers, and Header Format

contiguous portion of packet header to store its packet attributes. In common with most TclObjects, there are three classes related to each protocol speciﬁc header: a C++ class, an OTcl class, and a mapping class. – A C++ class (e.g., hdr_cmn or hdr_ip): provides a sturcture to store packet attributes. – An OTcl class (e.g., PacketHeader/Common or PacketHeader/IP): acts as an interface to the OTcl domain. NS2 uses this class to conﬁgure packet header from the OTcl domain. – A mapping class (e.g., CommonHeaderClass or IPHeaderClass): binds a C++ class to an OTcl class. We will discuss the details of protocol speciﬁc header later in Section 8.3.5. • Packet Header Manager: A packet header manager maintains a list of active protocols, and conﬁgures all active protocol speciﬁc headers to setup a packet header. It has an instvar hdrlen_ which indicates the length of packet header consisting of protocol speciﬁc headers. Instvar hdrlen_ is bound to a variable hdrlen_ of class Packet. Any change in one of these two variables will result in an automatic change in another. • Data Payload: From Line 4 in Program 8.1, the pointer data_ points to data payload, which is of class AppData. We will discuss the details of data payload in Section 8.4. 8.1.2 A Packet as an Event: A Delayed Packet Reception Event Derived from class Event (Line 1 in Program 8.1), class Packet can be placed on the simulation time line (see the details in Chapter 4). In Section 4.2, we mentioned two main classes derived from class Event: class AtEvent and class Packet. We also mentioned that an AtEvent object is an event created by a user from a Tcl simulation script. This chapter discusses the details of another derived class of class Event: class Packet. As discussed in Section 5.2.2, NS2 models a delayed action by placing an event corresponding to the action on the simulation timeline at a certain delayed time. Derived from class Event, class Packet can be placed on the simulation timeline to signify a delayed packet reception. For example, the following statement (see Line 8 in Program 7.4) schedules a packet reception event, where the NsObject *target_ receives a packet *p at txt+delay_ seconds in future: s.schedule(target_, p, txt + delay_) where function schedule(...) of class Scheduler deﬁned in Program 4.7 takes an Event pointer as its second input argument. A Packet pointer is cast to be an Event pointer before being fed as the second input argument. At the ﬁring time, the Scheduler dispatches the scheduled event (i.e., *p) and invokes target->handle(p), which executes “target_->recv(p)” to forward packet *p to the NsObject pointer *target_.

8.1 An Overview of Packet Modeling Principle

161

8.1.3 A Linked List of Packets Apart from the above 4 main packet components, a Packet object contains a pointer next_ (Line 11 in Program 8.1), which helps formulating a linked list of Packet objects (e.g., Packet List in Fig. 8.2). Program 8.2 shows the implementation of functions enque(p) and deque() of class PacketQueue. Function enque(p) (Lines 3–13) puts a Packet object *p to the end of the queue. If the PacketQueue is empty, NS2 sets head_, tail_, and p to point to the same place2 (Line 5). Otherwise, Lines 7–8 set p as the last packet in the PacketQueue, and shift variable tail_ to the last packet pointer p. Since the pointer tail_ is the last pointer of PacketQueue, Line 10 sets the pointer tail_->next_ to 0 (i.e, points to NULL).

Fig. 8.2. A linked list of packets and a free packet list.

Function deque() (Lines 14–21) retrieves a pointer to the packet at the head of the buﬀer. If there is no packet in the buﬀer, the function deque() will return a NULL pointer (Line 15). If the buﬀer is not empty, Line 17 will shift the pointer head_ to the next packet, Line 19 will decrease the length of PacketQueue object by one, and Line 20 will return the packet pointer p which was set to the pointer head_ in Line 16.

8.1.4 Free Packet List Unlike most NS2 objects, a Packet object, once created, will not be destroyed until the simulation terminates. NS2 keeps Packet objects which are no longer
2

Note that, head and tail are pointers to the ﬁrst and the last Packet objects, respectively, in a PacketQueue object.

162

8 Packets, Packet Headers, and Header Format

Program 8.2 Functions enque and deque of class PacketQueue.
1 2 3 4 5 6 7 8 9 10 11 12 13 //~/ns/common/queue.h class PacketQueue : public TclObject { ... virtual Packet* enque(Packet* p) { // Returns previous tail Packet* pt = tail_; if (!tail_) head_= tail_= p; // if the PacketQueue is empty else { tail_->next_= p; tail_= p; } tail_->next_= 0; ++len_; return pt; } virtual Packet* deque() { if (!head_) return 0; Packet* p = head_; head_= p->next_; // 0 if p == tail_ if (p == tail_) head_= tail_= 0; --len_; return p; } ...

14 15 16 17 18 19 20 21 22 23 };

in use in a free packet list (see Fig. 8.2). When NS2 needs a new packet, it ﬁrst checks whether the free packet list is empty. If not, it will take a Packet object from the list. Otherwise, it will create another Packet object. We will discuss the details of how to allocate and deallocate a Packet object later in Section 8.2. There are two variables which are closely related to the packet allocation/deallocation process: fflag_ and free_. Each Packet object uses a variable fflag_ (Line 6 in Program 8.1) to indicate whether it is in use. Variable fflag_ is set to true, when the Packet object is in use, and set to false otherwise. Shared by all the Packet objects, a static pointer free_ (Line 8 in Program 8.1) is a pointer to the ﬁrst packet on the free packet list. Each packet on the free packet list uses its variable next_ to help form a link list of free Packet objects. This linked list of free packets is referred to as a free packet list. Although NS2 does not return memory allocated to a Packet object to the system, it does return the memory used by packet header (i.e., bits_) and data payload (i.e., data_) to the system (see Section 8.2.2), when the packet is deallocated. Since most memory required to store a Packet object is consumed by packet header and data payload, maintaining a free packet list does not result in a signiﬁcant waste of memory.

8.2 Packet Allocation and Deallocation

163

8.2 Packet Allocation and Deallocation
Unlike most of the NS2 objects,3 a Packet object is allocated and deallocated using static functions alloc() and free(p) of class Packet, respectively. If possible, function alloc() takes a Packet object from the free packet list. Only when the free packet list is empty, does the function alloc() creates a new Packet object using new. Function free(p) deallocates a Packet object, by returning the memory allocated for packet header and data payload to the system and storing the not-in-use Packet pointer p in the free packet list for future reuse. The details of packet allocation and deallocation will be discussed in the next two sections. 8.2.1 Packet Allocation Program 8.3 shows the details of function alloc() of class Packet, the packet allocation function. Function alloc() returns a pointer to an allocated Packet object to the caller. This function consists of two parts: packet allocation in Lines 3–15, and packet initialization in Lines 16–24. Consider the packet allocation in Lines 3–15. Line 3 declares p as a pointer to a Packet object, and sets the pointer p to point to the ﬁrst packet on the free packet list4 . If the free packet list is empty (i.e., p = 0), NS2 will create a new Packet object (in Line 11), and allocate memory space with size “hdrlen_” bytes for the packet header in Line 12. Variable hdrlen_ is not conﬁgured during the construction of a Packet object. Rather, it is set up in the Network Conﬁguration Phase (see Section 8.3.8), and is used by the function alloc() to create packet header. Function alloc() does not allocate memory space for data payload. When necessary, NS2 creates data payload by using the function allocdata(n) (see Lines 8–14 in Program 8.4), which will be discussed in detail later in this section. If the free packet list is non-empty, function alloc() will execute Lines 5–9 in Program 8.3 (see also the diagram in Fig. 8.3). In this case, function alloc() ﬁrst makes sure that nobody is using the Packet object p, by asserting that fflag_ is false (Line 5).5 Then, Line 6 shifts the pointer free_ by one position. Lines 8–9 initialize two variables (uid_ and time_) of class Event (i.e., the mother class of class Packet) to be zero. Line 23 removes the packet from the free list by setting p->next_ to zero. After the packet allocation process is complete, Lines 16–24 initialize the allocated Packet object. Line 16 invokes function init(p), which initializes
3

4 5

Generally, NS2 creates and destroys most objects by using procedures new and delete, respectively. Again, free is the pointer to the ﬁrst packet on the free packet list. The C++ function assert(cond) can be used for an integrity check. It does nothing if the input argument cond is true. Otherwise, it will initiate an error handling process (e.g., showing an error on the screen).

164

8 Packets, Packet Headers, and Header Format

Program 8.3 Function alloc of class Packet.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 //~/ns/common/packet.h inline Packet* Packet::alloc() { //Packet Allocation Packet* p = free_; if (p != 0) { assert(p->fflag_ == FALSE); free_ = p->next_; assert(p->data_ == 0); p->uid_ = 0; p->time_ = 0; } else { p = new Packet; p->bits_ = new unsigned char[hdrlen_]; if (p == 0 || p->bits_ == 0) abort(); } //Packet Initialization init(p); // Initialize bits_[] (HDR_CMN(p))->next_hop_ = -2; // -1 reserved for IP_BROADCAST (HDR_CMN(p))->last_hop_ = -2; // -1 reserved for IP_BROADCAST p->fflag_ = TRUE; (HDR_CMN(p))->direction() = hdr_cmn::DOWN; /* setting all direction of pkts to be downstream as default; until channel changes it to +1 (upstream) */ p->next_ = 0; return (p);

16 17 18 19 20 21 22 23 24 25 }

Packet free_ next_

Packet next_ …

p NULL

Fig. 8.3. Diagram of packet allocation when the free packet list is non-empty. The dotted lines show the actions caused by function alloc of class Packet.

8.2 Packet Allocation and Deallocation

165

Program 8.4 Functions alloc, allocdata, and copy of class Packet.
1 2 3 4 5 6 7 //~/ns/common/packet.h inline Packet* Packet::alloc(int n) { Packet* p = alloc(); if (n > 0) p->allocdata(n); return (p); }

8 inline void Packet::allocdata(int n) 9 { 10 assert(data_ == 0); 11 data_ = new PacketData(n); 12 if (data_ == 0) 13 abort(); 14 } 15 inline Packet* Packet::copy() const 16 { 17 Packet* p = alloc(); 18 memcpy(p->bits(), bits_, hdrlen_); 19 if (data_) 20 p->data_ = data_->copy(); 21 return (p); 22 }

the header of packet *p. From Line 5 in Program 8.1, invocation of function init(p) executes “bzero(p-> bits_,hdrlen_)”, which clears bits_ to zero.6 Line 19 sets fflag_ to be true, indicating that the packet *p is now in use. Line 23 sets the pointer p->next_ to be zero. Lines 17, 18, and 20 initialize the common header. We will discuss packet header in greater detail in Section 8.3.2. Apart from function alloc(), other relevant functions include alloc(n), allocdata(n), and copy() (See Program 8.4). Function alloc(n) allocates a packet (Line 3), and invokes allocdata(n) (Line 5). Function alloc(n) creates data payload with size “n” bytes (by invoking new PacketData(n) in Line 11). We will discuss the details of data payload later in Section 8.4. Function copy() returns a replica of the current Packet object. The only diﬀerence between the current and the replicated Packet objects is the unique ID (uid_) ﬁeld. This function is quite useful, since we often need to create a packet which is the same as or slightly diﬀerent from an original packet. This
6

Function bzero takes two arguments – the ﬁrst is a pointer to the buﬀer and the second is the size of the buﬀer – and sets all values in a buﬀer to zero.

166

8 Packets, Packet Headers, and Header Format

function ﬁrst allocates a packet pointer p in Line 17. Then, it copies packet header and data payload to the packet *p in Lines 18 and 20, respectively. Despite its name, function refcopy() (Line 16 in Program 8.1) does not create a copy of a Packet object. Rather, it returns the pointer to the current Packet object. For example, suppose p is a Packet pointer. Then, x = p and x = p->refcopy() both store p in x. However, function refcopy() also keeps track of the number of objects which share the same Packet object, by using variable ref_count_ (Line 9 in Program 8.1). This variable is initialized to 0 in the constructor of class Packet (Line 13 in Program 8.1). It is incremented by one when function ref_copy() (Line 16 in Program 8.1) is invoked, indicating that a new object starts using the current Packet object. Similarly, it is decremented by one when function free(p) (see Section 8.2.2) is invoked, indicating that an object has stopped using the current Packet object. 8.2.2 Packet Deallocation When a packet *p is no longer in use, NS2 deallocates the packet by using function free(p). By deallocation, NS2 returns the memory used to store packet header and data payload to the system, sets the pointer data_ to zero, and stores the Packet object in the free packet list. Note that although the value of bits_ is not set to zero, the memory location stored in bits_ is no longer accessible by bits_. It is very important not to use bits_ after packet deallocation. Otherwise, NS2 will encounter a (memory share violation) runtime error. The details of function free(Packet*) are shown in Program 8.5. Before returning a Packet object to the free packet list, we need to make sure that (i) The packet is in use (i.e., p->fflag_ = 1 in Line 3), since there is no point in deallocating a packet which has already been deallocated. (ii) No object is using the packet; the variable ref_count_ is Zero (Line 4), where ref_count_ stores the number of objects which are currently using the packet. (iii) The packet is no longer on the simulation time line (i.e., p->uid_uid_) is non-positive, and therefore is no longer on the simulation timeline.7 (iv) The data payload pointer data_ must not point to NULL (p->data_=0 in Line 6), when returning the memory occupied by data payload to the system. NS2 allows more than one simulation object to share the same Packet object. To deallocate a packet, NS2 must ensure that the packet is no longer
7

From Fig. 4.2, an event with positive unique ID (e.g, uid is 2 or 6) was scheduled but has not been dispatched.

8.2 Packet Allocation and Deallocation

167

Program 8.5 Function free of class Packet.
//~/ns/common/packet.h 1 inline void Packet::free(Packet* p) 2 { 3 if (p->fflag_) { 4 if (p->ref_count_ == 0) { 5 assert(p->uid_ data_ != 0) { 7 delete p->data_; 8 p->data_ = 0; 9 } 10 init(p); 11 p->next_ = free_; 12 free_ = p; 13 p->fflag_ = FALSE; 14 } else { 15 --p->ref_count_; 16 } 17 } 18 }

used by any simulation object. Again, NS2 keeps the number of objects sharing a packet in variable ref_count_. If ref_count_>0–meaning an object is invoking function free(p) while other objects are still using the packet *p, function free(p) will simply reduce ref_count_ by one, indicating that one object stops using the packet (Line 15).8 On the other hand, if ref_count_ is zero–meaning no other object is using the packet, Lines 5–13 will then clear packet header and data payload, and store the Packet object in the free packet list. If all the above four conditions are satisﬁed, function free(p) will execute Lines 6–13 in Program 8.5. The schematic diagram for this part is shown in Fig. 8.4. Line 7 returns the memory used by data payload to the system. Line 8 sets the pointer data_ to zero. Line 10 returns the memory used by packet header of a packet *p to the system by invoking function init(p) (see Line 5 of Program 8.1). Function free(p) does not set the variable bit_ to zero. Do not try to access bit_ after this point, since doing so will cause a runtime error. Lines 11 and 12 place the packet as the ﬁrst packet on the free packet list. Finally, Line 13 sets fflag_ to false, indicating that the packet is no longer in use.

8

If the Packet object is deallocated when ref count > 0, simulation objects may later try to access the deallocated Packet object and cause a runtime error.

168

8 Packets, Packet Headers, and Header Format
Packet
p bits_ data_ next_

Packet next_ …

NULL NULL

Packet Header

Data Payload

… next_ free_

Packet

Fig. 8.4. The process of returning a packet to the packet free list. The dotted lines show the action caused by function free of class Packet.

8.3 Packet Header
As a part of a packet, packet header contains packet attributes such as packet unique ID, and IP address. Again, packet header is stored in variable bits_ of class Packet (see Line 3 of Program 8.1). The variable bits_ is declared as a string (i.e., a Bag of Bits (BOB)), and has no structure to store packet attributes. NS2 hence imposes a two-level structure on variable bits_, as shown in Fig. 8.5. The ﬁrst level divides the entire packet header into protocol speciﬁc headers. The location of each protocol speciﬁc header on bits_ is identiﬁed by its variable offset_. The second level imposes a packet attribute storing structure on each protocol speciﬁc header. On this level, packet attributes are stored as members of a C++ struct data type. In practice, a packet contains only relevant protocol speciﬁc headers. An NS2 packet on the other hand includes all protocol speciﬁc headers into a packet header, regardless of packet type. Every packet uses the same amount of memory to store the packet header. The amount of memory is stored in the variable hdrlen_ of class Packet in Line 12 of Program 8.1, and is declared as a static variable. The variable hdrlen_ has no relationship to simulation packet size. For example, TCP and UPD packets may have diﬀerent sizes. The values stored in the corresponding variable hdr_cmn::size_ may be diﬀerent; however, the values stored in the variable Packet::hdrlen_ for both TCP and UDP packets are the same. In the following, we ﬁrst discuss the ﬁrst level packet header composition in Section 8.3.1. Sections 8.3.2 and 8.3.3 shows examples of protocol speciﬁc headers: common packet header and IP packet header. Section 8.3.4 discusses

8.3 Packet Header
Packet
bits_ data_ next_

169

…

Packet Header

Data Payload

1st Level ... TCP Header IP Header Common Header ... 2nd Level uid_ ptype_ offset_ size_
... ...

Protocol specific header size is determined during the compilation.

hdr_tcp::offset_ hdr_ip::offset_ hdr_cmn::offset_ Packet header size is determined during the construction of the simulator

Fig. 8.5. Architecture of packet header.

one of the main packet attributes: packet type. Section 8.3.5 explains the details of protocol speciﬁc header (i.e., the second level packet header composition). Section 8.3.6 demonstrates how packet attributes stored in packet header are accessed. Section 8.3.7 discusses one of the main packet header component, a packet header manager, which maintains the active protocol list and sets up the oﬀset value for each protocol. Finally, Section 8.3.8 presents the packet header construction process. 8.3.1 An Overview of First Level Packet Composition: Oﬀseting Protocol Speciﬁc Header on the Packet Header On the ﬁrst level, NS2 puts together all relevant protocol speciﬁc headers (e.g., common header, IP header, TCP header) and composes a packet header (see Fig. 8.5). Conceptually, NS2 allocates a contiguous part on the packet header for a protocol speciﬁc header. Each protocol speciﬁc header is oﬀset from the beginning of packet header. The distance between the beginning of packet header and that of a protocol speciﬁc header is stored in the member variable offset_ of the protocol speciﬁc header. For example, hdr_cmn, hdr_ip, and hdr_tcp–which represent common header, IP header, and TCP header–

170

8 Packets, Packet Headers, and Header Format

store their oﬀset values of variables hdr_cmn::offset_, hdr_ip::offset_, and hdr_tcp::offset_, respectively. 8.3.2 Common Packet Header Common packet header contains packet attributes which are common to all packets. It employs C++ struct data type hdr_cmn to indicate how the packet attributes are stored. Program 8.6 shows a part of hdr_cmn declaration. The main member variables of hdr_cmn are as follows: Program 8.6 Declaration of C++ hdr cmn struct data type.
1 2 3 4 5 6 7 //~/ns/common/packet.h struct hdr_cmn { enum dir_t { DOWN= -1, NONE= 0, UP= 1 }; packet_t ptype_; // packet type int size_; // simulated packet size int uid_; // unique id dir_t direction_; // direction: 0=none, 1=up, -1=down static int offset_; // offset for this header inline static hdr_cmn* access(const Packet* p) { return (hdr_cmn*) p->access(offset_); } inline static int& offset() { return offset_; } inline packet_t& ptype() { return (ptype_); } inline int& size() { return (size_); } inline int& uid() { return (uid_); } inline dir_t& direction() { return (direction_); }

8 9 10 11 12 13 14 15 16 };

ptype_ The packet type (not the type of protocol speciﬁc header). size_ The packet size. Unlike actual packet transmission, the number of bits requires to hold a packet has no relationship to simulation packet size. During simulation, NS2 uses variable hdr_cmn::size_ as the packet size. uid_ The ID which is unique to every packet. dir_t The transmitting direction which can be downstream (–1), upstream (1), or not-in-use (0). By default, dir_t is set to downstream (see Line 20 in Program 8.3). offset_ The memory location relative to the beginning of packet header from which the common header is stored (see Section 8.3.1 and Fig. 8.5). From Fig. 8.6, most functions of class hdr_cmn act as an interface to access its variables. Apart from these functions, function access(p) in Lines 8–10

8.3 Packet Header

171

is perhaps the most important function of hdr_cmn. It is used to access a protocol speciﬁc header of the input Packet object *p. We will discuss the packet header access mechanism in greater detail in Section 8.3.6. Program 8.7 Declaration of C++ hdr ip struct data type.
//~/ns/common/ip.h 1 struct hdr_ip { 2 ns_addr_t src_; 3 ns_addr_t dst_; 4 int ttl_; 5 int fid_; 6 int prio_; 7 static int offset_; 8 inline static int& offset() { return offset_; } 9 inline static hdr_ip* access(const Packet* p) { 10 return (hdr_ip*) p->access(offset_); 11 } 12 ns_addr_t& src() { return (src_); } 13 nsaddr_t& saddr() { return (src_.addr_); } 14 int32_t& sport() { return src_.port_;} 15 ns_addr_t& dst() { return (dst_); } 16 nsaddr_t& daddr() { return (dst_.addr_); } 17 int32_t& dport() { return dst_.port_;} 18 int& ttl() { return (ttl_); } 19 int& flowid() { return (fid_); } 20 int& prio() { return (prio_); } 21 };

8.3.3 IP Packet Header Represented by C++ struct data type hdr_ip, IP packet header contains information about source and destination of a packet. Program 8.7 shows a part of hdr_ip declaration. IP packet header contains the following ﬁve main variables which contain IP-related packet information (see Lines 2–6 in Program 8.7): src_ Source node’s address of the packet dst_ Destination node’s address of the packet ttl_ Time to live for the packet fid_ Flow ID of the packet prio_ Priority level of the packet NS2 utilizes data type ns_addr_t deﬁned in ﬁle ˜ns/conﬁg.h to store node address. From Program 8.8, ns_addr_t is a struct data type, which contains

172

8 Packets, Packet Headers, and Header Format

two members: addr_ and port_. Both members are of type int32_t, which is simply an alias for int data type (see Line 5 and ﬁle ˜ns/autoconf-win32.h). While addr_ speciﬁes the node address, port_ identiﬁes the attached port (if any). Program 8.8 Declaration of C++ ns addr t struct data type, and its int32 t alias
1 2 3 4 //~/ns/config.h struct ns_addr_t { int32_t addr_; int32_t port_; }; //~/ns/autoconf-win32.h typedef int int32_t;

5

The variables src_ and dst_ of IP header are of class ns_addr_t. Hence, src_.addr_ and src_.port_ store the node address and the port of the sending agent, respectively. Similarly, the packet will be sent to a receiving agent attached to port dst_.port_ of a node with address dst_.addr_. Lines 7–11 in Program 8.7 declare variable offset_, function offset(off) and function access(p), which are essential to access IP header of a packet. We will discuss the packet access mechanism later in Section 8.3.6. Lines 12–20 in Program 8.7 are functions which return the values of the variables. 8.3.4 Packet Type Although stored in common header, packet type is attributed to an entire packet, not to a protocol speciﬁc header. Each packet corresponds to only one packet type but may contain several protocol speciﬁc headers. For example, a packet can be encapsulated by both TCP and IP protocols. However, its type can be either audio or TCP packet, but not both. NS2 stores a packet type in a member variable ptype_ of a common packet header. The type of the variable ptype_ is enum packet_t deﬁned in Program 8.9. Again, members of enum are integers which are mapped to strings. From Fig. 8.9, PT_TCP (Line 2) and PT_UDP (Line 3) are mapped to 0 and 1, respectively. Since packet_t declares PT_NTYPE (representing undeﬁned packet type) as the last member, the value of PT_NTYPE is Np − 1, where Np is the number of packet_t members. NS2 provides 60 built-in packet types, meaning the default value of PT_NTYPE is 59. From Lines 11–30 in Program 8.9, class p_info maps each member of packet_t to a description string. It has a static associative array variable, name_ (Line 28). The index and value of name_ are the packet type, and the corresponding description string, respectively. Class p_info also has one

8.3 Packet Header

173

Program 8.9 Declaration of enum packet t type and class p info.
//~/ns/common/packet.h 1 enum packet_t { 2 PT_TCP, 3 PT_UDP, 4 PT_CBR, 5 PT_AUDIO, 6 PT_VIDEO, 7 PT_ACK, 8 ... 9 PT_NTYPE // This MUST be the LAST one 10 } 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 class p_info { public: p_info() { name_[PT_TCP]= "tcp"; name_[PT_UDP]= "udp"; name_[PT_CBR]= "cbr"; name_[PT_AUDIO]= "audio"; name_[PT_VIDEO]= "video"; name_[PT_ACK]= "ack"; ... name_[PT_NTYPE]= "undefined"; } const char* name(packet_t p) const { if ( p ptype(); printf("Example Test: Class Agent allocates a packet with type %s\n", packet_info.name(pt)); 9 getchar(); 10 //------- End Additional Codes --------11 return (p); 12 } 1 2 3 4 5 6 7 8 where Lines 5–10 are added to the original codes. Line 6 retrieves the reference “ch” to the common packet header (see Section 8.3.6). Line 7 obtains the packet type stored in the common header by using function ptype(), and assigns the packet type to variable pt. Note that, variable packet_info is a global variable of class p_info. When the variable pt is fed as an input argument, function packet_info.name(pt) returns the description string corresponding to the packet_t object “pt” (Line 8). After re-compiling the code, the simulation should show the type of every allocated packet on the screen. For example, when running the Tcl simulation script in Programs 2.1–2.2 provided in Chapter 2, the following result should appear on the screen: >> ns myfirst_ns.tcl Example Test: Class Agent allocates a packet with type cbr Example Test: Class Agent allocates a packet with type cbr Example Test: Class Agent allocates a packet with type cbr . . . 8.3.5 Protocol Speciﬁc Headers A protocol speciﬁc header stores packet attributes relevant to the underlying protocol only. For example, common packet header holds basic packet attributes such as packet unique ID, packet size, packet type, and so on. IP packet header contains IP packet attributes such as source and destination IP addresses and port numbers. There are 48 classiﬁcations of packet headers.

8.3 Packet Header

175

The complete list of protocol speciﬁc headers with their descriptions is given in [15]. Each protocol speciﬁc header involves three classes discussed below. A Protocol Speciﬁc Header C++ Class In C++, NS2 uses a struct data type to represent a protocol speciﬁc header. It stores packet attributes and its oﬀset value in members of a struct data type. It also provides a function access(p) which returns the reference to the protocol speciﬁc header of a packet *p. Representing a protocol speciﬁc header, each struct data type is named using format hdr_, where XXX is an arbitrary string representing the type of a protocol speciﬁc header. For example, the C++ class name for common packet header is hdr_cmn. In the C++ domain, protocol speciﬁc headers are declared but not instantiated. Therefore, NS2 uses a struct data type (rather than a class) to represent protocol speciﬁc headers, and no constructor is required for a protocol speciﬁc header. Hereafter, we will refer to struct and class interchangeably. A Protocol Speciﬁc Header OTcl Class NS2 deﬁnes a shadow OTcl class for each C++ protocol speciﬁc header class. An OTcl class acts as an interface to the OTcl domain. It is named with the format PacketHeader/, where XXX is an arbitrary string representing a protocol speciﬁc header. For example, the OTcl class name for common packet header is PacketHeader/Common. A Protocol Speciﬁc Header Mapping Class A mapping class is responsible for binding OTcl and C++ class names together. All the packet header mapping classes derive from class PacketHeaderClass which is a child class of class TclClass. A mapping class is named with format HeaderClass, where XXX is an arbitrary string representing a protocol speciﬁc header. For example, the mapping class name for common packet header is CommonHeaderClass. Program 8.10 shows the declaration of class PacketHeaderClass, which has two key variables: hrdlen_ in Line 8 and offset_ in Line 9. The variable hdrlen_ represents the length of the protocol speciﬁc header.9 It is the system memory needed to store a protocol speciﬁc header C++ class. Variable offset_ indicates the location on packet header where the protocol speciﬁc header is used.
9

While variable hdrlen in class PacketHeaderClass represents the length of a protocol speciﬁc header, variable hdrlen in class Packet represents total length of packet header.

176

8 Packets, Packet Headers, and Header Format

Program 8.10 Declaration of class PacketHeaderClass.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 //~/ns/common/packet.h class PacketHeaderClass : public TclClass { protected: PacketHeaderClass(const char* classname, int hdrlen) : TclClass(classname), hdrlen_(hdrlen), offset_(0);{}; virtual int method(int argc, const char*const* argv); inline void bind_offset(int* off) { offset_ = off; }; inline void offset(int* off) {offset_= off;}; int hdrlen_; // # of bytes for this header int* offset_; // offset for this header public: TclObject* create(int argc,const char*const* argv){return 0;}; virtual void bind(){ TclClass::bind(); Tcl& tcl = Tcl::instance(); tcl.evalf("%s set hdrlen_ %d", classname_, hdrlen_); add_method("offset"); }; };

The constructor of class PacketHeaderClass in Lines 3–4 takes two input arguments. The ﬁrst input argument classname is the name of the corresponding OTcl class name (e.g., PacketHeader/Common). The second one, hdrlen, is the length of the protocol speciﬁc header C++ class. In Lines 3–4, the constructor feeds classname to the constructor of class TclClass, stores hdrlen in the member variable hdrlen_, and resets offset_ to zero. Function method(argc,argv) in Line 5 is an approach to take a C++ action from the OTcl domain. Functions bind_offset(off) in Line 6 and offset(off) in Line 7 are used to conﬁgure and retrieve, respectively, the value of variable offset_. Function create(argc,argv) in Line 11 does nothing, since no protocol speciﬁc header C++ object is ever. It will be overridden by the derived classes of class PacketHeaderClass. Function bind() in Lines 12–17 glues the C++ class to the OTcl class. Line 13 ﬁrst invokes function bind() of class TclClass, which performs the basic binding actions. Line 15 then exports variable hdrlen_ to the OTcl domain. Line 16 registers the OTcl method offset using function add_method(“offset”). Apart from the commands discussed in Section 3.4.4, an OTcl method is another way to invoke C++ functions from the OTcl domain. It is implemented in C++ via the following two steps. The ﬁrst step is to deﬁne a function method(ac,av). As can be seen from Program 8.11, the structure of function method is very similar to that of function command. A method “offset” sets the value of *offset_ to be what speciﬁed in the input argument (Line 7 in Program 8.11). The second step in method implementation is to register the name of the method by using a function “add_method(str)”, which

8.3 Packet Header

177

Program 8.11 Function method of class PacketHeaderClass.
//~/ns/common/packet.cc 1 int PacketHeaderClass::method(int ac, const char*const* av) 2 { 3 Tcl& tcl = Tcl::instance(); 4 ... 5 if (strcmp(argv[1], "offset") == 0) { 6 if (offset_) { 7 *offset_ = atoi(argv[2]); 8 return TCL_OK; 9 } 10 tcl.resultf("Warning: cannot set offset_ for %s",classname_); 11 return TCL_OK; 12 } 13 ... 14 return TclClass::method(ac, av); 15 }

takes the method name as an input argument. For class PacketHeaderClass, the method offset is registered from within function bind(...) (Line 16 of Program 8.10). A protocol speciﬁc header is implemented using a struct data type, and hence does not derive function command(...) from class TclObject10. It resorts to OTcl methods deﬁned in the mapping class to take C++ actions from the OTcl domain. We will show an example use of the method offset later in Section 8.3.8, when we discuss packet construction mechanism. Program 8.12 Declaration of class CommonHeaderClass.
//~/ns/common/packet.cc class CommonHeaderClass : public PacketHeaderClass { public: CommonHeaderClass() : PacketHeaderClass("PacketHeader/Common", sizeof(hdr_cmn)) { 4 bind_offset(&hdr_cmn::offset_); 5 } 6 } class_cmnhdr; 1 2 3

Consider, for example, a common packet header. Its C++, OTcl, and mapping classes are hdr_cmn, PacketHeader/Common, and CommonPacketHeader Class, respectively (see Table 8.1). Program 8.12 shows the declaration of
10

Since NS2 does not instantiate a protocol speciﬁc header object, it models a protocol speciﬁc header using struct data type.

178

8 Packets, Packet Headers, and Header Format Table 8.1. Classes and objects related to common packet header Class/Object C++ class OTcl class Mapping class Mapping variable Name hdr_cmn PacketHeader/Common CommonHeaderClass class_cmnhdr

class CommonPacketHeaderClass. As a child class of TclClass, a class mapping variable class_cmnhdr is instantiated at the declaration. Line 3 of the constructor invokes the constructor of its parent class PacketHeaderClass, which takes the OTcl class name (i.e., PacketHeader/Common) and the amount of memory needed to hold the C++ class (i.e., hdr_cmn) as input arguments. Here, “sizeof (hdr_cmn)” computes such the required amount of memory, which is fed as the second input argument. In Line 6 of Program 8.10, function bind_offset(&hdr_cmn::offset_) sets its variable offset_ to share the address with the input argument. Therefore, a change in hdr_cmn::offset_ will result in an automatic change in variable *offset_ of class CommonHeaderClass, and vice versa. 8.3.6 Packet Header Access Mechanism This section demonstrates how packet attributes stored in packet header can be retrieved and modiﬁed. NS2 employs a two-level packet header structure to store packet attributes. On the ﬁrst level, protocol speciﬁc headers are stored within a packet header. On the second level, each protocol speciﬁc header employs a C++ struct data type to store packet attributes. The header access mechanism consists of two major steps: (1) Retrieve a reference to a protocol speciﬁc header, and (2) Follow the structure of the protocol speciﬁc header to retrieve or modify packet attributes. In this section, we will explain the access mechanism through common packet header (see the corresponding class names in Table 8.1). Retrieving a Reference to Protocol Speciﬁc Header NS2 obtains a reference to a protocol speciﬁc header by of a packet *p using a function access(p) in the C++ class. A reference to the common header of a Packet object *p can be obtained by executing hdr_cmn::access(p) (see Example 8.2 below). Example 8.2. Consider function allocpkt() of class Agent shown in Program 8.13, which shows the details of functions allocpkt() and initpkt(p). Function allocpkt() in Lines 1–6 creates a Packet object and returns a pointer to the created object to the caller. Function allocpkt() ﬁrst invokes

8.3 Packet Header

179

Program 8.13 Functions allocpkt and initpkt of class Agent.
1 2 3 4 5 6 //~/ns/common/agent.cc Packet* Agent::allocpkt() const { Packet* p = Packet::alloc(); initpkt(p); return (p); }

7 Packet* Agent::initpkt(Packet* p) const 8 { 9 hdr_cmn* ch = hdr_cmn::access(p); 10 ch->uid() = uidcnt_++; 11 ch->ptype() = type_; 12 ch->size() = size_; 13 ... 14 hdr_ip* iph = hdr_ip::access(p); 15 iph->saddr() = here_.addr_; 16 iph->sport() = here_.port_; 17 iph->daddr() = dst_.addr_; 18 iph->dport() = dst_.port_; 19 ... 20 }

function alloc() of class Packet in Line 3 (see the details in Section 8.2.1). Then, Line 4 initializes the allocated packet, by invoking function initpkt(p). Finally, Line 5 returns the pointer p to the initialized Packet object to the caller. Function initpkt(p) follows the structure deﬁned in the protocol speciﬁc header C++ classes to set packet attributes to the default values. Lines 9 and 14 in Program 8.13 execute the ﬁrst step in the access mechanism: retrieve references to common packet header ch and IP header iph, respectively. After obtaining pointers ch and iph, Lines 10–12 and Lines 15–18 carry out the second step in the access mechanism: access packet attributes through the structure deﬁned in the protocol speciﬁc headers. In this step, the relevant packet attributes such as unique packet ID, packet type, packet size, source IP address and port, destination IP address and port, are conﬁgured through pointers ch and iph. Note that uidcnt (i.e., uid count) is a static member variable of class Agent which represents the total number of generated packets. We will discuss the details of class Agent later in Chapter 9. Figure 8.6 shows an internal mechanism of function hdr_cmn::access(p) where p is a Packet pointer. When hdr_cmn::access(p) is executed Line 9 in Program 8.6 executes p->access(offset_), where offset_ is the member variable of class hdr_cmn, specifying the location on the packet header allocated to the common header (see also Fig. 8.5). On the right hand side of

180

8 Packets, Packet Headers, and Header Format

Fig. 8.6. The internal mechanism of function access(p) of the hdr cmn struct data type, where p is a pointer to a Packet object.

Fig. 8.6, function access(off) simply returns &bits_[off], where bits_ is the member variable of class Packet storing the packet header. Since hdr_cmn feeds its variable offset_ as the input argument, function access(offset_) essentially returns &bits_[hdr_cmn::offset_], which is the reference to the common header stored in the Packet object *p. This reference is returned as an unsigned char* variable. Then, class hdr_cmn casts the returned reference to type hdr_cmn*, and returns it to the caller. Accessing Packet Attributes in a Protocol Speciﬁc Header After obtaining a reference to a protocol speciﬁc header, the second step is to access the packet attributes according to the structure speciﬁed in the protocol speciﬁc header C++ class. Since NS2 declares a protocol speciﬁc header as a struct data type, it is fairly straightforward to access packet attributes once the reference to the protocol speciﬁc header is obtained (see Example 8.2). 8.3.7 Packet Header Manager A packet header manager is responsible for keeping the list of active protocols and setting the oﬀset values of all the active protocols. It is implemented using a C++ class PacketHeaderManager which is bound to an OTcl class with the same name. Program 8.14 and Fig. 8.7 show the declaration of the C++ class PacketHeaderManager as well as the corresponding binding class, and the diagram of the OTcl class PacketHeaderManager, respectively.

8.3 Packet Header

181

Program 8.14 Declarations of C++ class PacketHeaderManager and mapping class PacketHeaderManagerClass.
1 2 3 4 //~/ns/common/packet.cc class PacketHeaderManager : public TclObject { public: PacketHeaderManager() {bind("hdrlen_", &Packet::hdrlen_);} };

5 static class PacketHeaderManagerClass : public TclClass { 6 public: 7 PacketHeaderManagerClass() : TclClass("PacketHeaderManager") {} 8 TclObject* create(int, const char*const*) { 9 return (new PacketHeaderManager); 10 } 11 } class_packethdr_mgr;

The C++ class PacketHeaderManager has only one constructor (Line 3) and has neither variables nor functions. The constructor binds the instvar hdrlen_ of OTcl class PacketHeaderManager to variable hdrlen_ of class Packet (see also Fig. 8.1). The OTcl class PacketHeaderManager has two main instvars: hdrlen_ and tab_. Instvar hdrlen_ stores the length of packet header. It is initialized to zero in Line 1 of Program 8.15, and is incremented as protocol speciﬁc headers are added to the packet header. Representing the active protocol list, instvar tab_ (Line 2 in Program 8.16) is an associative array whose indexes are protocol speciﬁc header OTcl class names and values
Packet Header Manager tab_ hdrlen_ bind Packet hdrlen_

Offset assignment

...

TCP Header

IP Header

Common Header

...

Fig. 8.7. Architecture of an OTel PacketHeaderManager object.

182

8 Packets, Packet Headers, and Header Format

Program 8.15 Initialization of a PacketHeaderManager object.
1 2 3 4 5 6 7 8 9 //~/tcl/ns-packet.tcl PacketHeaderManager set hdrlen_ 0 foreach prot { Common Flags IP ... } { add-packet-header $prot }

10 proc add-packet-header args { 11 foreach cl $args { 12 PacketHeaderManager set tab_(PacketHeader/$cl) 1 13 } 14 }

Program 8.16 Function create packetformat of class Simulator and function allochdr of class PacketHeaderManager.
//~/tcl/ns-packet.tcl 1 Simulator instproc create_packetformat { } { 2 PacketHeaderManager instvar tab_ 3 set pm [new PacketHeaderManager] 4 foreach cl [PacketHeader info subclass] { 5 if [info exists tab_($cl)] { 6 set off [$pm allochdr $cl] 7 $cl offset $off 8 } 9 } 10 $self set packetManager_ $pm 11 } 12 PacketHeaderManager instproc allochdr cl { 13 set size [$cl set hdrlen_] 14 $self instvar hdrlen_ 15 set NS_ALIGN 8 16 set incr [expr ($size + ($NS_ALIGN-1)) & ~($NS_ALIGN-1)] 17 set base $hdrlen_ 18 incr hdrlen_ $incr 19 return $base 20 }

8.3 Packet Header

183

are 1 if the protocol speciﬁc header is active (see Line 12 in Program 8.5). If the protocol speciﬁc header is inactive, the corresponding value of tab_ will not be available (i.e., NS2 unsets all entries corresponding to inactive protocol speciﬁc headers; see Line 7 in Program 8.20). 8.3.8 Protocol Speciﬁc Header Composition and Packet Header Construction Packet header is constructed through the following three-step process: Step 1: At the Compilation Time During the compilation, NS2 translates all C++ codes into an executable ﬁle. It sets up all necessary variables (including the length of all protocol speciﬁc headers) for all built-in protocol speciﬁc headers, and includes all built-in protocol speciﬁc headers into the active protocol list. There are three main tasks in this step. Task 1: Construct all mapping variables, conﬁgure the variable hdrlen , and register the OTcl class name, and binds the oﬀset value Since all mapping variables are instantiated at the declaration, they are constructed during the compilation using their constructors. As an example, consider the common packet header11 whose construction process shown in Fig. 8.8 proceeds as follows: 1. Store the corresponding OTcl class name (e.g., PacketHeader/Common) in the variable classname_ of class TclClass. 2. Determine the size (using function sizeof (...)) of the protocol speciﬁc header, and store it in the variable hdrlen_ of class PacketHeaderClass. 3. Bind the variable PacketHeader:: offset_ to that of the C++ class hdr_cmn. Task 2: Invocation of function bind() of class TclClass which exports the variable hdrlen The main NS2 function (i.e., main(argc,argv)) invokes function init(...) of class Tcl, which in turn invokes function bind() of class TclClass of all mapping variables. Function bind() registers and binds an OTcl class name to the C++ domain (see ﬁle ˜tclcl /Tcl.cc). This function is overridden by class PacketHeaderClass.
11

NS2 repeats the following process for all protocol speciﬁc headers. For brevity, we show the construction process through common packet header only.

184

8 Packets, Packet Headers, and Header Format
CommonHeaderClass class_cmnhdr;

PacketHeaderClass ::PacketHeaderClass (classname,hdrlen)

CommonHeaderClass ::CommonHeaderClass() PacketHeaderClass(

PacketHeader/Common (OTcl Class)

“PacketHeader/Common”, TclClass(classname)

hdr_cmn (C++ Class) offset_

hdrlen_(hdrlen) offset_(0)

sizeof(hdr_cmn)

)

PacketHeaderClass::bind_offset(off) offset_ = off; bind_offset(&hdr_cmn::offset_)

return

Fig. 8.8. Construction of the static mapping variable class cmnhdr.

As shown in Lines 12–17 of Program 8.10, class PacketHeaderClass overrides function bind() of class TclClass. Line 13 ﬁrst invokes the function bind() of class TclClass. Line 15 exports the variable hdrlen_ to the OTcl instvar with the same name. Finally, Line 16 registers the OTcl method offset. In case of class CommonHeaderClass, classname_ is PacketHeader/Common and hdrlen_ is 104 bytes. Therefore, Line 15 of Program 8.10 executes the following command in the OTcl domain: PacketHeader/Common set hdrlen_ 104 which sets instvar hdrlen_ of class PacketHeader/Common to be 104. Note that this instvar hdrlen_ is not bound to the C++ domain. After Task 1 and Task 2 are completed, the related protocol speciﬁc classes, namely, hdr_cmn, PacketHeader/Common, and CommonHeaderClass, would be as shown in Fig. 8.9. The mapping object class_cmnhdr is of class CommonHeaderClass, which derives from classes PacketHeaderClass and TclClass, respectively. It inherits variables classname_, hdrlen_, and offset_ from its parent classes. After object construction is complete, variable classname_ will store the name of the OTcl common header class (i.e., PacketHeader/Common), hdr_len_ will store the amount of memory in bytes needed to store common header, and offset_ will point to hdr_cmn::offset_. Here, variable offset_ of class CommonHeaderClass only points to variable offset_ of class hdr_cmn. However, at this moment, the oﬀset value is set to zero. The dashed arrow in Fig. 8.9 indicates that the value of variable hdr_cmn::offset_ will be later set to store an oﬀset from the beginning of a packet header to the point where the common packet header is stored. Also, after function Tcl::init() invokes function bind() of class

8.3 Packet Header

185

Fig. 8.9. A schematic diagram of a static mapping object class cmnhdr, class hdr cmn, class PacketHeader/Common, and class Packet.

PacketHeaderClass, instvar hdrlen_ of class PacketHeader/Common will store the value of variable hdrlen_ of class CommonHeaderClass. Note that tasks and 2 only set up C++ OTcl class, and mapping class. However, the packet header manager is not conﬁgured at this phase. Task 3: Sourcing the ﬁle ˜ ns/tcl/lib/ns-packet.tcl to setup an active protocol list As discussed in Section 3.7, NS2 sources all scripting Tcl ﬁles during the compilation process. In regards to packet header, Program 8.15 shows a part of the ﬁle ˜ns/tcl/lib/ns-packet. Here, Line 8 invokes procedure add-packetheader{prot} for all built-in protocol speciﬁc headers indicated in Lines 3–6. In Line 12, this procedure sets the value of the associated array tab_ whose index is the input protocol speciﬁc header name to be 1. Step 2: During the Network Conﬁguration Phase In regards to packet header construction, the main task in the Network Conﬁguration Phase is to setup variables offset_ of all active protocol speciﬁc headers and formulate a packet header format. Subsequent packet creation will follow the packet format created in this step.

186

8 Packets, Packet Headers, and Header Format

The oﬀset conﬁguration process takes place during the simulator construction. From Line 2 of Program 4.11, the constructor of the simulator invokes instproc create_packetformat{} of class Simulator shown in Program 8.16. Instproc create_packetformat{} creates a PacketHeaderManager object pm (Line 3), computes the oﬀset value of all active protocol speciﬁc headers using instproc allochdr{cl} (Line 6), and conﬁgures the oﬀset values of all protocol speciﬁc headers (Line 7). The foreach loop in Line 4 runs for all built-in protocol speciﬁc headers which are child classes of class PacketHeader. Then Line 5 ﬁlters out those which are not in the active protocol list (see Section 8.3.7). Lines 6–7 are executed for all active protocol speciﬁc headers speciﬁed in variable tab_ (which was conﬁgured in Step 1 – Task 3) of the PacketHeaderManager object “pm”. Line 7 conﬁgures oﬀset values by using the OTcl method offset (see Program 8.11) of protocol speciﬁc header mapping classes. The OTcl method offset stores the input argument in variable *offset_ of the protocol speciﬁc header mapping class (e.g., CommonHeaderClass). Lines 12–19 in Program 8.16 and Fig. 8.10 show the OTcl source codes and the diagram, respectively, of the instproc allochdr{cl} of an OTcl class PacketHeaderManager. Instproc allochdr{cl} takes one input argument “cl” (in Line 12) which is the name of a protocol speciﬁc header, and returns the oﬀset value corresponding to the input argument “cl”. Line 13 stores header length of a protocol speciﬁc header “cl”(e.g., variable hdrlen_

Fig. 8.10. A diagram representing instproc allochdr of class PacketHeaderManager. Line numbers shown on the left correspond to the lines in Program 8.16. The action corresponding to each line is shown on the right.

8.4 Data Payload

187

of class PacketHeader/Common) in a local variable size.12 Based on size, Lines 15–16 compute the amount of memory “incr” needed to store the header based on size.13 Line 17 stores the current packet header length (excluding the input protocol speciﬁc header) in a local variable “base”. Since “base” is an oﬀset distance from the beginning of packet header to the input protocol speciﬁc header, it is returned to the caller as the oﬀset value in Line 19. Finally, Line 18 increases the header length (i.e., the instvar hdrlen_ of class PacketHeaderManager) by “incr”. During the Simulator construction, the packet header manager also updates its variable hdrlen_ (Line 19 in Program 8.16). Note that the instvar hdrlen_ of class PacketHeaderManager was set to zero at the compilation (Line 1 in Program 8.15). As Lines 6–7 in Program 8.16 repeat for every protocol speciﬁc header, the oﬀset value is added to the instvar hdrlen_ of an OTcl class PacketHeaderManager. At the end, the instvar hdrlen_ will represent the total header length, which embraces all protocol speciﬁc headers. Step 3: During the Simulation Phase During the Simulation Phase, NS2 creates packets based on the format deﬁned in the former two steps. For example, an Agent object creates and initializes a packet using its function allocpkt(). Here, a packet is created using function alloc() of class Packet, and initialized using function initpkt(p) of class Agent. Again, function alloc() takes a packet from the free packet list, if it is non-empty. Otherwise, it will create a new packet using new. After retrieving a packet, it clears the values stored in the packet header and data payload. Function initpkt(p) assigns default values to packet attributes such as packet unique ID, packet type, and packet size (see Program 8.13). The initialization is performed by retrieving a reference (e.g., ch) to the relevant protocol speciﬁc header and accessing packet attributes using the predeﬁned structure.

8.4 Data Payload
Implementation of data payload in NS2 diﬀers from actual data payload. In practice, user information is transformed into bits, and are stored into data payload. Such a transformation is not necessary in simulation, since NS2 stores the user information in the packet header. NS2 rarely needs to maintain data payload. In Line 11 of Program 7.3, packet transmission time, i.e., the time packet size . Class LinkDelay required to send out a packet, is computed as bandwidth
12

13

Variable hdrlen of a protocol speciﬁc header OTcl class was conﬁgured in Step 1 – Task 2. Variable “incr” could be greater than size, depending on the underlying hardware.

188

8 Packets, Packet Headers, and Header Format

determines the size of a packet by hdr_cmn::size_ (not by counting the number of bits stored in packet header and data payload) to compute packet transmission time. In most cases, users do not need to explicity deal with data payload. Program 8.17 Declaration of enum AppDataType and class AppData.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 //~/ns/common/ns-process.h enum AppDataType { ... PACKET_DATA, HTTP_DATA, ... ADU_LAST }; class AppData { private: AppDataType type_; // ADU type public: AppData(AppDataType type) { type_ = type; } AppData(AppData& d) { type_ = d.type_; } virtual ~AppData() {} AppDataType type() const { return type_; } virtual int size() const { return sizeof(AppData); } virtual AppData* copy() = 0; };

NS2 also provides a support to hold data payload. In Line 4 of Program 8.1, class Packet provides a pointer data_ to an AppData object.14 Program 8.17 shows the declaration of an abstract class AppData. Class AppData has only one member variable type_ in Line 11. Among its functions, and one is a pure virtual function copy() shown in Line 18. Indicating the type of application, variable type_ is of type enum AppDataType deﬁned in Lines 1–8. Function copy() duplicates an AppData object to a new AppData object. It is a pure virtual function, and must be overridden by child instantiable classes of class AppData. Function size() in Line 17 returns the amount of memory required to store an AppData object. Class AppData provides two constructors. One is in Line 13, where the caller feeds an AppData type as an input argument. Another is in Line 14, where a reference to a AppData object is fed as an input argument. In both the cases, the constructor simply sets the variable type_ to a value as speciﬁed in the input argument.
14

However, no memory is allocated to the AppData object unless it is needed.

8.4 Data Payload

189

Program 8.18 Declaration of class PacketData.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 //~/ns/common/packet.h class PacketData : public AppData { private: unsigned char* data_; int datalen_; public: PacketData(int sz) : AppData(PACKET_DATA) { datalen_ = sz; if (datalen_ > 0) data_ = new unsigned char[datalen_]; else data_ = NULL; } PacketData(PacketData& d) : AppData(d) { datalen_ = d.datalen_; if (datalen_ > 0) { data_ = new unsigned char[datalen_]; memcpy(data_, d.data_, datalen_); } else data_ = NULL; } virtual ~PacketData() { if (data_ != NULL) delete []data_; } unsigned char* data() { return data_; } virtual int size() const { return datalen_; } virtual AppData* copy() { return new PacketData(*this); } };

Program 8.18 shows the declaration of class PacketData, a child class of class AppData. This class has two new member variables: data_ (a string variable which stores data payload) in Line 3 and datalen_ (the length of data_) in Line 4. Line 25 deﬁnes a function data() which simply returns data_. Lines 26 and 27 override the virtual functions size() and copy(), respectively, of class AppData. Function size() simply returns datalen_, while function copy() creates a new PacketData object which has the same content as the current PacketData object, and returns the pointer to the created object to the caller. Class PacketData has two constructors. One is to construct a new object with size “sz”, using the constructor in Lines 6–12. This constructor simply sets the default application data type to be PACKET_DATA (Line 6), stores “sz” in datalen_ (Line 7), and allocates memory of size datalen_ to data_ (Line

190

8 Packets, Packet Headers, and Header Format

9). Another construction method15 is to create a copy of an input PacketData object (Lines 13–20). In this case, the constructor feeds an input PacketData object “d” to the parent class (Line 13), copies the variable datalen_ (Line 14), and duplicates its data payload (Line 17).16 NS2 creates a PacketData object through two functions of class Packet: alloc(n) and allocdata(n). In Program 8.4, function alloc(n) allocates a packet in Line 3, and creates data payload using function allocdata(n) in Line 5. Function allocdata(n) creates a PacketData object of size “n”, by executing new PacketData(n) in Line 11. Program 8.19 Functions accessdata, userdata, setdata and datalen of class Packet.
//~/ns/common/packet.h 1 class Packet : public Event { 2 ... 3 public: 4 inline unsigned char* accessdata() const { 5 if (data_ == 0) 6 return 0; 7 assert(data_->type() == PACKET_DATA); 8 return (((PacketData*)data_)->data()); 9 } 10 inline AppData* userdata() const {return data_;} 11 inline void setdata(AppData* d) { 12 if (data_ != NULL) 13 delete data_; 14 data_ = d; 15 } 16 inline int datalen() const { return data_ ? data_->size() : 0; } 17 ... 18 };

Program 8.19 shows four functions which can be used to manipulate data payload. Functions accessdata() (Lines 4–9) and userdata() (Line 10) are both data payload access functions. The diﬀerence is that accessdata() returns a direct pointer to a string data_ which contains data payload while userdata() returns a pointer to an AppData object which contains data payload. Function setdata(d) (Lines 11–15) sets the pointer data_ to point to the input argument “d”. Finally, function datalen() in Line 16 returns the size of data payload.
15

16

Function copy() in Line 27 employs this constructor to create a copy of a PacketData object. Function memcpy(dst,src,num)) copies “num” data bytes from the location pointed by “src” to the memory block pointed by “dst”.

8.5 Customizing Packets

191

8.5 Customizing Packets
8.5.1 Creating Your Own Packet When designing a new protocol, a user may need to change the packet format. This section gives a guideline of how packet header, data payload, or both can be modiﬁed. Note that, it is recommended not to use data payload in simulation. If possible, include information related to the new protocol in a protocol speciﬁc header. Deﬁning a New Packet Header Suppose we would like to include a new protocol speciﬁc header, namely “My Header”, into the packet header. We need to deﬁne a C++ class (e.g., hdr_myhdr), an OTcl class (e.g., PacketHeader/MyHeader), and a mapping class (e.g., MyHeaderClass) for My Header, and include the OTcl class into the active protocol list. In particular, we need to perform the following four steps: • Step 1: Deﬁne a protocol speciﬁc header C++ struct hdr_myhdr (e.g., see Program 8.6). – Declare variable offset_. – Deﬁne function access(p) (see Lines 8–10 in Program 8.6). – Include all member variables required to hold new packet attributes. – [Optional] Include a new packet type into enum packet_t and class p_info (e.g., see Program 8.9). Again, a new packet type does not need to be added for every new protocol speciﬁc header. • Step 2: Deﬁne a protocol speciﬁc header OTcl class PacketHeader/MyHe ader. • Step 3: Derive a mapping class MyHeaderClass from class PacketHeader Class (e.g, see class CommonHeaderClass in Program 8.12). – At the construction, feed the corresponding OTcl class name (i.e., PacketHeader/MyHeader) and the size needed to hold the protocol speciﬁc header (i.e., sizeof(hdr_myhdr)) to the constructor of class PacketHeaderClass (e.g., see Line 3 in Program 8.12). – From within the constructor, invoke function bind_offset(...) feeding the address of the variable offset_ of the C++ struct data type as an input argument. (i.e., invoke bind_offset(&hdr_myhdr::offset_)). – Instantiate a mapping variable class_myhdr at the declaration. • Step 4: Activate My Header by including class PacketHeader/MyHeader into the active protocol list. The simplest way is to modify Lines 2–9 of Program 8.15 as follows: foreach prot { Common Flags

192

8 Packets, Packet Headers, and Header Format

... MyHeader } { add-packet-header $prot } where only the suﬃx of the new protocol speciﬁc header (i.e., MyHeader) is added to the foreach loop. Deﬁning a New Data Payload Data payload can be created in four levels: (i) None: NS2 rarely uses data payload in simulation. To avoid any complicacy, it is suggested not to use data payload in simulation. (ii) Use class PacketData: The simplest form of storing data payload is to use class PacketData (see Program 8.18). Class Packet has a variable data_ whose class is PacketData and provides functions (in Program 8.19) to manipulate the variable data_. (iii) Derive a class (e.g., class MyPacketData) from class PacketData: This option is suitable when new functionalities (i.e., functions and variables) in addition to those provided by class PacketData are needed. After deriving a new PacketData class, a user may also derive a new class (e.g., class MyPacket) from class Packet, and override the variable data_ of class Packet to be a pointer to a MyPacketData object. (iv) Deﬁne a new data payload class: A user can also deﬁne a new payload type if needed. This option should be used when the new payload has nothing in common with class PacketData. The followings are the main tasks needed to deﬁne and use a new payload type MY_DATA. • Include the new payload type (e.g., MY_DATA) into enum AppDataType data type (see Program 8.17). • Derive a new payload class MyData from class AppData. – Feed the payload type MY_DATA to the constructor of class AppData. – Include any other necessary functions and variables. – Override functions size() and copy(). • Derive a new class MyPacket from class Packet – Declare a variable of class MyData to store data payload. – Include functions to manipulate the above MyData variable. 8.5.2 Activate/Deactivate a Protocol Speciﬁc Header By default, NS2 includes all built-in protocol speciﬁc headers into packet header (see Program 8.15). This inclusion can lead to unnecessary wastage of memory especially in a packet-intensive simulation, where numerous packets are created. For example, common, IP, and TCP headers together use only 0.1 kB, while the default packet header consumes as much as 1.9 kB [15].

8.5 Customizing Packets

193

Again, NS2 does not return the memory allocated to a Packet object until the simulation terminates. Selectively including protocol speciﬁc header can lead to huge memory saving. The packet format is deﬁned when the Simulator is created. Therefore, a protocol speciﬁc headers must be activated/deactivated prior to the creation of the Simulator. NS2 provides the following OTcl procedures to activate/ deactivate protocol speciﬁc headers: • To add a protocol speciﬁc header PacketHeader/MH1, execute add-packet-header MH1 In Lines 10–14 of Program 8.15, the above statement includes PacketHeader/MH1 to the variable tab_ of class PacketHeaderManager. • To remove a protocol speciﬁc header PacketHeader/MH1 from the active list, execute remove-packet-header MH1 The details of procedure remove-packet-header{args} are shown in Lines 1–9 of Program 8.20. From Line 7, the above statement removes the entries with index PacketHeader/MH1 from the variable tab_ of class PacketHeaderManager. Program 8.20 Procedures remove-all-packet-header.
1 2 3 4 5 6 7 8 9

remove-packet-header,

and

//~/tcl/ns-packet.tcl proc remove-packet-header args { foreach cl $args { if { $cl == "Common" } { warn "Cannot exclude common packet header." continue } PacketHeaderManager unset tab_(PacketHeader/$cl) } }

10 proc remove-all-packet-headers {} { 11 PacketHeaderManager instvar tab_ 12 foreach cl [PacketHeader info subclass] { 13 if { $cl != "PacketHeader/Common" } { 14 if [info exists tab_($cl)] { 15 PacketHeaderManager unset tab_($cl) 16 } 17 } 18 } 19 }

194

8 Packets, Packet Headers, and Header Format

• To remove all protocol speciﬁc headers, execute remove-all-packet-header In Lines 10–19 of Program 8.20, the procedure remove-all-packetheader{} uses foreach to remove all protocol speciﬁc headers (except for common header) from the active protocol list (i.e., the instvar tab_).

8.6 Chapter Summary
Consisting of packet header and data payload, a packet is represented by a C++ class Packet. Class Packet consists of pointers bits_ to its packet header and data_ to its data payload. It employs a pointer next_ to form a linked list of packets. It also has a pointer free_ which points to the ﬁrst Packet object on the free packet list. When a Packet object is no longer in use, NS2 stores the Packet object in the free packet list for future reuse. Again, Packet objects are not destroyed until the simulation terminates. When allocating a packet, NS2 ﬁrst tries to take a Packet object from the free packet list. Only when the free packet list is empty, will NS2 create a new Packet object. During simulation, NS2 usually stores relevant user information (e.g., packet size) in packet header, and rarely uses data payload. It is recommended not to use data payload if possible, since storing all information in packet header greatly simpliﬁes the simulation yet yields the same simulation results. Packet header consists of several protocol speciﬁc headers. Each protocol speciﬁc header occupies a contiguous part in packet header, and identiﬁes the occupied location by using its variable offset_. NS2 employs a packet header manager (represented by an OTcl class PacketHeaderManager) to maintain a list of active protocols, and deﬁne packet header format using the list when the Simulator is created. The packet header construction process proceeds in the three following steps: (i) At the Compilation: NS2 deﬁnes the following three classes for each of protocol speciﬁc headers: • A C++ class: NS2 uses C++ struct data type to deﬁne how packet attributes are stored in a protocol speciﬁc header. One of the important member variables is offset_, which indicates the location of the protocol speciﬁc header on the packet header. • An OTcl class: During the Network Conﬁguration Phase, the packet header manager conﬁgures packet header from the OTcl domain. It accesses a protocol speciﬁc header from the OTcl class which acts as an interface from the OTcl to the C++ domains. • A mapping class: A mapping class binds the OTcl and C++ class together. It declares a method offset, which is invoked by a packet header manager from the OTcl domain to conﬁgure the value of variable offset_ of the C++ class PacketHeaderClass.

8.6 Chapter Summary

195

In this step, NS2 also stores all built-in protocol speciﬁc headers in instvar tab_ of class PacketHeaderManager, which represents the active protocol list. (ii) At the Network Conﬁguration Phase: A user may add/remove protocol speciﬁc headers to/from the active protocol list. When the Simulator is created, the packet header manager computes and assigns appropriate oﬀset values to all protocol speciﬁc headers speciﬁed in the active list. (iii) At the Simulation Phase: NS2 follows the above packet header deﬁnitions when allocating a packet.

9 Transport Control Protocols Part 1 – An Overview and User Datagram Protocol implementation

A typical communication system consists of applications, transport layer agents, and a low level network. An application models user demand to transmit data. Taking user demand as an input, a sending transport layer agent creates packets and forwards them to the associated receiving transport layer agent through a low-level network. Having discussed the details of low level networks in Chapters 5–7, the details of transport layer agents are presented here in Chapters 9–10. Also, the details of applications will be presented in Chapter 11. This chapter provides an overview of transport layer agents, and shows NS2 implementation of User Datagram Protocol (UDP) agents. In particular, Section 9.1 introduces two most widely used transport control protocols: Transmission Control Protocol (TCP) and User Datagram Protocol (UDP). Section 9.2 explains NS2 implementation of basic agents. Section 9.3 shows the implementation of UDP agents and Null agents. Finally, the chapter summary is given in Section 9.4.

9.1 UDP and TCP Basics
9.1.1 UDP Basics Deﬁned in [18] and [19], User Datagram Protocol (UDP) is a connectionless transport-layer protocol, where no connection setup is needed prior to data transfer. UDP oﬀers minimal transport layer functionalities – non-guaranteed data delivery – and gives applications a direct access to the network layer. Aside from the multiplexing/demultiplexing functions and some light error checking, it adds nothing to IP packets. In fact, if the application developer employs UDP as a transport layer protocol, then the application is communicating almost directly with the network layer.

T. Issariyakul, E. Hossain, Introduction to Network Simulator NS2, DOI: 10.1007/978-0-387-71760-9 9, c Springer Science+Business Media, LLC 2009

198

9 Transport Control Protocols Part 1– An Overview

UDP takes messages from an application process, attaches source and destination port for the multiplexing/demultiplexing service, adds two other ﬁelds of error checking and length information, and passes the resulting packet to the network layer [19]. The network layer encapsulates the UDP packet into a network layer packet and then delivers the encapsulated packet at the receiving host. When a UDP packet arrives at the receiving host, it is delivered to the receiving UDP agent identiﬁed by the destination port ﬁeld in the packet header. 9.1.2 TCP Basics As shown in Fig. 9.1, Transmission Control Protocol (TCP) [20] is a connectionoriented reliable transport protocol consisting of three phases of operations: connection setup, data transfer, and connection termination. In the connection setup phase, a TCP sender initiates a three-way handshake (i.e., sending SYN, SYN-ACK, and ACK messages). After a connection is established, TCP enters the data transfer phase where a TCP sender transfer data to a TCP receiver. Finally, after the data transfer is complete, TCP tears down the connection in the connection termination phase by using a four-way handshake (i.e., sending two pairs of FIN-ACK messages.)

SYN

SYN-ACK CONNECTION SETUP PHASE ACK

…
FIN ACK FIN ACK

DATA TRANSFER PHASE

CONNECTION TERMINATION PHASE

Fig. 9.1. Main phases of TCP operation: Connection setup, data transfer, and connection termination.

9.1 UDP and TCP Basics

199

The main operation of TCP lies in the data transfer phase, which implements two following mechanisms: (1) Error control using basic acknowledgement and timeout, and (2) Congestion control using a window-based mechanism. Error Control Using Basic Acknowledgement and Timeout As a reliable transport layer protocol, TCP provides connection reliability by means of acknowledgement (ACK). For every received packet, a TCP receiver returns an ACK packet to the sender. If an ACK packet is not received within a given timeout value, the TCP sender will assume that the packet is lost, and will retransmit the lost packet. Note that in the literature, a timeout period is also referred to as Retransmission TimeOut (RTO). Hereafter, we will refer to these two terms interchangably. TCP employs a cumulative acknowledgement mechanism. With this mechanism, a TCP receiver always acknowledges to the sender with the highest sequence number up to which all packets have been successfully received. For example, in Fig. 9.2, packet 3 is lost. Therefore, the TCP receiver returns ACK for packet 2 (A2) even when packets 4, 5, and 6 have been received. These ACK packets (e.g., A2), which acknowledge the same TCP packet (e.g., packet 2), are referred to as the duplicated acknowledgement packets. From Fig. 9.2, the TCP sender does not receive an ACK packet which acknowledges packet 3. After a period of RTO, the sender will assume that packet 3 was lost and will retransmit packet 3.
RTO TCP Sender 1 2 3 4 5 6 3

TCP Receiver

A1

A2

A2

A2

A2

Fig. 9.2. An example of TCP error control using acknowledgement: A TCP sender realizes the loss of TCP packet number 3 after transmitting the packet number 3 for a period of RTO (ie., timeout).

The RTO value is optimized according to the following tradeoﬀ: a small RTO value leads to unnecessary packet retransmission while a large RTO value results in high latency of packet loss detection. In general, an RTO value should be a function of network Round-Trip Time (RTT), which is the time required for a data bit to travel from a source node to the destination node and travel back to the source node. Due to network dynamics, RTT of one packet could be diﬀerent from that of another. In TCP, smoothed (i.e., average) RTT (t) and RTT variation (σt ) are computed based on the collected RTT samples, and are used to compute the RTO value. According to [21], instantaneous smoothed RTT, RTT variation, and instantaneous RTO are computed as follows. Let t(k) be the k th RTT sample

200

9 Transport Control Protocols Part 1– An Overview

collected upon ACK reception. Also, let t(k), σt (k), and RT O(k) be the values of t, σt , and RTO, respectively, when the k th RTT sample is determined. Then, t(k + 1) = α × t(k) + (1 − α) × t(k + 1), σt (k + 1) = β × σt (k) + (1 − β) × |t(k + 1) − t(k + 1)|, (9.1) (9.2)

RT O(k + 1) = min{ub, max{lb, γ × [t(k + 1) + 4 × σt (k + 1)]}} (9.3) where ub and lb are ﬁxed upper and lower bounds on the RTO value. The constants α ∈ (0, 1) and β ∈ (0, 1) are usually set to 7/8 and 3/4, respectively. The variable γ is a binary exponential backoﬀ (BEB) factor. It is initialized to 1, and doubled for every timeout event, and is reset to 1 when a new ACK packet arrives. Window-Based Congestion Control A transport layer protocol is also responsible for network congestion. It limits the transmission rate of a data ﬂow to help control network congestion. As a window-based congestion control protocol, TCP limits the transmission rate by adjusting the congestion window (i.e., transmission window) which basically refers to the amount of data that a sender can transmit without waiting for acknowledgement. For example, the congestion window size of the TCP connection in Fig. 9.2 is initialized to 4. Therefore, the TCP sender pauses after sending packets 1–4. After receiving ACK corresponding to packet 1 (i.e., A1), the number of unacknowledged packets becomes 3 and TCP is able to send out packet 5. Congestion window refers to a range of sequence numbers of TCP packets which can be transmitted at a moment. For example, the congestion window at the beginning of Fig. 9.2 is {1, 2, 3, 4} and the congestion window size is 4. When A1 is received, the congestion window becomes {2, 3, 4, 5}. In this case, we say that the congestion window slides to the right. Suppose that the congestion window changes to {2, 3, 4, 5, 6} (the size is 5). In this case, we say that the congestion window is opened by one unit. On the contrary, if the window becomes {2, 3, 4}, we say the congestion window is closed by one unit. Again, a larger window size allows the sender to transmit more data in a given interval implying a higher transmission rate at the transport layer. TCP increases and decreases its transmission rate by opening and closing its congestion window. Window Increasing Mechanism One of the key features of TCP is network-based rate adaptability. TCP slowly opens its congestion window to ﬁll up the underlying network, when the network is underutilized. When the network is overutilized, TCP rapidly closes the congestion window to help relieve the congestion. TCP window opening

9.1 UDP and TCP Basics

201

mechanism consists of two phases, each of which is identiﬁed by the current congestion window size (w) and a slow-start threshold (wth ): (i) Slow-start phase: If w < wth , TCP increases w by one for every received ACK packet. 1 (ii) Congestion avoidance phase: If w ≥ wth , TCP increases w by w(t) for every received ACK packet. Note that, TCP receiver may advertise its maximum window size (wmax ) which does not ﬁll its buﬀer too rapidly. This wmax acts as an upper-bound for the above window increasing mechanism. In NS2, congestion window (ω) evolves according to the above two phases, regardless of ωmax . However, TCP uses the minimum of ω and ωmax to determine amount of data it can transmit. Packet Loss Detection Mechanism In the literature, various TCP variants use diﬀerent combinations of the following packet loss detection mechanisms: • Timeout: As discussed earlier, TCP starts its retransmission timer for every transmitted packet, and assumes a packet loss upon timer expiration. • Fast Retransmit: By default, an RTO has granularity of 0.5 seconds, which could lead to large latency in packet loss detection. Fast Retransmit expedites the packet loss detection by means of duplicated acknowledgement detection. Upon detection of the k th (which is equal to 3 by default) duplicated acknowledgement (excluding the ﬁrst one which is a new acknowledgement), the TCP sender stops waiting for the timeout, assumes a packet loss, and retransmits the lost packet. From Fig. 9.2, if the fast retransmit mechanism is used, the TCP sender will assume that packet 3 is lost and retransmits packet 3 upon receiving the 4th A2 packet (i.e., the 3rd duplicated acknowledgement). Note that based on the cumulative acknowledgement principle, upon receiving the retransmitted packet 3, TCP receiver sends A6 back to the sender, since packets 4, 5, and 6 have been successfully received earlier. Window Decreasing Mechanism Originally conceived to combat congestion in a wired network, TCP assumes that all packet losses occurs due to congestion (i.e., buﬀer overﬂow at the routers in the network). It reacts to every packet loss by reducing its transmission rate (or window size) to lessen the congestion. The following approaches are among the most popular window decreasing mechanisms for a TCP variant used in the literature. • Reset to 1: Conventionally, TCP reacts to packet loss by resetting the window size to 1, and setting the slow-start threshold to half of the current congestion window size. However, this option is usually deemed too radical and could lead to TCP throughput degradation in presence of random packet loss.

202

9 Transport Control Protocols Part 1– An Overview

• Fast Recovery: Upon detection of a packet loss, the fast recovery mechanism sets both current window size and slow-start threshold to half of the current congestion window size. Then, it increases the window size by one for each incoming duplicated acknowledgement. At this moment, the sender may transmit a new packet if the congestion window allows. Upon receiving a new acknowledgement, the sender exits Fast Recovery and sets the window size to the slow-start threshold value, after which TCP operates normally in a congestion avoidance phase. TCP Variants There are numerous TCP variants in the literature. This section discusses only de facto TCP variants namely Old Tahoe, Tahoe, Reno, and new Reno. These TCP variants utilize the same window increasing mechanism (i.e., slow start and congestion avoidance). However, they diﬀer in how they detect a packet loss and decrease the window size. Table 9.1 shows the diﬀerences in window size adjustment mechanism, when packet loss is detected through timeout and Fast Retransmit (i.e., duplicated ACKs).
Table 9.1. Diﬀerences among basic TCP variants: Diﬀerent window closing mechanisms upon detection of a packet loss. TCP Variant Timeout Old-Tahoe Reset w to Tahoe Reset w to Reno Reset w to New Reno Reset w to 1 1 1 1 Loss Detection Fast Retransmit N/A Reset w to 1 Fast Recovery (single packet) Fast Recovery (all packets)

The very ﬁrst TCP variant, Old-Tahoe, detects packet loss through timeout only. When packet loss is detected it always resets congestion window size to 1. Developed from Old-Tahoe, TCP Tahoe uses the Fast Retransmit mechanism to expedite packet loss detection rather than waiting for the timeout. It always sets the window size to 1 upon detection of a packet loss. Both TCP Reno and New-Reno reset the window size to 1, when a packet loss is detected through timeout. However, they will employ Fast Recovery if packet loss is detected through Fast Retransmit. The diﬀerence between TCP Reno and TCP New- Reno is that TCP Reno exits the fast recovery process as soon as the lost packet which triggered Fast Retransmit is acknowledged. If there are multiple packet losses within a congestion window, Fast Recovery could be invoked for several times, and the window size will decrease signiﬁcantly. To avoid the multiple window closures, TCP New-Reno stays in the Fast Recovery phase until all packets in the loss window are acknowledged or until timeout occurs.

9.2 Basic Agents

203

9.2 Basic Agents
An agent is an NsObject which is responsible for creating and destroying packets. There are two main types of NS2 agents: routing agents and transportlayer agents. A routing agent creates and receives routing control packets, and commands routing protocols to act accordingly. Connecting an application to a low level network, a transport-layer agent controls the congestion and reliability of a data ﬂow based on an underlying transport layer protocol (e.g., UDP or TCP). This book focuses on transport layer agents only Program 9.1 Class AgentClass which binds OTcl and C++ class Agent.
1 2 3 4 5 6 7 //~/ns/common/agent.cc static class AgentClass : public TclClass { public: AgentClass() : TclClass("Agent") {} TclObject* create(int, const char*const*) { return (new Agent(PT_NTYPE)); } } class_agent;

NS2 implements agents in a C++ class Agent, which is bound to an OTcl class with the same name (see Program 9.1). In the following, we ﬁrst discuss the relationship among a transport-layer agent, an application, and a lowlevel network in Section 9.2.1. Agent conﬁguration and internal mechanisms are discussed in Sections 9.2.2 and 9.2.3, respectively. Finally, Section 9.2.4 provides guidelines to deﬁne a new transport-layer agent. 9.2.1 Applications, Agents, and a Low-level Network An agent acts as a bridge which connects an application and a low-level network. Based on the user demand provided by an application, a sending agent constructs packets and transmits them to a receiving agent through a low-level network. Figure 9.3 shows an example of such a connection. Consider Fig. 9.3. On the top level, a CBR (constant bit rate) application, which models a user demand to periodically transmit data, is used as an application. The demand is passed to a UDP sending agent, which in turn creates UDP packets. Here, the UDP agent stores source and destination IP addresses and transport layer ports in the packet header, and forwards the packet to the attached node (e.g., Node 1 in Figure 9.3). Using the pre-calculated routing table, the low-level network delivers the packet to the destination node (e.g., Node 3 in Fig. 9.3) speciﬁed in the packet header. The destination node employs its demultiplexer to forward the packet to the agent attached to the port speciﬁed in the packet header. Finally, a Null receiving agent simply destroys the received packet.

204

9 Transport Control Protocols Part 1– An Overview
CBR
agent_

attach-agent

Application: User demand indication Null Agent (Receiving) connect UDP Agent (Sending) app_ target_

0

app_ target_

Agents: Packet construction and destruction

A Network of NsObjects

attach-agent

attach-agent

Node 1 dmux_ classifier_ Link 2

Node 3 dmux_ classifier_

Low-Level Network: Packet forwarding

Link 1

Low-Level Network

Link 3

Node 2

Fig. 9.3. A CBR application over UDP conﬁguration.

From the above discussion, an agent can be used as a sending agent (e.g., a UDP agent) or a receiving agent (e.g., a Null agent). A sending agent has connections to both an application and a low-level network, while a receiving agent may not have a connection to an application (because it does not need any). An application (e.g., CBR) uses its variable agent_ as a reference to an agent (e.g., UDP and Null agents), while an agent uses its variable app_ as a reference to an application. It is mandatory to conﬁgure the variables agent_ and app_ (i.e., create the connection) for a sending agent, while it is optional for a receiving agent. This is mainly because the application needs to inform the agent of user demand (i.e., by invoking function sendmsg(...)), and the sending agent needs to inform the application of the completion of data transmission (i.e., by invoking function resume()). Since a receiving agent simply destroys the received packet, it does not need a connection to an application. Both sending and receiving agents connect to a low-level network in the same manner. They use a pointer target_, to point to the attached node. The Node, on the other hand, installs the agent slot number “port” of its demultiplexer dmux_ (which is of class PortClassifier), where “port” is the corresponding port number of the agent (see Section 6.6.3).

9.2 Basic Agents Table 9.2. Key diﬀerences between a sending and a receiving agent. Sending agent Upstream - object Application - packet forwarding function sendmsg Downstream object - object Node - packet forwarding function recv Receiving agent Node recv N/A N/A

205

Table 9.2 shows the key diﬀerences between a sending agent and a receiving agent. The upstream object of a sending agent is an application, which informs a sending agent of incoming user demand through function sendmsg(...) of the sending agent. The upstream object of a receiving agent, on the other hand, is a Node object, which forwards packets to the receiving agent via function recv(p,h). The downstream object of a sending agent is a Node object. The sending agent passes a packet *p to a Node object by executing target_->recv(p,h). A receiving agent usually has no downstream object, since it simply destroys the received packets. 9.2.2 Agent Conﬁguration From Fig. 9.3, agent conﬁguration consists of four main steps: (i) Create a sending agent, a receiving agent, and an application using “new{...}”. (ii) Attach agents to the application using OTcl Command attach-agent{agent} of class Application. (iii) Attach agents to the a low-level network using instproc attach-agent{node agent} of class Simulator. (iv) Associate the sending agent with the receiving agent using instproc connect{src dst} of class Simulator. Example 9.1 (A Network Construction Example). The example network in Fig. 9.3 employs CBR, a UDP agent, and a Null agent as an application, a sending agent, and a receiving agent, respectively. To setup the example network, we may use the Tcl simulation script in Program 9.2. While Lines 1–7 create a low-level network (see the details in Chapters 6 and 7), Lines 8–14 set up a CBR application, a UDP agent, and a Null agent on top of the low-level network. Again, there are 4 major steps to create connections among agent, an application, a low-level network: (i) Create agent and application objects (Lines 8–10). (ii) Use command attach-agent of class Application to create a connection between an application and a sending agent (Line 11).

206

9 Transport Control Protocols Part 1– An Overview

Program 9.2 A simulation script which creates the network in Fig. 9.3.
1 2 3 4 5 6 7 set set set set $ns $ns $ns ns [new Simulator] n1 [$ns node] n2 [$ns node] n3 [$ns node] duplex-link $n1 $n2 5Mb 2ms DropTail duplex-link $n2 $n3 5Mb 2ms DropTail duplex-link $n1 $n3 5Mb 2ms DropTail

8 9 10 11 12 13 14

#=== UDP-Null peering starts here === set udp [new Agent/UDP] set null [new Agent/Null] set cbr [new Application/Traffic/CBR] $cbr attach-agent $udp $ns attach-agent $n1 $udp $ns attach-agent $n3 $null $ns connect $udp $null

(iii) Use instproc attach-agent{node agent} of class Simulator to create a connection between each agent and a node entry (Lines 12 and 13). (iv) Use instproc connect{src dst} of class Simulator to associate a sending agent with a receiving agent (Line 14). 9.2.3 Internal Mechanism for Agents The internal mechanisms for agents are deﬁned in the C++ domain as follows: • A sending agent: Receive user demand by having the associated application invoke its function sendmsg(...). From within sendmsg(...), create packets using function allocpkt() and forward the created packets to the low-level network by executing target_->recv(p,h). • A receiving agent: Receive packets by having a low-level network demultiplexer invoke its function recv(p,h). Destroy received packets by invoking function free(p) of class Packet. In this section, we will discuss the detail of variables and functions required to perform the above mechanisms. Related C++ and OTcl Variables The main variables of C++ class Agent and their bound OTcl instvars are shown in Table 9.3. Of type ns_addr_t (see Section 8.3.3), variables here_ and dst_ contain addresses and ports of the Node attached to the agent and the peering agent, respectively. An IPv6 priority level is stored in variable prio_. Variable app_ acts as a reference to an Application object. Since

9.2 Basic Agents Table 9.3. The list of C++ and OTcl variables of class Agent. C++ Type ns_addr_t C++ variable OTcl instvar Description here_ here_.addr_ agent_addr_ Address of the attached node here_.port_ agent_port_ Port where the agent is attached dst_ dst_.addr_ dst_.port_ int int int int int packet_t int int fid_ prio_ flags_ defttl_ size_ type_ seqno_ uidcnt_ dst_addr_ dst_port_ fid_ prio_ flags_ ttl_ N/A N/A N/A N/A N/A

207

ns_addr_t

Address of the node attaching to a peering agent Port where the peering agent is attached Flow ID IPv6 priority ﬁeld (e.g., 0 = unspeciﬁed, 1 = background traﬃc) Flags Default time to live value Packet size Packet type Current sequence number A pointer to an application Total number of packets generated by all agents

Application* app_

class Agent is responsible for packet generation, it must assign a unique ID to every packet. Therefore, it maintains a static variable uidcnt_, which counts the total number of generated packets. When a packet is created, an Agent object sets the unique ID of the packet to be uidcnt_, and increases uidcnt_ by one (see function initpkt(p) in Line 10 of Program 9.3). Key C++ Functions A list of key C++ functions with their descriptions is given below (see the declaration of class Agent in ﬁle ˜ns/common/agent.cc,h). Since class Agent is a template for transport layer agents, it provides no implementation for some of its functions. The child classes of class Agent are responsible for implementing these functions. recv(p,h) send(p,h) send(nbytes) sendmsg(nbytes) Receive a packet “*p”. Send a packet “*p”. Send a message with “nbytes” bytes. Send a message with “nbytes” bytes (no implementation). timeout(tno) Action to be performed at timeout (No implementation)

208

9 Transport Control Protocols Part 1– An Overview

connect(dst) Connect to a dynamic destination dst (no implementation). close() Close a connection-oriented session (no implementation). listen() Wait for a connection-oriented session (no implementation). attachApp(app) Store app in the variable app_. allocpkt() Create a packet. initpkt(p) Initialize the input packet “*p”. recvBytes(bytes) Send data of “bytes” bytes to the attached application. idle() Tell the application that the agent has nothing to transmit. The Constructor Class Agent has no default constructor. Its only constructor takes a packet_t (see Section 8.3.4 and Program 8.9) object as an input argument (see Line 1 of Program 9.3). The constructor sets the variable type_ to be as speciﬁed in the input argument, and resets other variables to zero. This packet type setting implies that one agent is able to transmit packets of one type only. We need several agents to transmit packets of several types. Functions allocpkt() and initpkt(p) Shown in Program 9.3, function allocpkt() is the main packet construction function. It creates a packet by invoking function alloc() of class Packet in Line 4, and initializes the packet by invoking function initpkt(p) in Line 5. After initialization, function allocpkt() returns the constructed packet pointer p to the caller. The details of function initpkt(p) are shown in Lines 8–20 of Program 9.3. Function initpkt(p) sets the initial values in the packet header of the input packet “*p” to the default values. The uniqueness of the unique ID ﬁeld uid_ in the common header is assured by setting uid_ to be the total number of packets allocated so far. Class Agent stores the total number of allocated packet in its static variable unicnt_. Since the variable unicnt_ is distinct and unique to all agents, assigning this variable to the ﬁeld uid_ of the common header (Line 11) assures the uniqueness of packet unique ID. Other initialization includes setting up the packet type in the common header to be as speciﬁed in the variable type_ (Line 12). Also, Lines 14– 18 conﬁgure source and destination IP addresses and port numbers in the variables here_ and dst_. Function attachApp(app) Lines 1–4 in Program 9.4 show the details of function attachApp(app). To bind an application to an agent, function attachApp(app) stores the input

9.2 Basic Agents

209

Program 9.3 Constructor, function allocpkt, and function initpkt of class Agent.
1 2 3 4 5 6 //~/ns/common/agent.cc Agent::Agent(packet_t pkttype):size_(0),type_(pkttype),app_(0){} Packet* Agent::allocpkt() const { Packet* p = Packet::alloc(); initpkt(p); return (p); } void Agent::initpkt(Packet* p) { hdr_cmn* ch = hdr_cmn::access(p); ch->uid() = uidcnt_++; ch->ptype() = type_; ... hdr_ip* iph = hdr_ip::access(p); iph->saddr() = here_.addr_; iph->sport() = here_.port_; iph->daddr() = dst_.addr_; iph->dport() = dst_.port_; ... }

7 8 9 10 11 12 13 14 15 16 17 18

pointer “app” in its pointer to a Application object, app_. After this point, the agent may invoke public functions of the attached application through the pointer app_. Program 9.4 Functions attachApp and recv of class Agent.
1 2 3 4 //~/ns/common/agent.cc void Agent::attachApp(Application *app) { app_ = app; }

5 void Agent::recv(Packet* p, Handler*) 6 { 7 if (app_) 8 app_->recv(hdr_cmn::access(p)->size()); 9 Packet::free(p); 10 }

210

9 Transport Control Protocols Part 1– An Overview

Functions recv(p,h), send(p,h), and sendmsg(nbytes) These functions are used by sending and receiving agents in the packet forwarding process. On the sender side, an application informs a sending agent of user demand by invoking functions send(nbytes), and sendmsg(...) of class Agent. As an NsObject, the sending agent forwards an incoming packet *p to a downstream NsObject by executing target_->recv(p,h). As discussed earlier, these functions send(nbytes) and sendmsg(...) have no implementation in the scope of class Agent, and must be implemented by the child classes of class Agent. On the receiver side, an NsObject forwards packets to a receiving agent by invoking its function recv(p,h). Shown in Lines 5–10 of Program 9.4, function recv(p,h) deallocates the received packet (Line 9) and may inform the attached application (if it exists) of packet reception by invoking function recv(size) of the attached Application object (Lines 7–8), where size is the size of packet *p. 9.2.4 Guidelines to Deﬁne a New Transport Layer Agent Class Agent provides the basic functionalities necessary for most agents. A new agent can be created based on these functionalities, following the guidelines below: (i) Deﬁne an inheritance structure: Select a base class and derive a new agent class from the selected base class. Bind the C++ and OTcl agent class names together. (ii) Deﬁne necessary C++ variables and OTcl instvars. (iii) Implement the constructors of both C++ and OTcl classes. Bind the variables and the instvars if necessary. (iv) Implement the necessary functions including functions send(nbyte), send- msg(...), recv(p,h), and timeout(tno). Also deﬁne OTcl instprocs if necessary. (v) Deﬁne necessary OTcl commands as interfaces to the C++ domain from the OTcl domain. (vi) [Optional]Deﬁne a timer (see Section 12.1).

9.3 UDP (User Datagram Protocol) and Null Agents
UDP (User Datagram Protocol) is a connectionless transport layer protocol, which provides neither congestion control nor error control. In NS2, a UDP agent is used as a sending agent. It is usually peered with a Null (receiving) agent, which is responsible for packet destruction. Figure 9.3 shows a network conﬁguration example where a CBR (Constant Bit Rate) traﬃc source employs a UDP agent and a Null agent as its transport later agents. Here,

9.3 UDP (User Datagram Protocol) and Null Agents

211

the CBR asks the UDP agent to transmit a burst of packets for every ﬁxed interval. The UDP agent creates and forwards packets to the low-level network, irrespective of the network condition. On the receiving end, the Null agent simply destroys the packets received from the low-level network. In the following, we will discuss the details of UDP and Null agents. 9.3.1 Null (Receiving) Agents A Null agent is the simplest but one of the most widely-used receiving agents. The main responsibility of a Null agent is to deallocate packets, through function free(p) of class Packet (see Line 9 in Program 9.4). A Null agent is represented by an OTcl class Agent/Null which is derived directly from an OTcl class Agent (see ﬁle ˜ns/tcl/lib/ns-agent.tcl). Due to its simplicity, Null agents have no implementation in the C++ domain. 9.3.2 UDP (Sending) Agent A UDP agent is perhaps the simplest form of sending agents. It receives user demand to transmit data by having the attached application invoke its function (e.g., sendmsg(...)), creates packets based on the demand, and forwards the created packet to a low-level network. An application may use three following ways to tell a UDP agent to send out packets: via a C++ function sendmsg(...) of class UdpAgent, via an OTcl command send{...} of OTcl class Agent/UDP, or via an OTcl command sendmsg{...} of OTcl class Agent/UDP. Again, NS2 deﬁnes a UDP sending agent based on the guideline in Section 9.2.4. Since a UDP agent implements no acknowledgement mechanism and needs no timer, we can skip the last step in the guideline. Step 1: Deﬁne Inheritance Structure A UDP agent is represented by a C++ class UdpAgent and an OTcl class Agent/UDP. These two classes derive from class Agent in their domains, and are bound by using a mapping class UdpAgentClass (see Program 9.5). Step 2: Deﬁne C++ Variables and OTcl Instvars The key variable of class UdpAgent is seqno_ (Line 12 in Program 9.6), which counts the number of packets generated by a UdpAgent object. Note that every packet has a unique ID uid_. Also, every packet generated by the same agent has a unique sequence number seqno_. However, two packets generated by diﬀerent agents may have the same sequence number seqno_ but they must have diﬀerent unique ID uid_.

212

9 Transport Control Protocols Part 1– An Overview

Program 9.5 Mapping class UdpAgentClass which binds a C++ class UdpAgent to an OTcl class Agent/UDP.
1 2 3 4 5 6 7 //~/ns/apps/udp.cc static class UdpAgentClass : public TclClass { public: UdpAgentClass() : TclClass("Agent/UDP") {} TclObject* create(int, const char*const*) { return (new UdpAgent()); } } class_udp_agent;

Step 3: Implement the Constructors in the C++ and OTcl Domains NS2 implements constructors for a UDP agent in the C++ domain only. From Program 9.6, the default constructor in Lines 14–16 feeds UDP packet type (i.e., PT_UDP) to constructor of class Agent, essentially storing PT_UDP in the variable type . It also sets the sequence number (i.e., seqno ) to be –1. By specifying the packet type, the constructor in Lines 17–19 sets the packet type to be as speciﬁed in the input argument. The constructor in this case does not set the value of seqno_ since the packets of speciﬁed type may not have sequence number. For both constructors, the C++ variable size_, which speciﬁes the packet size, is bound to instvar packetSize_ in the OTcl domain (Lines 15 and 19). By default, the packet size is set to 1,000 bytes in ﬁle ˜ns/tcl/lib/ns-default.tcl (Line 20). Step 4: Deﬁne the Necessary C++ Functions As a sending agent, a UDP agent needs to deﬁne a function sendmsg(...) to receive a user demand from the application. Program 9.7 shows the details of function sendmsg(nbytes,data,flags), which takes three input arguments: nbytes, data, and flags. Function sendmsg(...) divides data payload with size nbytes bytes into “n” (see Line 4) or “n+1” parts (depending on nbytes), stores each part into a UDP packet (which contains a payload of size_ bytes), and transmits all (“n” or “n+1”) packets to a low-level network. Since NS2 rarely sends actual payload along with a packet (see Section 8.4), Line 8 only sets the size of packet created in Line 6 to be size_. Line 11 sends out the created packet, by executing target_->recv(p).1 Lines 6–11 are repeated for “n” times to transmit all “n” packets. After transmitting the ﬁrst “n” packets, the entire application payload is left with nbytes % size_, where % is a modulus operator. If the remainder is
1

Variable target is conﬁgured to point to a node entry during the network conﬁguration phase (see Section 9.2.2).

9.3 UDP (User Datagram Protocol) and Null Agents

213

Program 9.6 Declaration and the constructors of class UdpAgent as well as the default value of the instvar packetSize of class Agent/UDP.
1 2 3 4 5 6 7 8 9 10 11 12 13 //~/ns/apps/udp.h class UdpAgent : public Agent { public: UdpAgent(); UdpAgent(packet_t); virtual void sendmsg(int nbytes, const char *flags = 0){ sendmsg(nbytes, NULL, flags); } virtual void sendmsg(int nbytes, AppData* data, ... const char *flags = 0); virtual void recv(Packet* pkt, Handler*); virtual int command(int argc, const char*const* argv); protected: int seqno_; };

//~/ns/apps/udp.cc 14 UdpAgent::UdpAgent() : Agent(PT_UDP), seqno_(-1){ 15 bind("packetSize_", &size_); 16 } 17 UdpAgent::UdpAgent(packet_t type) : Agent(type){ 18 bind("packetSize_", &size_); 19 } //~/ns/tcl/lib/ns-default.tcl 20 Agent/UDP set packetSize_ 1000

nonzero, Lines 15–20 will transmit the remaining application payload in another packet. Finally, Line 22 invokes function idle() to inform the attached application that the UDP agent has ﬁnished data transmission. From Line 24, function idle() does so by invoking function resume() of the attached application (if any). There are two important notes for UDP agents. First, since a UDP agent is a sending agent its function recv(p,h) is generally not to be used. Secondly, in Program 9.7, function sendmsg(...) transmits packets, irrespective of network condition. Step 5: Deﬁne OTcl Commands and Instprocs Class Agent/UDP deﬁnes the two following OTcl commands deﬁned in Program 9.8:

214

9 Transport Control Protocols Part 1– An Overview

Program 9.7 Function sendmsg of class UdpAgent and function idle of class Agent.
//~/ns/apps/udp.cc 1 void UdpAgent::sendmsg(int nbytes,AppData* data,const char* flags) 2 { 3 Packet *p; 4 int n = nbytes / size_; 5 while (n-- > 0) { 6 p = allocpkt(); 7 /* packet header configuration */ 8 hdr_cmn::access(p)->size() = size_; 9 ... 10 /* --------------------------- */ 11 target_->recv(p); 12 } 13 14 15 16 17 18 19 20 21 22 23 } n = nbytes % size_; if (n > 0) { p = allocpkt(); /* packet header configuration */ hdr_cmn::access(p)->size() = n; ... /* --------------------------- */ target_->recv(p); } idle();

//~/ns/common/agent.cc 24 void Agent::idle() { if (app_) app_->resume(); } 25 }

• send{nbytes str}: Send a payload of size “nbytes” containing a message “str”. • sendmsg{nbytes str flags}: Similar to the command send but also passes the input ﬂag “flags” when sending a packet. Lines 5–8 in Program 9.8 show the details of the OTcl command send{...}. Line 5 creates a PacketData object. Line 6 stores the input message str in the created PacketData object. Line 7 sends out the application payload by invoking function sendmsg(...). Note that the size of application payload does not depend on the size of the message in the PacketData object (i.e., argv[3] or str). Rather, it is speciﬁed in the ﬁrst input argument (i.e., argv[2] or nbytes). The implementation of the OTcl command sendmsg(...) is similar to that of an OTcl command send{...}. However, it also feeds a ﬂag “flags” as an input argument of function sendmsg(...) (see Line 14).

9.4 Chapter Summary

215

Program 9.8 OTcl Commands send and sendmsg of class Agent/UDP.
//~/ns/apps/udp.cc 1 int UdpAgent::command(int argc, const char*const* argv) 2 { 3 if (argc == 4) { 4 if (strcmp(argv[1], "send") == 0) { 5 PacketData* data = new PacketData(1 + strlen(argv[3])); 6 strcpy((char*)data->data(), argv[3]); 7 sendmsg(atoi(argv[2]), data); 8 return (TCL_OK); 9 } 10 } else if (argc == 5) { 11 if (strcmp(argv[1], "sendmsg") == 0) { 12 PacketData* data = new PacketData(1 + strlen(argv[3])); 13 strcpy((char*)data->data(), argv[3]); 14 sendmsg(atoi(argv[2]), data, argv[4]); 15 return (TCL_OK); 16 } 17 } 18 return (Agent::command(argc, argv)); 19 }

9.3.3 Setting Up a UDP Connection A UDP connection can be created by the network conﬁguration method provided in Section 9.2.2. An example connection where a UDP agent, a Null agent, and a CBR traﬃc source are used as a sending agent, a receiving agent, and an application is shown in Example 9.1.

9.4 Chapter Summary
An agent is a connector which bridges an application to a low-level network. Its main responsibilities are to create packets based on user demand received from an application, to forward packets to a low-level network, and to destroy packets received from a low-level network. From this point of view, an agent can be used to model transport layer protocols and routing protocols. This chapter focuses on transport layer (protocol) agents only. Class Agent is a base class, which represents both sending and receiving agents. It connects to an application and a low-level network using pointers app_ and target_. An application also has a pointer agent_ to an agent, while a low-level network uses a pointer target_ as a reference to an agent. Class Agent provides basic functionalities for creating, forwarding, and destroying packets. Its functions send(...) and sendmsg(...) are invoked by an attached application to pass on user demand. An agent creates packets

216

9 Transport Control Protocols Part 1– An Overview

based on the demand, and forwards the created packet to a low-level network by executing target_->recv(p,h). A low level network sends a packet to a receiving agent which may destroy the packet by invoking function recv(p,h) of the receiving agent. User Datagram Protocol (UDP) and Transmission Control Protocol (TCP) are among the most widely used transport layer protocols. UDP is a simple transport layer protocol and it can be ﬂexibly used by other protocols. In NS2, UDP is implemented in the C++ class UdpAgent which is bound to an OTcl class Agent/UDP. A UDP agent is usually peered with a Null agent, which simply destroys a received packet. TCP is a reliable transport control protocol. Its main features are end-toend error control and network congestion control. It implements timeout and acknowledgement to provide end-to-end error control, and adopts a windowbased rate adjustment to control network congestion. We will discuss the details of TCP implementation in NS2 in the next chapter.

10 Transport Control Protocols Part 2 – Transmission Control Protocol (TCP)

As a transport control protocol, TCP (Transmission Control Protocol) bridges an application to a low-level network, controls network congestion, and provides reliability to an end-to-end connection. This chapter discusses the details of TCP agents. Section 10.1 gives an overview of TCP agents. Here, we show a TCP network conﬁguration method, a brief overview of TCP internal mechanism, TCP header format, and the main steps in deﬁning TCP senders and TCP receivers. Sections 10.2 and 10.3 discuss the implementation of TCP receivers and senders, respectively. Sections 10.4–10.7 presents the implementation of four main functionalities of a TCP sender. Finally, the chapter summary is provided in Section 10.8.

10.1 An Overview of TCP Agents in NS2
Based on user demand from an application, a TCP sender creates and forwards packets to a low-level network. It controls the congestion by limiting the rate (i.e., by adjusting the congestion window) at which packets are fed to the low-level network. It enforces an acknowledgment mechanism to provide connection reliability. A TCP receiver must acknowledge every received TCP packet. Based on the acknowledgment pattern, a TCP sender determines whether the transmitted packet was lost or not. If so, it will retransmit the packet. A TCP sender is responsible for sending packets as well as controlling the transmission rate, while the role of a TCP receiver is only to return acknowledgments to the associated TCP sender. 10.1.1 Setting Up a TCP Connection As a transport layer agent, TCP can be incorporated into a network by using the method discussed in Section 9.2.2.

T. Issariyakul, E. Hossain, Introduction to Network Simulator NS2, DOI: 10.1007/978-0-387-71760-9 10, c Springer Science+Business Media, LLC 2009

218

10 Transport Control Protocols Part 2 – Transmission Control Protocol

Example 10.1. Consider Fig. 9.3. Replace the CBR application with FTP (File Transfer Protocol), the UDP agent with a TCP sender, and the Null agent with a TCP receiver. The modiﬁed network can be created by using the following Tcl simulation script. 1 2 3 4 5 6 7 set set set set $ns $ns $ns ns [new Simulator] n1 [$ns node] n2 [$ns node] n3 [$ns node] duplex-link $n1 $n2 5Mb 2ms DropTail duplex-link $n2 $n3 5Mb 2ms DropTail duplex-link $n1 $n3 5Mb 2ms DropTail

8 9 10 11 12 13 14 15

#=== TCP connection setup starts here === set tcp [new Agent/TCP] set sink [new Agent/TCPSink] set ftp [new Application/FTP] $ns attach-agent $n1 $tcp $ns attach-agent $n3 $sink $ftp attach-agent $tcp $ns connect $tcp $sink $ns at 0.0 "$ftp start"

Similar to those in Example 9.1, Lines 8–14 above create a TCP connection on top of a low-level network. 10.1.2 Packet Transmission and Acknowledgment Mechanism TCP provides connection reliability by means of acknowledgment and packet retransmission. Figure 10.1 shows a diagram for TCP packet transmission and acknowledgment mechanisms. The process starts when an application (e.g., FTP) informs a TCP sender (e.g., TcpAgent) of user demand by invoking function sendmsg(nbytes) of the TcpAgent object through its variable agent_. The TCP sender creates TCP packets, and forwards them to its downstream object by executing target_->recv(p,h). The low-level network delivers the packets to the destination node attached to the TCP receiver (i.e., TcpSink). The destination node forwards the packet to the TCP receiver (i.e., a TcpSink object) by invoking function recv(p,h) of the TCP receiver installed in its demultiplexer (e.g., dmux_). Upon receiving a TCP packet, the TCP receiver creates an ACK packet and returns it to the TCP sender by executing target_->recv(p,h). The low-level network delivers the ACK packet to the sending node, which forwards the ACK packet to the TCP sender via its demultiplexer. If a TCP packet or an ACK packet is lost (or delayed for a long period of time), the TCP sender will assume that the packet is lost. In this case, the

10.1 An Overview of TCP Agents in NS2
FTP Application agent_ 219

User demand

TcpAgent (Sending)
Agents

TcpAgent (Receiving) recv target_ ACK ACK Packet

sendmsg target_ Packet

recv

Node
Low-Level Network

Node recv dmux_ entry_ dmux_

recv entry_ Fig. 10.1. TCP packet transmission and acknowledgment mechanisms.

TCP sender will retransmit the lost TCP packet using target_->recv(p,h) (see the description of the retransmission process in Section 9.1.2). 10.1.3 TCP Header TCP packet header is deﬁned in the “hdr_tcp” struct data type shown in Program 10.1. The key variables of hdr_tcp include seqno_ TCP sequence number ts_ Timestamp: The time when the packet was generated ts_echo_ Timestamp echo: The time when the peering TCP received the packet reason_ Reason for packet transmission (e.g., 0 = normal transmission) In common with other packet header, hdr_tcp contains function access(p) (Lines 8–10), which can be used to obtain the reference to a TCP header stored in the input packet *p. This reference can then be used to access the attributes of a TCP packet header. 10.1.4 Deﬁning TCP Sender and Receiver We follow the guidelines provided in Section 9.2.4 to deﬁne a TCP sender and a TCP receiver.

220

10 Transport Control Protocols Part 2 – Transmission Control Protocol

Program 10.1 Declaration of hdr tcp struct data type.
//~/ns/tcp/tcp.h 1 struct hdr_tcp { 2 double ts_; /*time packet generated (at source)*/ 3 double ts_echo_; /*the echoed timestamp*/ 4 int seqno_; /*sequence number */ 5 int reason_; /*reason for a retransmit */ 6 static int offset_; // offset for this header 7 inline static int& offset() { return offset_; } 8 inline static hdr_tcp* access(Packet* p) { 9 return (hdr_tcp*) p->access(offset_); 10 } 11 int& seqno() { return (seqno_); } 12 ... 13 };

Step 1: Deﬁne the Inheritance Structure NS2 deﬁnes TCP sender in a C++ class TcpAgent which is bound to an OTcl class Agent/TCP through a mapping class TcpClass, as shown in Lines 1–7 of Program 10.2. Similarly, TCP receiver is deﬁned in a C++ class TcpSink which is bound to an OTcl class Agent/TCPSink through a mapping class TcpSinkClass, as shown in Lines 8–14 of Program 10.2. Step 2: Deﬁne Necessary C++ and OTcl Variables While class TcpSink has only one C++ key variable acker_ which is of class Acker,1 class TcpAgent has several variables. We classify the key C++ variables of class TcpAgent into four categories. First, C++ variables, whose values change dynamically during a simulation, are shown in Table 10.1. Secondly, C++ variables, which are usually conﬁgured once, are listed in Table 10.2. Thirdly, Table 10.3 shows variables which are related to TCP timer mechanism. Finally, Table 10.4 shows the other non-classiﬁed variables of class TcpAgent. Step 3: Implement the Constructor The constructors of both TCP senders and TCP receivers set their variables to the default values, and bind C++ variables to OTcl instvars as speciﬁed in Tables 10.1–10.3. In addition, the constructor of the TCP sender invokes the constructor of its parent class (i.e., Agent) with an input argument PT_TCP, setting the instantiated TcpAgent object to transmit TCP packet only. It also initializes the retransmission timer rtx_timer_ with the pointer “this”
1

We will be discuss the details of class Acker later in Section 10.2.1.

10.1 An Overview of TCP Agents in NS2

221

Program 10.2 Class TcpClass which binds a C++ class TcpAgent and an OTcl class Agent/TCP together, class TcpSinkClass, which binds a C++ class TcpSink and an OTcl class Agent/TCPSink together, and the constructor of class TcpSink.
1 2 3 4 5 6 7 //~/ns/common/tcp.cc static class TcpClass : public TclClass { public: TcpClass() : TclClass("Agent/TCP") {} TclObject* create(int , const char*const*) { return (new TcpAgent()); } } class_tcp;

//~/ns/common/tcp-sink.cc 8 static class TcpSinkClass : public TclClass { 9 public: 10 TcpSinkClass() : TclClass("Agent/TCPSink") {} 11 TclObject* create(int, const char*const*) { 12 return (new TcpSink(new Acker)); 13 } 14 } class_tcpsink; 15 TcpSink::TcpSink(Acker* acker) : Agent(PT_ACK), acker_(acker) {...}

to itself. The details of TcpAgent construction and timers are given in ﬁle ˜ns/tcp/tcp.cc and Section 12.1. A TCP receiver is somewhat diﬀerent from a TCP sender, since it does not have a default constructor. From Line 15 of Program 10.2, the constructor takes a pointer to an Acker object as an input argument (see Section 10.2.1), and initializes its variable ack_ with this input pointer. It also initializes its parent constructor with PT_ACK, an ACK packet type. Finally, it binds few C++ variables to OTcl instvars (see the detailed construction of class TcpSink in ﬁle ˜ns/tcp/tcp-sink.cc). Steps 3, 4, and 5: Implement Necessary Functions, OTcl Commands, and Instprocs, and Deﬁne Timers if Necessary The detailed implementation of C++ functions of TCP receivers are shown in the next section, while those of TCP senders are given in Sections 10.3–10.7. For brevity, we will not discuss the details of implementation of OTcl command and instproc. The readers are encouraged to study the details of TCP senders and TCP receivers in ﬁles ˜ns/tcp/tcp.cc,h, ˜ns/tcp/tcpsink.cc,h, and ˜ns/tcl/lib/ns-agent.tcl.

222

10 Transport Control Protocols Part 2 – Transmission Control Protocol Table 10.1. Key operating variables of class TcpAgent.

C++ variable OTcl variable t_seqno_ curseq_ t_seqno_ seqno_

Default Description value 0 Current TCP sequence number 0 Total number of packets need to be transmitted speciﬁed by the application. A TCP sender transmits packets as long as its sequence number is less than curseq_. 0 0 0 0 0 0 0 0 0 0 0 –1 0 0 0 N/A Highest ACK number (not frozen during Fast Recovery) Highest ACK number (frozen during Fast Recovery) Congestion window size in packets Slow-start threshold Duplicated ACK counter Highest transmitted sequence number RTT sample Smoothed RTT RTT deviation Current RTO backoﬀ multiplicative factor Status of the RTT collection process Time at which the packet is transmitted Sequence number of the tagged packet Current value of unbounded retransmission timeout Latest timestamp provided by the peering TCP receiver Retransmission timer object

highest_ack_ ack_ lastack_ cwnd_ ssthresh_ dupacks_ maxseq_ t_rtt_ t_srtt_ t_rttvar_ t_backoff_ rtt_active_ rtt_ts_ rtt_seq_ t_rtxcur_ ts_peer_ rtx_timer_ N/A cwnd_ ssthresh_ dupacks_ maxseq_ rtt_ srtt_ rttvar_ backoff_ N/A N/A N/A N/A N/A N/A

10.2 TCP Receiver
A TCP receiver is responsible for deallocating received TCP packets and returning cumulative ACK packets to the TCP sender. As discussed in Section 9.1.2, a cumulative ACK packet acknowledges a TCP packet with the highest contiguous sequence number. Upon receiving a cumulative ACK packet, the TCP sender assumes that all packets whose sequence numbers are lower than or equal to that of the ACK packet have been successfully received. A cumulative ACK packet has the capability of acknowledging multiple packets. For example, suppose Packet 3 in Fig. 9.2 is not lost but is delayed and that it arrives the receiver right after Packet 6 is received. Upon receiving Packet 3, the receiver acknowledges with A6, since it has received Packets 4–6 earlier.

10.2 TCP Receiver Table 10.2. Key variables of class TcpAgent. C++ variable wnd_ numdupacks_ OTcl variable window_ numdupacks_ Default value 20 3 Description

223

Upper bound on window size Number of duplicated ACKs which triggers Fast Retransmit Initial value of window size TCP packet size in bytes TCP basic header size in bytes If true, TCP and IP header size will be added to packet size Maximum number of bytes that a TCP sender can transmit in one transmission Upper bound on cwnd_ If set to 1, do not open the congestion window when the network is limited (See Section 10.5).

wnd_init_ size_ tcpip_base_ useHeaders_

windowInit_ packetSize_ tcpip_base_hdr_size_ useHeaders_

2 1,000 40 true

maxburst_

maxburst_

0

maxcwnd_ control_

maxcwnd_ control_increase_

0 0

increase_ Table 10.3. Timer related variables of class TcpAgent. C++ variable srtt_init_ rttvar_init_ rtxcur_init_ T_SRTT_BITS T_RTTVAR_BITS rttvar_exp_ decrease_num_ increase_num_ tcpTick_ maxrto_ minrto_ OTcl variable Default Description value srtt_init_ 0 Initial value of t_srtt_ rttvar_init_ 12 Initial value of t_rttvar_ rtxcur_init_ 3.0 Initial value of t_rtxcur_ T_SRTT_BITS 3 Multiplicative factor for smoothed RTT T_RTTVAR_BITS 2 Multiplicative factor for RTT deviation rttvar_exp_ 2 Multiplicative factor for RTO computation decrease_num_ 0.5 Window decreasing factor increase_num_ 1.0 Window increasing factor tcpTick_ 0.01 Timer granularity in seconds maxrto_ 100,000 Upper bound on RTO in seconds minrto_ 0.2 Lower bound on RTO in seconds

224

10 Transport Control Protocols Part 2 – Transmission Control Protocol Table 10.4. Miscellaneous variables of class TcpAgent.

C++ variable cong_action_ sigledup_

Default Description value 0 true when the congestion has occurred. 1 If set to 1, the TCP sender will transmit new packets upon receiving ﬁrst few duplicated ACK packets. Sequence number of an ACK packet received prior to the current ACK packet. The latest action on congestion window The highest transmitted sequence number during the previous packet loss event

prev_highest_ack_ last_cwnd_action_ recover_

N/A N/A N/A

In NS2, C++ implementation of TCP receivers involves two main classes: Acker and TcpSink. Class Acker is a helper class responsible for generating ACK packets. Class TcpSink contains an Acker object, and acts as interfaces to a peering TCP sender and the OTcl domain. 10.2.1 Class Acker

Program 10.3 Declaration of class Acker.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 //~/ns/tcp/tcp-sink.h class Acker { public: Acker(); virtual ~Acker() { delete[] seen_; } inline int Seqno() const { return (next_ - 1); } inline int Maxseen() const { return (maxseen_); } int update(int seqno, int numBytes); protected: int next_; int maxseen_; int wndmask_; int *seen_; int is_dup_; };

Program 10.3 shows the declaration of a C++ class Acker.2 Class Acker stores necessary information required to generate cumulative ACK packets in the following variables:
2

Class Acker is not implemented in the OTcl domain.

10.2 TCP Receiver

225

seen_ next_ maxseen_ wndmask_ is_dup_

An array whose index and value are the sequence number and the corresponding packet size, respectively Expected sequence number Highest sequence number ever received Modulus mask, initialized to maximum window size-1 (set to 63 by default; see Section 12.4.1) True if the latest received TCP packet was received earlier

Figure 10.2 shows an example of information stored in an Acker object. In this case, Packets 1, 2, 3, 5, and 7 are received, but Packets 4 and 6 are missing. Therefore, next_ and maxseen_ are set to 4 and 7, respectively. Also, variable seen_ stores the size in bytes of Packets 1–7 in its respective entries. To determine whether packet n is missing, class Acker checks the value of seen_[n]. The packet is missing if and only if seen_[n] is zero. Suppose a TCP receiver receives a TCP packet number 4 when the status of the Acker object is as in Fig. 10.2. The Acker object will generate an ACK packet with sequence number 5. However, if the sequence number of the received packet is not 4, the Acker object will create an ACK packet with sequence number 3. As discussed in Section 9.1.2, a TCP connection can have at most w unacknowledged packets in a network, where w is the current congestion window size. Let MWS be the Maximum Window Size in a simulation (see Line 6 in Program 10.4). Then, w ∈ {0, · · · , MWS} and there can be at most MWS unacknowledged packets during the entire simulation. An Acker object needs only MWS entries in the array variable seen_ to store information about unacknowledged packets. Program 10.4 shows the constructor of the C++ class Acker. The constructor resets next_ and maxseen_ to zero in Line 1. Line 3 allocates memory space for array variable seen_ with MWS entries. Line 4 clears the allocated memory to zero. Also, wndmask_ is set to MWM (Maximum Window Mask which is set to 63 in Line 7). seen_ [seqno] (bytes)

Case I: seq < 3

Case II: seq = [next_,maxseen_]

Case III: seq > 7

1

2

3

5

7

seqno

next_ = 4

maxseen_ = 7

Fig. 10.2. Information necessary to generate a cumulative acknowledgement.

226

10 Transport Control Protocols Part 2 – Transmission Control Protocol

Program 10.4 The constructor of class Acker.
1 2 3 4 5 //~/ns/tcp/tcp-sink.cc Acker::Acker() : next_(0), maxseen_(0), wndmask_(MWM) { seen_ = new int[MWS]; memset(seen_, 0, (sizeof(int) * (MWS))); }

//~/ns/tcp/tcp-sink.cc 6 #define MWS 64 7 #define MWM (MWS-1)

The above MWS (set by default to 64 in Line 6 of Program 10.4) entries of seen_ are reused to store the packet size corresponding to all incoming TCP sequence numbers. Class Acker employs a modulus operation to map a sequence number to an array index. Upon receiving a TCP packet with sequence number seqno, an Acker object stores the packet size in the entry seqno % MWS (which is the remainder of seqno/MWS), of the array seen_, where “%” is a modulus operator. When seqno exceeds MWS, seqno % MWS will be restarted from the ﬁrst entry to reuse the memory allocated to seen_. As discussed in Section 12.4.1, a modulus operation can also be implemented by bit masking. In particular, seqno % MWS is in fact equivalent to seqno & wndmask_, where wndmask_ is set initially to MWM in the constructor (Line 1 in Program 10.4), and MWM (Maximum Window Mask) is deﬁned as 63 (Lines 6–7 in Program 10.4). To facilitate the understanding, readers may simply regard seqno & wndmask_ as seqno % 63. Class Acker has two key functions: Seqno() and update(seq,numBytes). Function Seqno() (Line 5 in Program 10.3) returns the highest sequence number of a burst of contiguously received packets. As shown in Program 10.5, function update(seq,numBytes) updates its internal variables according to the input arguments. Function update(seq,numBytes) takes two input arguments: seq and numBytes which are the sequence number and the size of an incoming TCP packet, respectively. It updates variables next_, maxseen_, seen_, and is_dup_ and returns the number of in-sequence bytes which is ready to be delivered to the application. From Fig. 10.2, “seq” can be (I) less than next_, (II) between next_ and maxseen_, and (III) greater than maxseen_. Function update(seq,numBytes) reacts to these three cases as follows: (i) If seq < next_, function update(seq,numBytes) will set is_dup_ to be true (Line 17). This case implies that this packet was received earlier, and therefore, this packet is a duplicated packet. (ii) If seq lies in between next_ and maxseen_, function update(seq, numBytes) will execute Lines 19–26. Line 19 determines whether seq was received earlier. This happens to be true under the two following

10.2 TCP Receiver

227

Program 10.5 Function update of class Acker.
//~/ns/tcp/tcp-sink.cc 1 int Acker::update(int seq, int numBytes) 2 { 3 bool just_marked_as_seen = FALSE; 4 is_dup_ = FALSE; 5 int numToDeliver = 0; 6 if (seq > maxseen_) { 7 int i; 8 for (i = maxseen_ + 1; i < seq; ++i) 9 seen_[i & wndmask_] = 0; 10 maxseen_ = seq; 11 seen_[maxseen_ & wndmask_] = numBytes; 12 seen_[(maxseen_ + 1) & wndmask_] = 0; 13 just_marked_as_seen = TRUE; 14 } 15 int next = next_; 16 if (seq < next) 17 is_dup_ = TRUE; 18 if (seq >= next && seq maxseen_, implying a new TCP packet, function update(...) will execute Lines 7–13. Lines 8–9 and 12 clear the seen_[maxseen_+1] through seen_[seq-1]. It updates maxseen_ in Line 10 and stores the
3

Bit masking with wndmask has the same impact as a modulus with wndmask +1 does.

228

10 Transport Control Protocols Part 2 – Transmission Control Protocol

packet size in seen_[seq & wndmask_] in Line 11. Since Line 10 stores seq in maxseen_, the condition in Line 18 is satisﬁed and Lines 19–26 are to be executed. If Case (iii) is executed, Case (ii) will also be executed. Therefore, Line 13 sets just_marked_as_seen to be true, which simply indicates that the current packet is not a duplicated packet, and Line 20 should be skipped. 10.2.2 Class TcpSink Representing TCP receivers, class TcpSink reacts to received TCP packets as follows: (i) Extract the sequence number (seq) from the received TCP packet, (ii) Inform the Acker object of the sequence number (seq) and the size of the TCP packet (numBytes) through function update(seq,numBytes) of class Acker, (iii) Create and send an ACK packet to the TCP sender by invoking function ack(p) of class TcpSink. The sequence number in the ACK packet is obtained from function Seqno() of the Acker object (invoked from within function ack(p)). Program 10.6 shows the declaration of a C++ class Tcpsink, which is bound to an OTcl class Agent/TCPSink. The only key variable of class TcpSink is a pointer to an Acker object, acker_ in Line 8. Two main functions of class TcpSink include recv(p,h) and ack(p). Program 10.6 Declaration of class TcpSink.
1 2 3 4 5 6 7 8 9 //~/ns/tcp/tcp-sink.cc class TcpSink : public Agent { public: TcpSink(Acker*); void recv(Packet* pkt, Handler*); int command(int argc, const char*const* argv); protected: void ack(Packet*); Acker* acker_; };

Shown in Program 10.7, function recv(p,h) is invoked by an upstream object to hand a TCP packet over to a TcpSink object. Lines 4–6 inform the Acker object, acker_, of an incoming TCP packet “pkt”. Here, the sequence number (i.e., th->seqno()) and packet size (i.e., numBytes) are passed to acker_ through this function. Again, function update(seq,numBytes) returns the number of in-order bytes which can be delivered to the application. If

10.2 TCP Receiver

229

this number is nonzero, it will be delivered to the application through function recvBytes(bytes) in Line 8. Line 9 invokes function ack(pkt) to generate an ACK packet and send it to the TCP sender. Finally, Line 10 deallocates the received TCP packet. Program 10.7 Function recv of class TcpSink.
//~/ns/tcp/tcp-sink.cc 1 void TcpSink::recv(Packet* pkt, Handler*) 2 { 3 int numToDeliver; 4 int numBytes = hdr_cmn::access(pkt)->size(); 5 hdr_tcp *th = hdr_tcp::access(pkt); 6 numToDeliver = acker_->update(th->seqno(), numBytes); 7 if (numToDeliver) 8 recvBytes(numToDeliver); 9 ack(pkt); 10 Packet::free(pkt); 11 }

Program 10.8 shows the details of function ack(p). In this function, variables whose name begins with “o” and “n” are used for an old packet and a new packet, respectively. Line 6 puts an ACK number in the ACK packet. Lines 7–8 and 9–11 conﬁgure timestamp and ﬂow ID of the ACK packet, respectively. Finally, the conﬁgured packet is sent out using function send(npkt,0) of class Agent in Line 12, where a new packet npkt is transmitted along with a Null handler. Program 10.8 Function ack of class TcpSink.
//~/ns/tcp/tcp-sink.cc 1 void TcpSink::ack(Packet* opkt) 2 { 3 Packet* npkt = allocpkt(); 4 hdr_tcp *otcp = hdr_tcp::access(opkt); 5 hdr_tcp *ntcp = hdr_tcp::access(npkt); 6 ntcp->seqno() = acker_->Seqno(); 7 double now = Scheduler::instance().clock(); 8 ntcp->ts() = now; 9 hdr_ip* oip = hdr_ip::access(opkt); 10 hdr_ip* nip = hdr_ip::access(npkt); 11 nip->flowid() = oip->flowid(); 12 send(npkt, 0); 13 }

230

10 Transport Control Protocols Part 2 – Transmission Control Protocol

10.3 TCP Sender
A TCP sender has the following four main responsibilities: • Packet transmission: Based on user demand from an application, a TCP sender creates and forwards TCP packets to a TCP receiver. • ACK processing: A TCP sender observes a received ACK pattern and determines whether transmitted packets were lost. If so, it will retransmit the lost packets. From the ACK pattern, it can also estimate the network condition (e.g., end-to-end bandwidth) and adjust the congestion window accordingly. • Timer related mechanism: A retransmission timer is used to provide connection reliability. Unless reset by an ACK packet arrival, the retransmission timer informs the TCP sender of packet loss after the packet has been transmitted for a period of Retransmission TimeOut (RTO). • Window adjustment: Based on the ACK pattern and timeout event, a TCP sender adjusts its congestion window to fully utilize network resource and prevent network congestion. The details of these four responsibilities will be discussed in the next four sections.

10.4 TCP Packet Transmission Functions
Class TcpAgent provides the following four main packet transmission functions: • sendmsg(nbytes): Send nbytes of application payload. If nbytes=-1, the payload is assumed to be inﬁnite. • sendmuch(force,reason,maxburst): out a packet whose sequence number is t_seqno_. Keep sending out packets as long as the congestion window allows and the total number of transmitted packets during a function invocation does not exceed maxburst. • output(seqno,reason): Create and send a packet with a sequence number and a transmission reason as speciﬁed by seqno and reason, respectively. • send_one(): Send a TCP packet with a sequence number t_seqno_. Among the above functions, function sendmsg(nbytes) is the only public function derived from class Agent, while the other three functions are internal to class TcpAgent. Again, function sendmsg(nbytes) is invoked by an application to inform a TcpAgent object of user demand. Function sendmsg(nbytes) does not directly send out packets. Rather, it computes the number of TCP

10.4 TCP Packet Transmission Functions

231

packets required to hold “nbytes” of data payload, and increases variable curseq_ by the computed value. In NS2, a TcpAgent object keeps transmitting TCP packets as long as the sequence number does not exceed curseq_. Increasing curseq_ is therefore equivalent to feeding data payload to a TcpAgent object. Another important variable is t_seqno_, which contains the default TCP sequence number. Unless otherwise speciﬁed, a TCP sender always transmits a TCP packet with the sequence number stored in t_seqno_. Both functions sendmuch(force,reason,maxburst) and send_one() use function output(t_seqno_,reason) to send out a TCP packet whose sequence number is t_seqno_. Function send_much(...) acts as a foundation for TCP packet transmission. In most cases, TCP agent ﬁrst stores the sequence number of packet to be transmitted in t_seqno_. Then, it invokes function send_much(...) to send TCP packets starting with that with sequence number t_seqno_ as long as the transmission window permit. As we shall see in Program 10.10, each packet transmission is carried out using function output(t_seqno_, reason). 10.4.1 Function sendmsg(nbytes) Function send_msg(nbytes) is the main data transmission interface function derived from class Agent. A user (e.g., application) informs a TCP sender of transmission demand through this function. Function sendmsg(nbytes) usually takes one input argument, nbytes, which is the amount of application payload in bytes that a user needs to send to the TCP receiver. When the user has inﬁnite demand, nbytes is speciﬁed as –1. Program 10.9 shows the details of function sendmsg(nbytes). Lines 4–7 transform the input user demand to the number of TCP packets to be Program 10.9 Function sendmsg of class TcpAgent.
//~/ns/tcp/tcp.h 1 #define TCP_MAXSEQ 1073741824 //~/ns/tcp/tcp.cc 2 void TcpAgent::sendmsg(int nbytes, const char* /*flags*/) 3 { 4 if (nbytes == -1 && curseq_ ts() = Scheduler::instance().clock(); 9 tcph->ts_echo() = ts_peer_; 10 tcph->reason() = reason; 11 tcph->last_rtt() = int(int(t_rtt_)*tcp_tick_*1000); 12 int databytes = hdr_cmn::access(p)->size(); 13 if (cong_action_ && !is_retransmit) { 14 hdr_flags* hf = hdr_flags::access(p); 15 hf->cong_action() = TRUE; 16 cong_action_ = FALSE; 17 } 18 if (seqno == 0) { 19 if (syn_) { 20 databytes = 0; 21 curseq_ += 1; 22 hdr_cmn::access(p)->size() = tcpip_base_hdr_size_; 23 } 24 } else if (useHeaders_ == true) { 25 hdr_cmn::access(p)->size() += headersize(); 26 }

10.4 TCP Packet Transmission Functions

235

Program 10.12 Function output of class TcpAgent (Cont.).
27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 } send(p, 0); ++ndatapack_; ndatabytes_ += databytes; if (seqno == curseq_ && seqno > maxseq_) idle(); if (seqno > maxseq_) { maxseq_ = seqno; if (!rtt_active_) { rtt_active_ = 1; if (seqno > rtt_seq_) { rtt_seq_ = seqno; rtt_ts_ = Scheduler::instance().clock(); } } } else { ++nrexmitpack_; nrexmitbytes_ += databytes; } if (highest_ack_ == maxseq_) force_set_rtx_timer = 1; if (!(rtx_timer_.status() == TIMER_PENDING) || force_set_rtx_timer) set_rtx_timer();

packet (with seqno =0 and syn =1) contains no pay-load, its size is set to be tcpip_base_hdr_size_ bytes (Line 22). The following TCP header ﬁelds are conﬁgured in Lines 6–12: sequence number, timestamp, timestamp echo, transmitting reason, and latest observed round trip time (RTT). Finally, function output(...) conﬁgures only the congestion ﬂag in the ﬂag header (Lines 13–16). This congestion ﬂag is set to be true if both of the following conditions are true (i.e., Line 13 is true): (i) Congestion has occurred: During network congestion, TCP sender closes the congestion window by invoking function slowdown(how), within which the variable cong_action_ is set to true. If variable cong_action_ is true, Lines 13–17 will presume that congestion has occurred. (ii) TCP sender is transmitting a new packet (is_retx = false): This ﬂag set to true, when a regular packet (not a retransmitted packets) is experiencing congestion.6
6

For example, a router in the network may drop packets marked with a congestion action ﬂag to help relieve network congestion. However, dropping a retransmitted packet may lead to TCP connection reset. Therefore, a TCP sender does not mark retransmitted packets with congestion action.

236

10 Transport Control Protocols Part 2 – Transmission Control Protocol

The third part of function output(seqno,reason) is used to send out the conﬁgured packet using function send(p,h) of class Agent in Line 27. The fourth part updates the relevant variables of the TcpAgent object in Lines 28–48. If the condition in Line 30 is true, TCP sender will no longer have data to transmit. In this case, Line 31 informs the application so by invoking function idle() of class Agent. Relevant variables to be updated are ndatapack_, ndatabytes_, nrexmitpack_, nremitbytes_, in Lines 28, 29, 42, and 43, respectively. The former two variables denote the data transmitted by the TcpAgent object in packets and bytes, while the latter two are those corresponding to the retransmitted packets only. Lines 33-39 update the related variables when seqno > maxseq_. These variables include maxseq_ and other RTT estimation variables. We will discuss about the RTT estimation later in Section 10.6 The ﬁnal part is to start the retransmission timer by invoking function set_rtx_timer() in Line 48. Note that each TCP sender has only one retransmission timer. Under a normal situation, the timer is started only when it is idle (i.e., its status is not TIMER_PENDING). However, it is also started when highest_ack_ == maxseq_, regardless of the timer’s status (see Line 47). Program 10.13 Function send one of class TcpAgent.
//~/ns/tcp/tcp.cc 1 void TcpAgent::send_one() 2 { 3 if (t_seqno_ last_ack_ yes no

sendmuch(0,0,maxburst_)

retrun

Fig. 10.3. Function recv(p,h) of class TcpAgent.

10.5.1 Function recv(p,h) Figure 10.3 and Program 10.14 show the diagram and implementation, respectively, for function recv(p,h). Function recv(p,h) pre-processes the received ACK packets in Lines 6–14, where t_seqno_ and cwnd_ are adjusted. Depending of the received ACK type (i.e., new or duplicated), Lines 6–14 (ACK pre-processing) process an ACK packet according to the following three cases: • Case I (New ACK): If a new ACK packet is received (i.e., Line 6 returns true), Line 7 will invoke function recv_newack_helper(p) to adjust congestion window (cwnd_) and prepare a new sequence number (t_seqno_) for packet transmission. • Case II (Duplicated ACK): In this case, a duplicated ACK packet is received (i.e., Line 6 returns false) but the number of duplicated ACK packets received so far has not reached numdupacks_ (i.e., Line 9 returns false). Line 12 will invoke function send_one() to transmit new TCP packets under the congestion window inﬂated by the number of received duplicated ACK packets. Note that variable sigledup_ is an NS2 option for congestion window inﬂation. The above actions are executed when singledup_ is true only. If singledup_ is false, the TCP sender will not send a new packet for every received ACK packet.

10.5 ACK Processing Functions

239

• Case III (Fast retransmit): If the received ACK is the last (i.e., numdupacks_th) duplicated ACK packet, the TCP sender will enter the Fast Retransmit phase, by invoking function dupack_action() (Line 10). Note that an option ﬂag noFastRetrans_ is an NS option for a fast retransmit phase. The TCP sender will not enter a Fast Retransmit phase, if noFastRetrans_ is true. After executing one of the above three cases, Line 17 deallocates the ACK packet *pkt by invoking function free(pkt). If the received ACK is valid (i.e., valid_ack_=1), Line 19 will create and transmit TCP packets using function send_much(0,0,maxburst_). Here a received ACK packet is said to be valid if it is a new ACK packet (i.e., tcph->seqno() > last_ack_) or a duplicated ACK (i.e., tcph->seqno() = last_ack_). If an ACK packet is invalid, a TCP sender will only destroy the ACK packet, but will not create and forward new packets. Program 10.14 Function recv of class TcpAgent.
//~/ns/tcp/tcp.cc 1 void TcpAgent::recv(Packet *pkt, Handler*) 2 { 3 hdr_tcp *tcph = hdr_tcp::access(pkt); 4 int valid_ack = 0; 5 ++nackpack_; 6 if (tcph->seqno() > last_ack_) { 7 recv_newack_helper(pkt); 8 } else if (tcph->seqno() == last_ack_) { 9 if (++dupacks_ == numdupacks_ && !noFastRetrans_) { 10 dupack_action(); 11 } else if (dupacks_ < numdupacks_ && singledup_ ) { 12 send_one(); 13 } 14 } 15 if (tcph->seqno() >= last_ack_) 16 valid_ack = 1; 17 Packet::free(pkt); 18 if (valid_ack) 19 send_much(0, 0, maxburst_); 20 }

10.5.2 Function recv newack helper(pkt) Function recv_newack_helper(pkt) is a helper function invoked when a new ACK packet is received. As shown in Program 10.15, the function recv_newack_helper(pkt) ﬁrst invokes function newack(pkt) in Line 2 to update relevant variables and to process the retransmission timer. When Explicit Congestion Notiﬁcation (ECN) is not enabled (i.e., by default ECT

240

10 Transport Control Protocols Part 2 – Transmission Control Protocol

(ECN Capable Transport System) is set to zero), Line 5 will open the congestion window (by invoking function opencwnd()) when at least one of the following conditions is true (Line 4): • control_increase_ = 0: Variable control_increase_, when set to 1, suppresses the congestion window opening. When control_increase_ is zero, a TCP sender can freely increase the congestion window. Program 10.15 Function recv newack helper of class TcpAgent.
//~/ns/tcp/tcp.cc void TcpAgent::recv_newack_helper(Packet *pkt) { newack(pkt); if (!ect_) { if (!control_increase_ || (control_increase_ && (network_limited() == 1))) 5 opencwnd(); 6 } 7 if ((highest_ack_ >= curseq_-1) && !closed_) { 8 closed_ = 1; 9 finish(); 10 } 11 } 1 2 3 4

• control_increase_ = 0 and network is limited: When control_ increase_ is 1, the TCP sender is allowed to open the congestion window only when the previous congestion window is not suﬃcient to transmit the current packet (i.e., the network is limited). In NS2, a network is said to be limited when t_seqno_ is less than prev_highest_ack_ + win, where prev_highest_ack_ is the ACK number prior to the reception of the current ACK packet and win is the current congestion window (see Program 10.16). In this case, it is necessary to open the congestion window, even if control_increase_ is enabled. Note that if the TCP sender stops transmission due to any reason other than the reason that the network is limited, function recv_newack_helper(pkt) will not open the congestion window. Program 10.16 Function network limited of class TcpAgent.
//~/ns/tcp/tcp.cc 1 int TcpAgent::network_limited() { 2 int win = window () ; 3 if (t_seqno_ > (prev_highest_ack_ + win)) 4 return 1; 5 else 6 return 0; 7 }

10.5 ACK Processing Functions

241

Finally, if the TCP sender no longer has data to transmit, Line 8 in Program 10.15 will close the connection by setting closed_ to 1, and Line 9 will invoke function finish(). 10.5.3 Function newack(pkt) Program 10.17 shows the details of function newack(pkt). Lines 5–10 update variables dupack_, last_ack_, prev_highest_ack_, highest_ack_, and t_seqno_. Lines 12–19 update RTT estimation variables and timeout backoﬀ value. Finally, Line 20 starts a retransmission timer for the transmitting packet. Again, we will discuss the details of RTT estimation and retransmission timer later in Section 10.6. Program 10.17 Function newack of class TcpAgent.
//~/ns/tcp/tcp.cc 1 void TcpAgent::newack(Packet* pkt) 2 { 3 double now = Scheduler::instance().clock(); 4 hdr_tcp *tcph = hdr_tcp::access(pkt); 5 dupacks_ = 0; 6 last_ack_ = tcph->seqno(); 7 prev_highest_ack_ = highest_ack_ ; 8 highest_ack_ = last_ack_; 9 if (t_seqno_ < last_ack_ + 1) 10 t_seqno_ = last_ack_ + 1; 11 hdr_flags *fh = hdr_flags::access(pkt); 12 if (rtt_active_ && tcph->seqno() >= rtt_seq_) { 13 if (!ect_) { 14 t_backoff_ = 1; 15 ecn_backoff_ = 0; 16 } 17 rtt_active_ = 0; 18 rtt_update(now - rtt_ts_); 19 } 20 newtimer(pkt); 21 }

Function dupack action() The main responsibilities of function dupack_action() are to: (1) decrease congestion window size, (2) set t_seqno_ to the sequence number of the lost TCP packet, and (3) restart retransmission timer. Program 10.18 shows the details of function dupack_action(). Line 5 registers fast retransmission event (i.e., FAST_RETX) for tracing. Line 6 records CWND_ACTION_DUPACK as the

242

10 Transport Control Protocols Part 2 – Transmission Control Protocol

latest window adjustment action (i.e., last_cwnd_action_). Line 7 closes the congestion window by invoking function slowdown( CLOSE_SSTHRESH_HALF | CLOSE_CWND_ONE), feeding how the slow start threshold and congestion window are to be conﬁgured as an input argument. Finally, Line 8 invokes function reset_rtx_timer(0,0) to set t_seqno_ to highest_ack_+1, and restarts the retransmission timer. The details of functions reset_rtx_timer(...) and slowdown(...) will be discussed in Sections 10.6 and 10.7, respectively. Program 10.18 Function dupack action of class TcpAgent.
//~/ns/tcp/tcp.cc 1 void TcpAgent::dupack_action() 2 { 3 if (highest_ack_ > recover_) { 4 recover_ = maxseq_; 5 trace_event("FAST_RETX"); 6 last_cwnd_action_ = CWND_ACTION_DUPACK; 7 slowdown(CLOSE_SSTHRESH_HALF|CLOSE_CWND_ONE); 8 reset_rtx_timer(0,0); 9 return; 10 }

TCP Tahoe reacts to a duplicated ACK packet diﬀerently. Lines 4–9 in Program 10.18 are executed only when all the packets transmitted during the previous packet loss have been acknowledged. Here, variable recover_ records the highest TCP sequence number (i.e., maxseq_) transmitted during the previous packet loss event. Line 4 sets recover_ to be maxseq_ so that it can be used in the next packet loss event. The condition in Line 3, highest_ack_ > recover_, implies that the TCP packet with highest sequence number transmitted during the previous loss must be acknowledged. If this condition is not satisﬁed, the TCP sender will wait for timeout and retransmit the lost packet.

10.6 Timer Related Functions
Another responsibility of a TCP sender is to use a retransmission timer to provide connection reliability. The main components of this part include estimation of smoothed RTT (round trip time) and RTT variation, computation of RTO (retransmission timeout), implementation of BEB (binary exponential backoﬀ), utilization of a retransmission timer, and deﬁning actions to be performed at timeout. 10.6.1 RTT Sample Collection A TCP sender needs to collect RTT samples in order to estimate smoothed RTT and RTT variation, and to compute retransmission timeout (RTO) value.

10.6 Timer Related Functions output(seqno,reason) 243

ACTIVE

dupack_action() reset_rtx_timer(mild, backoff)

INACTIVE

newack(pkt) & seqno = rtt_seq _

RTT Sample = now – rtt_ts_

Fig. 10.4. The RTT sampling process.

An RTT sample is measured as the time diﬀerence between the point where a packet is transmitted and the point where the associated ACK packet arrives the sender. In NS2, each TCP sender has only one set of variables including variables rtt_active_,rtt_ts_, and rtt_seq_ (see Table 10.1) to track RTT samples. It can collect only one RTT sample at a time – meaning not all the packets are used to collect RTT samples. Figure 10.4 shows the diagram of the RTT collection process. The process starts in the inactive state where rtt_active_=0. The collection is activated (i.e., the process enters the active state) when a TCP sender sends out a new packet using function output(seqno,reason). From Program 10.2, Line 35 sets rtt_active_ to be 1.7 Lines 37 and 38 record the TCP sequence number and the current time in the variables rtt_seq_ and rtt_ts_, respectively. An RTT sample is collected when the associated ACK packet returns (see Lines 12–19 of function newack(pkt) in Program 10.17). Given that the collection process is active (i.e., rtt_active_=1), Line 12 determines whether the incoming ACK packet belongs to the same collecting sample. It is so if the sequence number in the received ACK packet is the same as that stored in rtt_seq_ (set at the beginning of the collecting process). Note that the logical relation here is “>=” rather than “==”, since some TCP variants may not generate an ACK packet for every received TCP packet. At the end of the collection process, Line 17 sets rtt_active to zero indicating that the collecting process has completed (i.e., the process moves back to inactive state), and Line 18 takes an RTT sample by invoking rtt_update(now-rtt_ts_) (deﬁned in Program 10.22). The above RTT collection process operates fairly well under normal situations. However, a packet loss may inﬂate an RTT sample, and aﬀect the collecting accuracy. In this case, the measured RTT would be the RTT value plus the time used to retransmit the lost packets. To avoid complication, NS2 simply cancels the RTT collection process, when a packet loss
7

If the rtt active is nonzero, TCP sender will skip the collection process.

244

10 Transport Control Protocols Part 2 – Transmission Control Protocol

occurs. In particular, functions dupack_action() (Line 8 in Program 10.18) and timeout(tno) (Lines 14 and 16 in Program 10.26) invoke function reset_rtx_timer(...) to set rtt_active_ to zero, essentially cancelling the RTT collecting process. 10.6.2 RTT Estimation After collecting an RTT sample, a TCP sender feeds a sample “tao” to function rtt_update(tao) to estimate smoothed RTT (t_srtt_), RTT variation (t_rttvar_), and unbounded RTO (t_rtxcur_)8 based on Eqs. (9.1)–(9.3), where α = 7/8, β = 3/4 and γ = 1. Instead of directly computing these three variables, NS2 manipulates Eqs. (9.1)–(9.3) such that each term in these equations is multiplied with 2n , where n is an integer. As discussed in Section 12.4.2, multiplication and division by 2n can be implemented in C++ by shifting the binary value to the left and right, respectively, by n bits. This bit shifting technique is used in function rtt_update(tao) to compute t_srtt_, t_rttvar_, and t_rtxcur_. At time k, let t(k) be the RTT sample, t(k) be the smoothed RTT value, σt (k) be the RTT variation, and Δ refer to t(k + 1) − t(k). From (9.1)–(9.3), 1 7t(k) + t(k + 1) 8 1 7t(k) + t(k) + t(k + 1) − t(k) = 8 1 8t(k) + Δ = 8 1 σt (k + 1) = (3σt (k) + |Δ|) 4 1 = (3σt (k) − 4σt (k) + 4σt (k) + |Δ|) 4 1 = (−σt (k) + 4σt (k) + |Δ|) 4 RT Ou (k + 1) = γ × [t(k + 1) + 4σt (k + 1)] t(k + 1) =

(10.1)

(10.2) (10.3)

where RT Ou (k + 1) is an unbounded RTO. Equations (10.1)–(10.3) are now arranged in the multiple of 2n , n = {0, 2, 3} (i.e., the multiple of 1, 2, and 4). NS2 uses bit shifting operation in place of multiplication to implement Eqs. (10.1)–(10.3). 10.6.3 Overview of State Variables State variables contain the current status of a TCP agent. Related timer state variables are shown in Tables 10.1 and 10.3. Most the variables are well
8

An actual value of RTO must be bounded by a minimum and a maximum value.

10.6 Timer Related Functions

245

explained by their description. We now discuss a few points related to these variables. First, C++ timer variables are initialized in function rtt_init() (Lines 1–8 in Program 10.19). OTcl timer instvars, on the other hand, are initialized in ﬁle ˜ns/tcl/lib/ns-default.tcl shown in Lines 9–19 of Program 10.19. Program 10.19 Default values for the timer-related variables.
1 2 3 4 5 6 7 8 //~/ns/tcp/tcp.cc void TcpAgent::rtt_init() { t_rtt_ = 0; t_srtt_ = int(srtt_init_ / tcp_tick_) T_SRTT_BITS); 9 if ((t_srtt_ += delta) > T_RTTVAR_BITS); 14 if ((t_rttvar_ += delta) T SRTT BITS 8 t rttvar σt = = t rttvar >>T RTTVAR BITS 4 Δ = t rtt − t = t rtt − (t srtt >>T SRTT BITS) t=

(10.4) (10.5) (10.6)

where T_SRTT_BITS, T_RTTVAR_BITS, and rttvar_exp_ are deﬁned in Program 10.19 as 3, 2, and 2, respectively. Again, variables t_srtt_ and t_rttvar_ are stored in multiples of 8 and 4 (see Lines 4–5 in Program 10.19). Therefore, their relationship to actual smoothed RTT (t) and RTT variation (σt ) is given by Eqs. (10.4) and (10.5), respectively. Based on the above variables, Lines 8–15 compute the smoothed RTT value. In Eqs. (10.1) and (10.2), we rearrange the variables t_srtt_ and t_rttvar_ as follows: t srtt (k + 1) = 8t(k + 1) = 8t(k) + Δ(k) = t srtt (k) + Δ(k) t rttvar (k + 1) = 4σt (k + 1) = |Δ| − σt (k) + 4σt (k) = |Δ| − [t rttvar >>T SRTT BITS] (k) + t rttvar (k). (10.8) (10.7)

250

10 Transport Control Protocols Part 2 – Transmission Control Protocol

In Program 10.22, Line 8 computes delta (i.e., Δ) as indicated in (10.6). Line 9 updates t_srtt_ according to Eq. (10.7) and Lines 11–12 compute |Δ|. Lines 13–14 update t_rttvar_ according to Eq. (10.8). From Lines 9–10 and 14–15, both t_srtt_ and t_rttvar_ will be set to 1, if their updated values are less than zero. Also, Lines 8–15 are invoked when t_srtt_ is nonzero only. When t_srtt_ is zero, t_srtt_ and t_rttvar_ are simply set to 8 times (Line 17) and twice of (Line 18) the RTT sample (i.e., t_rtt_), respectively. NS2 computes (using Eq. (10.3)) and stores the unbounded value of RTO in variable t_rtxcur_ (Line 20). It is computed as t + 4σt shown in Eq. (9.3). The upper-bound and the lower-bound in Eq. (9.3) will be implemented when an unbounded RTO is assigned to the retransmission timer (e.g., in function rtt_timeout()). The computation of t_rtxcur_ in Line 20 consists of 4 steps: (i) Scale t_rttvar_: Variables t_srtt_ and t_rttvar_ are stored as multiples of 2T SRTT BITS = 8 and 2T RTTVAR BITS = 4, respectively. Line 20 converts the scale of t_rtt_var_ into the same scale of t_srtt_ as follows: t rttvar → t rttvar × 8 4 = t rttvar >>T RTTVAR BITS tcph->seqno() || tcph->seqno() < maxseq_) 5 set_rtx_timer(); 6 else 7 cancel_rtx_timer(); 8 }

10.6.11 Function timeout(tno) Function timeout(tno) is invoked when a retransmission timer expires. It adjusts congestion window as well as slow start threshold, and retransmits the lost packet. Again, function expire(e) is invoked when the timer expires. From Line 10 in Program 10.20, function expire(e) of class RtxTimer simply invokes function timeout(TCP_TIMER_RTX) of the associated TcpAgent object. As shown in Lines 1–19 of Program 10.26, function timeout(tno) takes a timer option (tno) as an input argument, where the possible values of tno are deﬁned in Lines 20–25 of Program 10.26. In this section, we are interested in TCP Tahoe. Therefore, we will discuss the case where only timeout(TCP_TIMER_RTX) is invoked. The basic operation of function timeout(tno) is to close the congestion window (in Line 10), restart the retransmission timer (in Lines 14 and 16), and retransmits the lost packet (in Line 18). We will discuss the details of function slowdown(...) which closes the congestion window in Section 10.7. The retransmission timer is restarted by using the function reset_rtx_timer(mild,backoff) (see Program 10.24). For zero value of “mild” this function sets t_seqno_ to highest_ack_+1. The zero and nonzero values of the second input argument “backoff” inform function reset_rtx_timer(mild,backoff) to and not to (respectively) update the binary exponential backoﬀ multiplicative factor (t_backoff_). Again, the

254

10 Transport Control Protocols Part 2 – Transmission Control Protocol

Program 10.26 Function timeout of class TcpAgent and the possible values of its input argument tno.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 //~/ns/tcp/tcp.cc void TcpAgent::timeout(int tno) { ... if (cwnd_ < 1) cwnd_ = 1; if (highest_ack_ == maxseq_ && !slow_start_restart_) { } else { recover_ = maxseq_; if (highest_ack_ < maxseq_) { ++nrexmit_; last_cwnd_action_ = CWND_ACTION_TIMEOUT; slowdown(CLOSE_SSTHRESH_HALF|CLOSE_CWND_RESTART); } } if (highest_ack_ == maxseq_) reset_rtx_timer(0,0); else reset_rtx_timer(0,1); last_cwnd_action_ = CWND_ACTION_TIMEOUT; send_much(0, TCP_REASON_TIMEOUT, maxburst_); } //~/ns/tcp/tcp.h #define TCP_TIMER_RTX #define TCP_TIMER_DELSND #define TCP_TIMER_BURSTSND #define TCP_TIMER_DELACK #define TCP_TIMER_Q #define TCP_TIMER_RESET

20 21 22 23 24 25

0 1 2 3 4 5

TCP sender assumes that all packets with sequence number lower than highest_ack_ are successfully transmitted. At a timeout event, it assumes that the ﬁrst lost packet (i.e., the packet to be retransmitted) is the packet with sequence number highest_ack_+1. After preparing t_seqno_ (i.e., set to highest_ack_+1) for retransmission, Line 18 invokes function send_much(0, TCP_REASON_TIMEOUT, maxburst_) to transmit the lost packet. After a TCP sender transmits all the packets provided by an attached application, its variable t_seqno_ is equal to variable curseq_, and variable maxseq_ stops increasing. After the last packet (with sequence number maxseq_) is acknowledged, variable highest_ack_ is equal to maxseq_. At this point, the TCP sender enters an idle state. Its retransmission timer, however, does not stop at this moment. It keeps expiring for every period of RTO. From Line 14 of Program 10.26, function timeout(tno) will invoke reset_rtx_timer(0,0), which stores the value of highest_ack_+1 in vari-

10.7 Window Adjustment Functions

255

able t_seqno_ but does not change the multiplicative factor t_backoff_. Also, function send_much(0, TCP_REASON_TIMEOUT, maxburst_) will not send out any packet since t_seqno is not less than curseq_ (see Program 10.10). When the application sends more user demand (i.e., data payload) by invoking sendmsg(nbytes), variable curseq_ is incremented and the TCP connection becomes active. In this case, function send_much(0,0,maxburst_) will send out packets, starting with the packet with sequence number t_seqno_= max_seq_+1 = highest_ack_ + 1. There are two important details in function timeout(tno). One is that regardless of whether connection is busy or idle, Line 17 sets the variable last_cwnd_action_ which records the latest action imposed on the congestion window to be CWND_ACTION_TIMEOUT. Another is related to variable recover_. Recall that recover contains the highest sequence number among all the transmitted TCP packets at the latest loss event (i.e., either timeout or Fast Retransmit). Line 6 hence records the highest TCP sequence number transmitted so far in the variable recover_.

10.7 Window Adjustment Functions
From Section 9.1.2, a TCP sender dynamically adjusts congestion window to fully utilize the network resource. When the network is under utilized, a TCP sender increases transport-level transmission rate by opening the congestion window. In the slow start phase, where the congestion window (cwnd_) is less than the slow start threshold (ssthresh_), a TCP sender increases the congestion window by one for every received ACK packet. If cwnd_ is not less than ssthresh_, on the other hand, a TCP sender will be in the congestion avoidance phase, and the congestion window is increased by 1/cwnd_ for every received ACK packet. When the network is congested, a TCP sender closes the congestion window to help relieve network congestion. As discussed in Section 9.1.2, TCP may decrease the window by half or may reset the congestion window size to one, depending on the situation. Class TcpAgent provides two main functions, which can be used to adjust the congestion window: • opencwnd(): Increases the size of the congestion window. The increasing method depends on cwnd_ and ssthresh_. • slowdown(how): Decreases the size of the congestion window by the method speciﬁed in “how”. The possible values of “how” are deﬁned in Program 10.27. All possible values of how contain 32 bits, and conform to the following format: 1 of “one” bit and 31 of “zero” bits. The diﬀerence among the values deﬁned in Program 10.27 lies in the position of the “one” bit. This format acts as a simple

256

10 Transport Control Protocols Part 2 – Transmission Control Protocol

identiﬁcation of the input method “how” through an “AND” operator. For example, suppose the input argument how is set to CLOSE_CWND_ONE (=2). Let x be a variable which can be any value in Program 10.27. Then, how & x would be nonzero if and only if x=CLOSE_CWND_ONE. This assignment is also able to contain several slowdown methods in one variable using an “OR” operator. For example, let how be CLOSE_CWND_ONE|CLOSE_SSTHRESH_HALF. Then, how & x would be nonzero if and only if x=CLOSE_CWND_ONE or x=CLOSE_SSTHRESH_ HALF. Program 10.27 Possible values of how – the input argument of function slowdown.
1 2 3 4 5 6 7 8 9 10 11 12 //~/ns/tcp/tcp.h #define CLOSE_SSTHRESH_HALF #define CLOSE_CWND_HALF #define CLOSE_CWND_RESTART #define CLOSE_CWND_INIT #define CLOSE_CWND_ONE #define CLOSE_SSTHRESH_HALVE #define CLOSE_CWND_HALVE #define THREE_QUARTER_SSTHRESH #define CLOSE_CWND_HALF_WAY #define CWND_HALF_WITH_MIN #define TCP_IDLE #define NO_OUTSTANDING_DATA 0x00000001 0x00000002 0x00000004 0x00000008 0x00000010 0x00000020 0x00000040 0x00000080 0x00000100 0x00000200 0x00000400 0x00000800

10.7.1 Function opencwnd() Function opencwnd() is invoked when a new ACK packet is received (see function recv_newack_helper() in Line 5 of Program 10.15). It opens the congestion window, and allows the TCP sender to transmit more packets without waiting for acknowledgement. Program 10.28 shows the details of function opencwnd(). From Line 3, if cwnd_ is less than ssthresh_, the TCP sender will be in the slow start phase and cwnd_ will be increased by 1. Otherwise, the TCP sender must be in a congestion avoidance phase, and cwnd_ will be increased by 1/cwnd_ (Lines 6–7), where increase_num_ is usually set to 1. In both cases, Lines 9–10 bound cwnd_ within maxcwnd_, the predeﬁned maximum congestion window size. 10.7.2 Function slowdown(how) Function slowdown(how) closes the congestion window based on the method speciﬁed in the input argument how. It is invoked from within function dupack_action() and timeout(tno) to decrease transport layer transmission

10.7 Window Adjustment Functions

257

Program 10.28 Function opencwnd of class TcpAgent.
//~/ns/tcp/tcp.cc 1 void TcpAgent::opencwnd() 2 { 3 if (cwnd_ < ssthresh_) { 4 cwnd_ += 1; 5 } else { 6 double increment = increase_num_ / cwnd_; 7 cwnd_ += increment; 8 } 9 if (maxcwnd_ && (int(cwnd_) > maxcwnd_)) 10 cwnd_ = maxcwnd_; 11 }

rate. Function dupack_action() invokes function slowdown(how) feeding how = CLOSE_SSTHRESH_HALF | CLOSE_CWND_ONE (Line 7 in Program 10.18) as an input argument. From Program 10.29, this invocation halves the current slow start threshold (Lines 9–13) and resets the congestion window to 1 (Line 26). Function timeout(tno) on the other hand invokes function slowdown(how) with an input argument how = CLOSE_SSTHRESH_HALF | CLOSE_CWND_RESTART as an input argument (Line 10 in Program 10.26). From Program 10.29, this invocation halves the current slow start threshold (Lines 9–13) and resets the congestion window to a prediﬁned window-restart value (Line 24). In both cases, NS2 employs an “OR” operator to combine how to adjust slow start threshold and how to adjust congestion window, and feed it as an input argument to function slowdown(how). The details of function slowdown(how) are shown in Program 10.29. In this function, Lines 4–6 ﬁrst set a variable slowstart to zero and one when TCP is in the slow start phase (i.e., cwnd_< ssthresh_) and in the congestion avoidance phase (i.e., cwnd_>= ssthresh_), respectively. Line 7 stores half of the window size in a variable halfwin and the window size in a variable win. Variable decrease_num_ in Line 8 is set to 0.5 by default. Therefore, the local variable decreasewin is half of the current congestion window. The variable decrease_num_ provides an option for window decrement, where diﬀerent TCP variants may set the value of decrease_num_ diﬀerently (e.g., 0.3, 0.7). Lines 9–26 show diﬀerent window closing method, which will be invoked according to the input argument “how”. Line 27 ensures that the minimum slow start threshold is 2. Line 29 sets the variable cong_action_ to be true if the window adjustment method, how, is either of CLOSE_CWND_HALF, CLOSE_CWND_RESTART, CLOSE_CWND_INIT, or CLOSE_CWND_ONE. Again, the variable cong_action_ is used in function output(seqno,reason) to set the congestion ﬂag of the transmitted packet. Finally, Line 32 sets first_decrease_ to zero, indicating TCP has decreased the congestion window at least once.

258

10 Transport Control Protocols Part 2 – Transmission Control Protocol

Program 10.29 Function slowdown of class TcpAgent.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 } //~/ns/tcp/tcp.cc void TcpAgent::slowdown(int how) { double win, halfwin, decreasewin; int slowstart = 0; if (cwnd_ < ssthresh_) slowstart = 1; halfwin = windowd() / 2; win = windowd(); decreasewin = decrease_num_ * windowd(); if (how & CLOSE_SSTHRESH_HALF) if (first_decrease_ == 1||slowstart || last_cwnd_action_ == CWND_ACTION_TIMEOUT) ssthresh_ = (int) halfwin; else ssthresh_ = (int) decreasewin; else if (how & THREE_QUARTER_SSTHRESH) if (ssthresh_ < 3*cwnd_/4) ssthresh_ = (int)(3*cwnd_/4); if (how & CLOSE_CWND_HALF) if (first_decrease_==1||slowstart||decrease_num_==0.5){ cwnd_ = halfwin; } else cwnd_ = decreasewin; else if (how & CWND_HALF_WITH_MIN) { cwnd_ = decreasewin; if (cwnd_ < 1) cwnd_ = 1; } else if (how & CLOSE_CWND_RESTART) cwnd_=int(wnd_restart_); else if (how & CLOSE_CWND_INIT) cwnd_ = int(wnd_init_); else if (how & CLOSE_CWND_ONE) cwnd_ = 1; if (ssthresh_ < 2) ssthresh_ = 2; if (how & (CLOSE_CWND_HALF|CLOSE_CWND_RESTART| CLOSE_CWND_INIT|CLOSE_CWND_ONE)) cong_action_ = TRUE; if (first_decrease_ == 1) first_decrease_ = 0;

Lines 9–15 adjust the slow start threshold (ssthresh_) based on the value of “how”: • CLOSE_SSTHRESH_HALF (Lines 11 and 13): Sets the slow start threshold ssthresh_ to the half of the current congestion window size cwnd_. • THREE_QUARTER_SSTHRESH (Line 15): Sets the slow start threshold ssthresh_ to at least 3/4 of its current value.

10.8 Chapter Summary

259

Similarly, Lines 16–29 adjust the congestion window (cwnd_) based on the value of “how”: • CLOSE_CWND_HALF (Lines 17–20): Decreases the current congestion window size (i.e., cwnd_) by half (i.e., either halfwin or decreasewin). • CWND_HALF_WITH_MIN (Lines 22–23): Sets the current congestion window size to decreasewin but not less than 1. • CLOSE_CWND_RESTART (Line 24): Sets the current congestion window size to the prediﬁned window-restart value wnd_restart_. • CLOSE_CWND_INIT (Line 25): Sets the current congestion window size to wnd_init_ (i.e., initial value of congestion window size). • CLOSE_CWND_ONE (Line 26): Sets the current congestion window size to 1.

10.8 Chapter Summary
TCP is a reliable connection-oriented transport layer protocol. It provides a connection with end-to-end error control and congestion control. NS2 implements TCP senders and TCP receivers using C++ classes TcpAgent and TcpSink, which are bound to OTcl classes Agent/TCP and Agent/TCPSink, respectively. A TCP sender has four main responsibilities. First, based on user demand, it creates and forwards packets to a TCP receiver. Secondly, it provides an end-to-end connection with reliability by means of packet retransmission. Thirdly, it implements timer-related components to estimate round trip time (RTT) and retransmission timeout (RTO), used to determine whether a packet is lost. Finally, it dynamically adjusts transport-level transmission rate to fully utilize the network resource without causing network congestion. ATCP receiver is responsible for creating (cumulative) ACK packets and forwards them back to the TCP sender.

11 Application: User Demand Indicator

Operating on top of a transport layer agent, an application models user demand for data transmission. A user is assumed to create bursts of data payload or application packets. These payload bursts are transformed into transport layer packets which are then forwarded to a transport layer receiving agent. Applications can be classiﬁed into traﬃc generator and simulated application. A traﬃc generator creates user demand based on a predeﬁned schedule. A simulated application, on the other hand, creates the demand as if the application is running. In the followings, we ﬁrst discuss the relationship between an application and a transport layer agent in Section 11.2. Sections 11.3 and 11.4 discuss the detailed implementation of traﬃc generators and simulated applications, respectively. Finally, the chapter summary is given in Section 11.5.

11.1 Relationship Between an Application and a Transport Layer Agent
From time to time, an application needs to exchange user demand information with a transport layer agent. An application declares a pointer agent_ to an attached agent. Similarly, an agent deﬁnes a pointer app_ to an attached application. The user demand information is exchanged between an application and an agent through these two pointers. Section 9.2.2 gives a four-step agent conﬁguration method, which binds an application and a transport layer agent together. The details of these four steps are given below: Step 1: Create a Sending Agent, a Receiving Agent, and an Application An agent and an application can be created by using instproc new{..} as follows:

T. Issariyakul, E. Hossain, Introduction to Network Simulator NS2, DOI: 10.1007/978-0-387-71760-9 11, c Springer Science+Business Media, LLC 2009

262

11 Application: User Demand Indicator

set agent [new Agent/] set app [new Application/] where and denote the type of an agent (e.g., TCP or UDP) and an application (e.g. Traffic/CBR or FTP), respectively. Step 2: Connect an Agent to an Application A two way connection between an application and an agent can be created by using of class Application whose syntax is shown below: $app attach-agent $agent where $app and $agent are Application and Agent objects. The details of instproc attach-app{s_type} are shown in Program 11.1. Line 7 stores an input Agent object in the variable agent_. Line 12 invokes function attachApp(this) of class Agent, while Lines 19–22 create a connection from the Agent object to the Application object. From Line 21, function Program 11.1 An OTcl command attach-agent of class Application and function attachApp of class Agent.
//~/ns/apps/app.cc 1 int Application::command(int argc, const char*const* argv) 2 { 3 Tcl& tcl = Tcl::instance(); 4 ... 5 if (argc == 3) { 6 if (strcmp(argv[1], "attach-agent") == 0) { 7 agent_ = (Agent*) TclObject::lookup(argv[2]); 8 if (agent_ == 0) { 9 tcl.resultf("no such agent %s", argv[2]); 10 return(TCL_ERROR); 11 } 12 agent_->attachApp(this); 13 return(TCL_OK); 14 } 15 ... 16 } 17 return (Process::command(argc, argv)); 18 } //~/ns/common/agent.cc 19 void Agent::attachApp(Application *app) 20 { 21 app_ = app; 22 }

11.1 Relationship Between an Application and a Transport Layer Agent

263

attachApp(app) stores an input Application object app in the variable app_ of the Agent object. Since Line 12 feeds the pointer this to function attachApp(...) of the Application object, which simply sets the pointer agent_->app_ to point to itself. Step 3: Attaching an Agent to a Low-Level Network Here, we consider the case where an agent is connected to a node in a low-level network. As discussed in Section 6.6.3, an agent can be attached to a node by using instproc attach-agent{node agent} of class Simulator, where node and agent are the Node, and Agent objects, respectively. This instproc creates a two-way connection between a Node object node and an Agent object agent. It sets variable agent::target_ to point to node and installs agent in the demultiplexer (i.e., instvar dmux_) of node. The process of attaching an agent to a node involves three OTcl classes: Simulator, Node, and RtModule. Figure 11.1 shows the main operation when “$ns attach-agent $node $agent” is invoked: (i) Instproc attach-agent{node agent} of class Simulator invokes $node attach $agent. (ii) Instproc attach{agent} of class Node allocates a port for an input agent $agent, conﬁgures instvar agent_addr_ and agent_port_ of the input
$ns attach-agent $node $agent Class Simulator (i) Instproc attach-agent {node agent} $node attach $agent Class Node (ii) Instproc attach {agent port} - set port [$dmux_ alloc-port [[Simulator instance] nullagent]] - $agent set agent_addr_ [AddrParams addr2id $address_] - $agent set agent_port_ $port - $self add-target $agent $port (iii) Instproc add-target{agent port} foreach m [$self set ptnotif_] { $m attach $port $agent } Class RtModule (iv) Instproc attach {agent port} - $agent target [[$self node] entry] - [[$self node] dmux] install $port $agent

Fig. 11.1. Internal mechanism of instproc attach-agent{node agent} of class Simulator.

264

11 Application: User Demand Indicator

agent $agent, and invokes instproc add-target{agent port} to inform every routing module stored in the instvar ptnotif_. (iii) Instproc add-target{agent port} of class Node invokes instproc attach{ agent port} of each routing module (of class RtModule) stored in the instvar ptnotif_. (iv) Instproc attach{agent port} of class RtModule creates a connection between a node and an agent. Here, $agent sets its $target_ to point to the entry of $node, while $node installs $agent in the slot “port” of its demultiplexer “dmux_”. This connection is created for both sending and receiving agents. Step 4: Associating a Sending Agent with a Receiving Agent To associate a sending agent with a receiving agent, NS2 employs instproc connect of class Simulator, whose syntax is shown below: $ns connect $s_agent $r_agent where $ns, $s_agent, and $r_agent are Simulator, sending Agent, and receiving Agent objects, respectively. Program 11.2 shows the details of instproc connect{src dst}. Lines 3 and 4 invoke instproc simplex-connect{src dst}, which set up a connection from src to dst_1, and simplex-connect{dst src} which creates a connection from dst back to src. Program 11.2 Instprocs connect and simplex-connect of class Simulator.
1 2 3 4 5 6 //~/ns/tcl/lib/ns-lib.tcl Simulator instproc connect {src dst} { ... $self simplex-connect $src $dst $self simplex-connect $dst $src ... }

7 Simulator instproc simplex-connect { src dst } { 8 ... 9 $src set dst_addr_ [$dst set agent_addr_] 10 $src set dst_port_ [$dst set agent_port_] 11 ... 12 }

1

From Table 9.3, instvars dst addr and dst port are bound to the C++ variables dst ::addr and dst ::port , respectively, in the C++ domain.

11.2 Details of Class Application

265

Instvars dst_addr_ and dst_port_ are conﬁgured in Lines 9–10. When an agent creates a packet, it stores values in variables dst_.addr_ and dst_.port_ in the packet header. During a packet forwarding process, a lowlevel network delivers packets to the agent corresponding to whose address and port are speciﬁed in the packet header.

11.2 Details of Class Application
An application is deﬁned in a C++ class Application as shown in Program 11.3. Class Application has only one key variable agent_ which is a pointer to class Agent. Other two variables, enableRecv_ and enableResume_, are ﬂag variables, which indicate whether an Application object should react to functions recv(nbytes) and resume(), respectively. These two ﬂag variables are set to zero by default. Program 11.3 Declaration of class Application.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 //~/ns/apps/app.h class Application : public Process { public: Application(); virtual void send(int nbytes); virtual void recv(int nbytes); virtual void resume(); protected: virtual int command(int argc, const char*const* argv); virtual void start(); virtual void stop(); Agent *agent_; int enableRecv_; int enableResume_; };

11.2.1 Functions of Classes Application and Agent After their connection is created, an application and an agent may invoke public functions of each other through the pointers agent_ and app_, respectively. The key public functions of class Application include functions send(nbytes), recv(nbytes), and resume(), while those of class Agent are functions send(nbytes), sendmsg(nbytes), close(), listen(), and set_pkttype(pkttype). Apart from these public functions, class Application also provides protected functions start() and stop() to start and stop an Application

266

11 Application: User Demand Indicator

object, respectively. Finally, there are ﬁve key OTcl commands for class Application which can be invoked from the OTcl domain: start{}, stop{}, agent{}, send{nbytes}, and attach-agent{agent}. 11.2.2 Public Functions of Class Application Program 11.4 shows the details of the three following public functions of class Application: • send(nbytes): Inform the attached transport layer agent that a user needs to send nbytes of data payload. Line 3 sends the demand to the attached agent by executing “agent_->sendmsg(nbytes)”. • recv(nbytes): Receive “nbytes” bytes from a receiving transport layer agent. A UDP agent speciﬁes nbytes as the number of bytes in a received packet. In case of UDP, nbytes is equal to packet size; on the other hand, TCP speciﬁes “nbytes” as the number of in-sequence received bytes. Due to possibility of out-of-order packet delivery, nbytes can be greater than the size of one packet. • resume(): Invoked by a sending agent, this function indicates that the agent has sent out all data corresponding to the user demand. For a TCP Program 11.4 Implementation of functions send, recv, and resume of class Application.
1 2 3 4 //~/ns/apps/app.cc void Application::send(int nbytes) { agent_->sendmsg(nbytes); }

5 void Application::recv(int nbytes) 6 { 7 if (! enableRecv_) 8 return; 9 Tcl& tcl = Tcl::instance(); 10 tcl.evalf("%s recv %d", name_, nbytes); 11 } 12 void Application::resume() 13 { 14 if (! enableResume_) 15 return; 16 Tcl& tcl = Tcl::instance(); 17 tcl.evalf("%s resume", name_); 18 }

11.2 Details of Class Application

267

sender, this function is invoked when it sends out all the packets regardless of whether the transmitted packets have been acknowledged. Note that both functions recv(nbytes) and resume() will do nothing if enableRecv_ = 0 and enableResume_ = 0, respectively. Otherwise, Line 10 and 17 in Program 11.5 will invoke OTcl commands or instprocs recv{nbytes} (Line 10) and resume{} (Line 18) in the OTcl domain, respectively. By default, both enableRecv_ and enableResume_ are set to zero, and functions recv(nbytes) and resume() simply do nothing. A user may specify actions to be done upon invocation of functions recv(nbytes) and resume() by (i) Setting enableRecv_ and/or enableResume_ to one. (ii) Specifying the actions in (a) Functions recv(nbytes) and/or resume(), (b) Instprocs recv{nbytes} and/or resume{} in the OTcl domain, or (c) OTcl commands recv{nbytes} and/or resume{} in function command(). It is important to perform both the steps above. Failing to perform the second step will result in a run-time error, since commands or instprocs recv{nbytes} and resume{} are undeﬁned in class Application Exercise 11.1. Modify (1) C++ functions, (2) OTcl commands, and (3) OTcl instprocs. Force an application to print out a message when its functions recv(nbyte) and resume() are invoked. Show simulation results to verify the modiﬁcation.

11.2.3 Related Public Functions of Class Agent Class Application may invoke the following functions of class Agent through variable agent_: • send(nbytes): Send “nbytes” of application payload (i.e., user demand) to a receiving agent. If nbytes=-1, the user demand would be inﬁnite. • sendmsg(nbytes,flags): Similar to function send(nbytes), but also feed flags as an input variable. • close(): Ask an agent to close the connection (applicable only to TCP) • listen(): Ask an agent to listen to (i.e., wait for) a new connection (applicable only to Full TCP) • set_pkttype(pkttype): Set the variable type_ of the attach agent to be pkttype.

268

11 Application: User Demand Indicator

11.2.4 OTcl Commands of Class Application Deﬁned in function command, OTcl commands associated with class Application are as follows: • • • • start{}: Invoke function start() to start the application. stop{}: Invoke function stop() to stop the application. agent{}: Return the name of the attached agent. send{nbytes}: Send nbytes bytes of user payload to the attached agent by invoking function send(nbytes). • attach-agent{agent}: Create a two-way connection between itself and the input Agent object (agent). The details of the above OTcl command can be found in ﬁle ˜ns/apps/app.cc.

11.3 Traﬃc Generators
A traﬃc generator models user behavior which follows a predeﬁned schedule. In particular, it sends a demand to transmit one burst of user payload to an attached agent at a time speciﬁed in the schedule, regardless of the state of the agent. In NS2, there are four main traﬃc generators: • Constant Bit Rate (CBR): Send a ﬁxed size payload to the attached agent. By default, the interval between two payloads (i.e., the sending rate) is ﬁxed, but it can be optionally randomized. • Exponential On/Oﬀ: Send ﬁxed size payloads for every randomized interval to an attached agent during an ON period. Stop sending during an OFF period. ON and OFF periods are exponentially distributed, and are alternated when one period terminates. • Pareto On/Oﬀ: Similar to the Exponential On/Oﬀ traﬃc generator. However, the durations of ON and OFF periods follow a Pareto distribution. • Traﬃc Trace: Generate traﬃc according to a given trace ﬁle, which contains a series of inter-burst transmission intervals and payload burst sizes. 11.3.1 An Overview of Class TrafficGenerator NS2 implements traﬃc generators using class TrafficGenerator. Program 11.5 shows the declaration of the abstract class TrafficGenerator, where function next_interval(size) (Line 4) is pure virtual. Class Traffic Generator consists of the following variables: timer_ A TrafficTimer object, which determines when a new burst of payload is created. nextPkttime_ Simulation time that the next payload will be passed to the attached transport layer agent

11.3 Traﬃc Generators

269

Program 11.5 Declaration of class TrafficGenerator.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 //~/ns/tools/trafgen.h class TrafficGenerator : public Application { public: TrafficGenerator(); virtual double next_interval(int &) = 0; virtual void init() {} virtual double interval() { return 0; } virtual int on() { return 0; } virtual void timeout(); virtual void recv() {} virtual void resume() {} protected: virtual void start(); virtual void stop(); double nextPkttime_; int size_; int running_; TrafficTimer timer_; };

size_ running_

Application payload size true if the TrafficGenerator object is running

Class TrafficGenerator derives and overrides the following four key functions of class Application. It derives functions recv(nbytes) and resume() (i.e., share the implementation) from class Application, and overrides functions start(), and stop() of class Application. Functions start() and stop() inform the TrafficGenerator object to start and stop, respectively, generating user payload. In Program 11.6, function start() initializes the TrafficGenerator object by invoking function init()2 in Line 3, and sets running_ to 1 in Line 4. It computes and stores the time until the next payload burst is generated in variable nextPkttime_ in Line 5. Finally, it sets the timer_ to expire at nextPkttime_ seconds in future (Line 6). From Lines 8 to 13 in Program 11.6, function stop() does the opposite of function start(). It cancels the pending timer (if any) in Line 11, and sets running_ to 0 in Line 12. Class TrafficGenerator also deﬁnes the following three new functions: next_interval(size) Takes payload size “size” as an input argument, and returns the delay time after which a new payload is generated (Line 4). This function is pure virtual and must be implemented by the instantiable derived classes of class TrafficGenerator.
2

In Line 5 of Program 11.5, function init() simply does nothing.

270

11 Application: User Demand Indicator

Program 11.6 Functions TrafficGenerator.
1 2 3 4 5 6 7

start,

stop,

and

timeout

of

class

//~/ns/tools/trafgen.cc void TrafficGenerator::start() { init(); running_ = 1; nextPkttime_ = next_interval(size_); timer_.resched(nextPkttime_); }

8 void TrafficGenerator::stop() 9 { 10 if (running_) 11 timer_.cancel(); 12 running_ = 0; 13 } 14 void TrafficGenerator::timeout() 15 { 16 if (! running_) 17 return; 18 send(size_); 19 nextPkttime_ = next_interval(size_); 20 if (nextPkttime_ > 0) 21 timer_.resched(nextPkttime_); 22 else 23 running_ = 0; 24 }

init() Initializes the traﬃc generator. timeout() Sends a user payload to the attached application and restart timer_. This function is invoked when timer_ expires. The details of function timeout() are shown in Lines 14–24 of Program 11.6. Function timeout() does nothing if the TrafficGenerator object is not running (Lines 16–17). Otherwise, it will send “size_” bytes of user payload to the attached agent using function send(nbytes) (deﬁned in Program 11.4). Then, Line 19 will compute nextPkttime_. If nextPkttime_ > 0, Line 21 will inform timer_ to expire after a period of nextPkttime_. Otherwise, Line 23 will stop the TrafficGenerator by setting running_ to zero. 11.3.2 Main Mechanism of a Traﬃc Generator Figure 11.2 illustrates the main mechanism of a traﬃc generator, which relies heavily on the variable timer_ whose class is TrafficTimer derived from

11.3 Traﬃc Generators

271

Fig. 11.2. Main mechanism of a traﬃc generator.

class TimerHandler. As discussed in Section 12.1, class TimeHandler consists of three states: TIMER_IDLE, TIMER_PENDING, and TIMER_HANDLING. Each of these states corresponds to one of two TrafficGenerator states: Idle (i.e., running_=0) and Active (i.e., running_=1). While state TIMER_IDLE corresponds to the idle state of a TrafficGenerator object, the other two timer states are within the active state of a TrafficGenerator object. Starting in an idle state, a traﬃc generator moves to active state when function start() is invoked. Here the timer_ state is set to TIMER_PENDING. At the expiration, timer_ moves to state TIMER_HANDLING, and invokes function timeout() of class TrafficGenerator. After executing function timeout(), it reschedule itself, changes the state to TIMER_PENDING, reschedules itself, and repeats the above process. When timer_ state is TIMER_PENDING or TIMER_HANDLING, the traﬃc generator can be stopped by invoking function stop(). Program 11.7 shows the declaration of class TrafficTimer, which derives from class TimerHandler (see Section 12.1). Class TrafficTimer has a key variable tgen_, a pointer to a TrafficGenerator object (Line 6). At the expiration, NS2 invokes function expire(e) of timer_ (Lines 8–11), which in turn invokes function timeout() of the associated TrafficGenerator object (i.e., *tgen_). A two-way connection between TrafficGenerator and TrafficTimer objects is created as follows. Class TrafficGenerator declares timer_ as its pointer to a TrafficTimer object (Line 17 in Program 11.5). A TrafficGenerator object instantiates timer_ by feeding a pointer to itself (i.e., this)

272

11 Application: User Demand Indicator

Program 11.7 Declaration of class TrafficTimer, function expire of class TrafficTimer, and the constructor of class TrafficGenerator.
1 2 3 4 5 6 7 //~/ns/tools/trafgen.h class TrafficTimer : public TimerHandler { public: TrafficTimer(TrafficGenerator* tg) : tgen_(tg) {} protected: void expire(Event*); TrafficGenerator* tgen_; };

//~/ns/tools/trafgen.cc 8 void TrafficTimer::expire(Event *) 9 { 10 tgen_->timeout(); 11 } 12 TrafficGenerator::TrafficGenerator() : nextPkttime_(-1), running_(0), timer_(this) {}

as an input argument (Line 12 in Program 11.7). The construction of variable timer_ in turn assigns the input pointer (i.e., this) to its pointer to a TrafficGenerator object, tgen_ (Line 3 in Program 11.7), creating a connection back to the TrafficTimer object.

11.3.3 Built-in Traﬃc Generators in NS2 Constant Bit Rate (CBR) A CBR traﬃc generator creates a ﬁxed size payload burst for every ﬁxed interval. As shown in Program 11.8, NS2 implements CBR traﬃc generators by using a C++ class CBR_Traffic which is bound to an OTcl class Application/Traffic/CBR, whose key instvars with their default values are shown in Table 11.1. Note that, by default the inter-burst transmission interval, which is the interval between the beginning of two successive payload bursts, can be computed by dividing the payload burst size by the sending rate. By default, the inter-burst transmission interval is 210 × 8/488.000 ≈ 3.44 ms. The detailed mechanism for class CBR_Traffic will be discussed in Section 11.3.4.

11.3 Traﬃc Generators

273

Program 11.8 Class CBRTrafficClass which binds C++ class CBR Traffic and OTcl class Application/Traffic/CBR together.
//~/ns/tools/cbr_traffic.cc 1 static class CBRTrafficClass : public TclClass { 2 public: 3 CBRTrafficClass() : TclClass("Application/Traffic/CBR") {} 4 TclObject* create(int, const char*const*) { 5 return (new CBR_Traffic()); 6 } 7 } class_cbr_traffic;

Table 11.1. Instvars of a CBR traﬃc generator. Instvar Default value Description packetSize_ 210 Application payload size in bytes rate_ random_ maxpkts_ 488 × 103 0 (false) 167 Sending rate in bps If true, introduce a random time (either positive or negative) to the inter-burst transmission interval. Maximum number of application payload packet that CBR can send

Exponential On/Oﬀ An exponential on/oﬀ traﬃc generator acts as a CBR traﬃc generator during an ON interval and does not generate any payload during an OFF interval. ON and OFF periods are both exponentially distributed. As shown in Program 11.9, NS2 implements Exponential On/Oﬀ traﬃc generators by using a C++ class EXPOO_Traffic which is bound to an OTcl class Program 11.9 Class EXPTrafficClass which binds C++ class EXPOO Traffic and OTcl class Application/Traffic/Exponential together.
//~/ns/tools/expoo.cc 1 static class EXPTrafficClass : public TclClass { 2 public: 3 EXPTrafficClass() : TclClass("Application/ Traffic/Exponential") {} 4 TclObject* create(int, const char*const*) { 5 return (new EXPOO_Traffic()); 6 } 7 } class_expoo_traffic;

274

11 Application: User Demand Indicator

Application/Traffic/Exponential, whose key instvars with their default values are shown in Table 11.2.
Table 11.2. Instvars of an exponential on/oﬀ traﬃc generator. Instvar packetSize_ rate_ burst_time_ idle_time_ Default value 210 64 × 103 0.5 0.5 Description Application payload size in bytes Sending rate in bps during an ON period Average ON period in seconds Average OFF period in seconds

Pareto On/Oﬀ A Pareto On/Oﬀ traﬃc generator does the same as an Exponential On/Oﬀ generator but the ON and OFF periods conform to a Pareto distribution. As shown in Program 11.10, NS2 implements Pareto On/Oﬀ traﬃc generators by using a C++ class POO_Traffic which is bound to an OTcl class Application/Traffic/Pareto, whose key instvars with their default values are shown in Table 11.3. Program 11.10 Class POOTrafficClass which binds C++ class POO Traffic and OTcl class Application/Traffic/Pareto together.
//~/ns/tools/pareto.cc 1 static class POOTrafficClass : public TclClass { 2 public: 3 POOTrafficClass() : TclClass("Application/Traffic/Pareto") {} 4 TclObject* create(int, const char*const*) { 5 return (new POO_Traffic()); 6 } 7 } class_poo_traffic;

Table 11.3. Instvars of a Pareto/Oﬀ traﬃc generator. Instvar packetSize_ rate_ burst_time_ idle_time_ shape_ Default value 210 64 × 103 0.5 0.5 1.5 Description Application payload in bytes Sending rate in bps during an ON period Average ON period in seconds Average OFF period in seconds A “Shape” parameter of a Pareto distribution

11.3 Traﬃc Generators

275

Traﬃc Trace A traﬃc trace generates payload bursts according to a given trace ﬁle. As shown in Program 11.11, NS2 implements traﬃc trace by using the C++ class TrafficTrace which is bound to an OTcl class Application/Traffic/Trace. Unlike other traﬃc generators described before, we need to specify a traﬃc trace ﬁle in the OTcl domain using command attach-tracefile of class Application/Traffic/Trace (see Example 11.2). Program 11.11 Class TrafficTraceClass which binds C++ class TrafficTrace and OTcl class Application/Traffic/Trace together.
//~/ns/trace/traffictrace.cc 1 static class TrafficTraceClass : public TclClass { 2 public: 3 TrafficTraceClass() : TclClass("Application/Traffic/Trace") {} 4 TclObject* create(int, const char*const*) { 5 return(new TrafficTrace()); 6 } 7 } class_traffictrace;

Example 11.2. A CBR traﬃc generator in Example 9.1 can be replaced with a traﬃc trace traﬃc generator by substituting Lines 10–12 in Program 9.2 with the following lines: set tfile [new Tracefile] $tfile filename example-trace set tt [new Applicaiton/Traffic/Trace] $tt attach-tracefile $tfile $tt attach-agent $udp A traﬃc trace ﬁle is a pure binary ﬁle. A codeword in the binary ﬁle consists of two 32-bits ﬁelds. The ﬁrst ﬁeld indicates inter-burst transmission interval in microseconds, while the second is the payload size in bytes (see ﬁle ˜ns/tcl/ex/example-trace as an example traﬃc trace ﬁle). 11.3.4 Class CBR Traffic: An Example Traﬃc Generator This section presents a C++ implementation of class CBR_Traffic whose declaration is shown in Program 11.12). Class CBR_Traffic derives from class TrafficGenerator, and has the following main variables: rate_ CBR sending rate in bps interval_ Packet inter-arrival time in seconds

276

11 Application: User Demand Indicator

random_ If true, the inter-arrival time will be random seqno_ CBR sequence number maxpkts_ Upper bound on the sequence number Based on the main mechanism discussed in Section 11.3.2, NS2 activates a traﬃc generator by invoking function start(). When activated, a trafﬁc generator invokes its function timeout(), which generates an application payload, periodically. An interval between two consecutive timeout() invocations is determined by the function next_interval(size). The timeout() invocations occur repeatedly until the traﬃc generator is deactivated (by an invocation of function close()). As shown in Program 11.12, function start() invokes function init() (Line 17) to initialize the traﬃc generator, sets running_ to 1 (Line 18), and Program 11.12 Functions start and init of class CBR Traffic.
//~/ns/tools/cbr_traffic.cc 1 class CBR_Traffic : public TrafficGenerator { 2 public: 3 CBR_Traffic(); 4 virtual double next_interval(int&); 5 inline double interval() { return (interval_); } 6 protected: 7 virtual void start(); 8 void init(); 9 double rate_; 10 double interval_; 11 double random_; 12 int seqno_; 13 int maxpkts_; 14 }; 15 void CBR_Traffic::start() 16 { 17 init(); 18 running_ = 1; 19 timeout(); 20 } 21 void CBR_Traffic::init() 22 { 23 interval_ = (double)(size_ get_pkttype() != PT_TCP && agent_->get_pkttype() != PT_TFRC) 26 agent_->set_pkttype(PT_CBR); 27 }

11.4 Simulated Applications

277

invokes function timeout() (Line 19). The details of function init() are shown in Lines 21–28 of Program 11.12. Line 23 computes the inter-burst transmission interval as transmission rate (rate_) divided by payload burst size “size_handle (...). Since variable handle_ of the dispatched event_ points to the TimerHandler object (see Fig. 12.3), NS2 invokes function handle(e) associated with the TimerHandler object at the ﬁring time. Function handle(e) of class TimerHandler in turn invokes function expire(e) (Line 6 of Program 12.4) which takes expiration actions speciﬁed by the derived classes of class TimerHandler. Function _cancel() does the opposite of what function _sched(delay) does. It removes the timer expiration event from the simulation timeline. From Line 19 in Program 12.3, it invokes function cancel(&event_) of class Scheduler to remove the event “event” from the simulation timeline. Expiration Actions At the ﬁring time, the Scheduler dispatches a timer expiration event by invoking function handle(e) of the associated timer (see also Fig. 12.3). The details of function handle(e) are shown in Program 12.4. Line 3 ﬁrst checks whether the current status_ is TIMER_PENDING. If so, Line 5 will change the variable status_ to TIMER_HANDLING, and Line 6 will invoke function expire(e) to take expiration actions. After returning from the function expire(e), variable status_ is set by default to TIME_IDLE (Line 8). However, if status_ has already changed (e.g., when the timer is rescheduled;

12.1 Timers

289

status_=TIMER_HANDLING in Line 7), function handle(e) will not change variable status_. Program 12.4 Function handle of class TimerHandler.
//~/ns/common/timer-handler.cc 1 void TimerHandler::handle(Event *e) 2 { 3 if (status_ != TIMER_PENDING) 4 abort(); 5 status_ = TIMER_HANDLING; 6 expire(e); 7 if (status_ == TIMER_HANDLING) 8 status_ = TIMER_IDLE; 9 }

In Line 10 of Program 12.3, function expire(e) is pure virtual. Therefore, derived instantiable classes of class TimerHandler are responsible for providing expiration actions by overriding this function. For example, class MyTimer below derives from class TimerHandler, and overrides function expire(e): void MyTimer::expire(Event *e) { printf("MyTimer has just expired!!\n"); } which prints the statement “MyTimer has just expired!!” on the screen upon timer expiration. Interface Functions to Start, Restart, and Cancel a Timer

Program 12.5 Function sched of class TimerHandler.
1 2 3 4 5 6 7 8 9 //~/ns/common/timer-handler.cc void TimerHandler::sched(double delay) { if (status_ != TIMER_IDLE) { fprintf(stderr,"Couldn’t schedule timer"); abort(); } _sched(delay); status_ = TIMER_PENDING; }

290

12 Related Helper Classes

The details of function sched(delay) of class TimerHandler is shown in Program 12.5. Function sched(delay) takes one input argument delay, and sets the timer to expire at delay seconds in the future by feeding delay into function _sched(delay) (Line 7). Note that function sched(delay) must be invoked when the status_ of the timer is TIMER_IDLE. Otherwise, Lines 4-5 will show an error message and exit the program. Program 12.6 shows the details of functions resched(delay) and cancel() of class TimerHandler. Function resched(delay) is very similar to function sched(delay). In fact, when invoked with status_ = TIMER_PENDING, it does the same as function sched(delay) does (i.e., starts the timer). However, when status_=TIMER_PENDING (Line 3)–meaning event_ was placed on the simulation timeline prior to the invocation–function resched(delay) removes the timer expiration event from the simulation time line, by invoking function _cancel(), and (re)starts the timer (Lines 4 and 5, respectively). Program 12.6 Functions resched and cancel of class TimerHandler.
1 2 3 4 5 6 7 //~/ns/common/timer-handler.cc void TimerHandler::resched(double delay) { if (status_ == TIMER_PENDING) _cancel(); _sched(delay); status_ = TIMER_PENDING; }

8 void TimerHandler::cancel() 9 { 10 if (status_ != TIMER_PENDING) { 11 ... 12 abort(); 13 } 14 _cancel(); 15 status_ = TIMER_IDLE; 16 }

Lines 8–16 of Program 12.6 show the details of function cancel() of class TimerHandler. Function cancel() invokes function _cancel() in Line 14 to remove the pending timer expiration event from the simulation timeline. Function cancel() must not be invoked, when event_ is not on the simulation timeline (i.e., status_ is either TIMER_IDLE or TIMER_HANDLING). Otherwise, NS2 will show an error message on the screen and exit the program (Lines 11-12).

12.1 Timers

291

Cross Referencing a Timer with Another Object In most cases, the usefulness of a timer stands out when it is cross-referenced with another object. In this case, the object employs a timer as a waiting tool, which starts, restarts, and cancels the waiting process as necessary. The timer, on the other hand, informs the object of timer expiration, upon which the object may take expiration actions. A typical cross-reference between a timer and an object can be created as follows: (i) Declare a timer as a variable of an object class. (ii) Declare a pointer to the object as a member of the timer class. (iii) Deﬁne a non-default constructor for the timer class. Store the input argument of the constructor in its member pointer variable (which points to the associated object). (iv) Instantiate a timer object from within the constructor of the associated object. Use the non-default constructor of the timer class deﬁned above. Feed the pointer this (i.e., the pointer to the object) as an input argument to the constructor of the timer. We now conclude this section with a simple timer example. Example 12.1. Consider a process of counting the number of customers who enter a store during a day. Let class Store represent a convenience store (i.e., an object class), and let class StoreHour represent the number of opening hours of a day (i.e., a timer class). The opening hours is speciﬁed when the store is opened. The objective here is to count the number of visiting customers during a day, and print out the result when the store is closed. Classes Store and StoreHour From Program 12.7, class Store also has 3 variables. First, hours_ (Line 17) contains opening hours of the store and is set to zero at the construction. Secondly, count_ (Line 18) records the number of customers who have entered the store so far and is set to zero at the construction. Finally, variable timer_ is a StoreHour object. Function close() (Lines 12–13) of class Store is invoked when the store is being closed. It prints out the opening hours and number of visiting customers for today on the screen. Declared in Line 1–8, class StoreHour has only one variable store_ (Line 7) which is a pointer to a Store object. Cross-Referencing Store and StoreHour Objects The process of cross-referencing a Store object and a StoreHour object is shown in Fig. 12.4. The constructor of class Store constructs its variable timer_ with the pointer this to the Store object (see Line 11). The constructor of class StoreHour stores the input pointer in variable store_. Since

292

12 Related Helper Classes

Program 12.7 Declaration of classes Store and StoreHour.
1 2 3 4 5 6 7 8 //store.h class Store; class StoreHour : public TimerHandler { public: StoreHour(Store *s) { store_ = s; }; virtual void expire( Event *e ); protected: Store *store_; };

9 class Store : public TclObject { 10 public: 11 Store() : timer_(this) { hours_ = -1; count_ = 0; }; 12 void close(){ 13 printf("The number of customers during %2.2f hours is %d\n", hours_,count_); 14 }; 15 int command(int argc, const char*const* argv); 16 protected: 17 double hours_; 18 int count_; 19 StoreHour timer_; 20 }

the input argument is the pointer to the Store object, the constructor of the StoreHour object essentially sets the variable store_ to point back to the Store object. Due to the cross-referencing, the compiler needs to recognize one of these two classes when declaring another. If Line 1 was removed, the compiler would not recognize class Store when compiling Line 7, and would show a compilation error message on the screen. After compiling Line 2, the compiler recognizes class StoreHour and can compile Line 19 without error.

Fig. 12.4. A diagram which represents the process of cross-referencing a Store object and a StoreHour object.

12.1 Timers

293

Program 12.8 Function expire of class StoreHour as well as OTcl Commands open and new-customer of class Store.
1 2 3 //store.cc void StoreHour::expire(Event*) { store_->close(); };

4 int Store::command(int argc, const char*const* argv) 5 { 6 if (argc == 3) { 7 if (strcmp(argv[1], "open") == 0) { 8 hours_ = atoi(argv[2]); 9 count_ = 0; 10 timer_.sched(hours_); 11 return (TCL_OK); 12 } 13 } else if (argc == 2) { 14 if (strcmp(argv[1], "new-customer") == 0) { 15 count_++; 16 return (TCL_OK); 17 } 18 } 19 return TclObject::command(argc,argv); 20 }

It is also important to note that when compiling Lines 2–8, the compiler recognizes only Store class name. Any attempt to invoke functions (e.g., close()) of class Store will result in a compilation error. This is the reason why we need to separate C++ codes into header and C++ ﬁles. Again, since a header ﬁle is included at the top of a C++ ﬁle, the compiler ﬁrst goes through the header ﬁle and recognizes all the variables and functions speciﬁed in the header ﬁle. With this knowledge, the compiler can compile the C++ ﬁle without error. Deﬁning Expiration Actions Derived from class TimerHandler, class StoreHour overrides function expire( e) as shown in Lines 1–3 of Program 12.8. At the expiration, the timer (i.e., StoreHour object) simply invokes function close() of the associated Store object. Creating OTcl Interface We bind the C++ class Store to an OTcl class with the same name using a mapping class StoreClass shown in Program 12.9. Lines 4–20 in Program 12.8 also show OTcl interface commands open{hours} and new-customer{}. With

294

12 Related Helper Classes

opening hours hours as an input argument, the OTcl command open{hours} (Lines 8–11) is invoked when the store is opened. Line 8 stores the opening hours in variable hours_, Line 9 resets the number of visiting customers to zero, and Line 10 tells timer_ to expire at “hours_” hours in future. The OTcl command new-customer{} is invoked as a customer enters the store. In Line 15, this command simply increases count_ by one. Again, at the timer expiration, the timer invokes function close() through the pointer store_ and prints out the opening hours (i.e., hours_) as well as the number of visiting customers (i.e., count_) for today (see function expire(e) in Line 2 of Program 12.8). Program 12.9 A mapping class StoreClass which binds C++ and OTcl classes Store.
//store.cc 1 static class StoreClass : public TclClass { 2 public: 3 StoreClass() : TclClass("Store") {} 4 TclObject* create(int, const char*const*) { 5 return (new Store); 6 } 7 } class_store;

Testing the Codes After deﬁning ﬁles store.cc and store.h, we include store.o to the Make File and run make at NS2 root directory to include classes Store and StoreHour into NS2 (see Section 2.7). For, deﬁne a test Tcl simulation script in a ﬁle store.tcl, 1 2 3 4 5 6 7 8 9 //store.tcl set ns [new Simulator] set my_store [new Store] $my_store open 10.0 $ns at 1 "$my_store new-customer" $ns at 5 "$my_store new-customer" $ns at 6 "$my_store new-customer" $ns at 8 "$my_store new-customer" $ns at 11 "$my_store new-customer" $ns run

We, run the script store.tcl, and obtain the following results: >>ns store.tcl The number of customers during 10.0 hours is 4

12.1 Timers

295

From the above script, when Line 2 creates a Store object, NS2 automatically creates a shadow C++ Store Object. Line 3 invokes command open with input argument 10.0, essentially opening the store for 10.0 hours. From Program 12.8, an OTcl command open{10.0} and tells the associated timer to expire at 10.0 hours in future, and clears the variable count_. Lines 4–8 invoke command new-customer{} at 1st, 5th, 6th, 8th, and 11th hours. Each of these lines increases the number of visiting customers (i.e., count_) by one. By the end of 11th hour in future, variable count_ should be 5. However, the program shows that the number of visiting customers is 4. This is because the timer expires and invokes function close() at the 10th hour. 12.1.4 Guidelines for Implementing Timers in NS2 We now summarize the process of deﬁning a new timer. Suppose that we would like to deﬁne a new timer class StoreHour. Suppose further that a Store object is responsible for starting, restarting, and canceling the StoreHour object, and for taking expiration actions. Then, the implementation of the above timer classes proceeds as follows: From class StoreHour • Step 1: Design class structure: – Derive class StoreHour from class TimerHandler. – Declare a pointer (e.g., store_) to class Store. The public function of class Store is accessible through the above pointer (e.g., store_) to class Store. • Step 2: Bind the reference to class Store in the constructor. • Step 3: Deﬁne expiration actions in function expire(e). From class Store • Step 1: Design class structure: – Derive class Store from class TclObject only if an interface to OTcl is necessary. – Declare a StoreHour variable (e.g., timer_) as a member variable. • Step 2: Instantiate the above StoreHour variable (e.g., timer_) with the pointer “this”. At runtime, we only need to instantiate a Store object. The internal mechanism of class Store will automatically create and conﬁgure a StoreHour object. Also, we do not need any global (or OTcl) reference to the StoreHour object, since the timer is usually manipulated by class Store.

296

12 Related Helper Classes

12.2 Implementation of Random Numbers in NS2
This section demonstrates implementation of random number generators in NS2. In principle, NS2 employs so-called Random Number Generator (RNG) to generate random numbers. An RNG sequentially picks numbers from a stream of psudo-random numbers. A set of generated random numbers is characterized by the point where the RNG starts picking the numbers–called “seed”. By default, NS2 sets the seed to 1. Therefore, the results obtained from every run are essentially the same. Random numbers can also be transformed to conform to a given distribution. Such the transformation is carried out through instprocs in the OTcl domain, and through classes derived from class RandomVariable in the C++ domain. We will discuss the details of RNGs and the seeding mechanism in Sections 12.2.1 and 12.2.2, respectively. Section 12.2.3 shows the implementation of RNGs in NS2. Section 12.2.4 discusses diﬀerent simulation scenarios, where RNGs are set diﬀerently. Section 12.2.5 explains the implementation of a C++ class RandomVariable which transforms random numbers according to a given distribution. Finally, Section 12.2.6 gives a guideline to deﬁne a new RNG and a new random variable in NS2. 12.2.1 Random Number Generation NS2 generates random numbers by sequentially picking numbers from a stream of pseudo-random number (as discussed in Section 1.3.2). It uses the combined multiple recursive generator (MRG32k3a) proposed by L’Ecuyer [22] as a pseudo-random number generator. Generally speaking, an MRG32k3a generator contains streams of pseudo-random numbers from which the numbers picked sequentially seem to be random. In Fig. 12.5, an MRG32k3a generator provides 1.8 × 1019 independent streams, each of which consists of 2.3 × 1015 substreams. Each substream contains 7.6 × 1022 random numbers (i.e., the period of each substream is 7.6 × 1022 ). In summary, an MRG32k3a generator can create 3.1 × 1057 numbers which appear to be random. 12.2.2 Seeding a Random Number Generator As mentioned in Section 1.3.2, “seed” is one of the main ingredients of Random Number Generator (RNG). Loosely speaking, a seed speciﬁes the location on a stream of pseudo-random numbers, where an RNG starts picking random numbers sequentially. When seeded diﬀerently, two RNGs start picking pseudo-random numbers from diﬀerent locations, and therefore generate two distinct sets of random numbers. On the other hand, if seeded with the same number, two RNGs will start picking random numbers from the same location, and therefore, generate the same set of random numbers. By default, NS2 always uses only one OTcl variable defaultRNG as a default RNG, and always seeds defaultRNG with 1. Therefore, the simulation

12.2 Implementation of Random Numbers in NS2

297

Fig. 12.5. Streams and substreams of an MRG32k3a generator.

results for every run are essentially the same. To collect independent simulation results, we must seed diﬀerent runs diﬀerently. Example 12.2. In the following, we run NS2 for three times to show NS2 seeding mechanism. 1 2 3 4 5 6 7 8 >>ns >>$defaultRNG seed 1 >>$defaultRNG next-random 729236 >>$defaultRNG next-random 1193744747 >>exit

### RESTART NS2 ### 9 >> ns 10 >>$defaultRNG seed 11 1 12 >>$defaultRNG next-random 13 729236 14 >>$defaultRNG next-random 15 1193744747 16 >>exit

298

12 Related Helper Classes

### RESTART NS2 ### 17 >>ns 18 >>$defaultRNG seed 101 19 >>$defaultRNG next-random 20 72520690 21 >>$defaultRNG next-random 22 308637100 23 >>exit In the ﬁrst run (Lines 1–8), variable defaultRNG (i.e., the default RNG) is used to generate two random numbers. In Line 2, instproc seed returns the current seed which is set (by default) to 1. Lines 4 and 6 use instproc next-random{} to generate two random numbers, 729236 and 1193744747, respectively. Finally, Line 8 exits the NS2 environment. Lines 9–16 repeat the process in Lines 1–8. In Lines 10–11, we can observe that the seed is still 1. As expected, the ﬁrst and the second random numbers generated are 729236 and 1193744747, respectively. These two numbers are the same as those in the ﬁrst run. Essentially, the ﬁrst run and the second run generate the same results. To generate diﬀerent results, we need to seed the simulation diﬀerently. Lines 17–22 show the last run, where the seed is set diﬀerently (to 101). The ﬁrst and the second random number generated in this case are 72520690 and 308637100, respectively. These two numbers are diﬀerent from those in the ﬁrst two runs, since Line 15 sets the seed of defaultRNG to 101. The key points about seeding the mechanism in NS2 are as follows: • A seed speciﬁes the starting location on a stream of psudo-random numbers, and hence characterizes an RNG. • To generate two independent simulation results, each simulation must be seeded diﬀerently. • At initialization, NS2 creates a variable defaultRNG as the default RNG, and seeds defaultRNG with 1. By default, NS2 generates the same simulation result for every run. • When seeded with zero, an RNG replaces the seed with current time of the day and counter. Despite their tendency to be independent, two runs may pick the same seed and generate the same result. To ensure independent runs, we must seed the RNG manually. • NS2 seeds a new RNG object to the beginning of the next random stream. Therefore, every RNG object is independent of each other.

12.2.3 OTcl and C++ Implementation NS2 employs a C++ class RNG (which is bound to an OTcl class with the same name) to generate random numbers (see Program 12.10). In most cases, it is

12.2 Implementation of Random Numbers in NS2

299

not necessary to understand the details of the MRG32k3a generator. This section shows only the key conﬁguration and implementation in the OTcl and C++ domains. The readers may ﬁnd the detailed implementation of an MRG32k3a generator in ﬁles ˜ns/tools/rng.cc,h. Program 12.10 A mapping class RNGClass which binds OTcl and C++ classes RNG.
//~/ns/tools/rng.cc 1 static class RNGClass : public TclClass { 2 public: 3 RNGClass() : TclClass("RNG") {} 4 TclObject* create(int, const char*const*) { 5 return(new RNG()); 6 } 7 } class_rng;

OTcl Commands and Instprocs In the OTcl domain, class RNG deﬁnes the following OTcl commands: Return the seed of RNG. Set the the seed of RNG to be n. Return a random number. Advance to the beginning of the next substream. reset-start-substream{} Return to the beginning of the current substream. normal{avg std} Return a random number normally distributed with average avg and standard deviation std. lognormal{avg std} Return a random number log-normally distributed with average avg and standard deviation std. seed{} seed{n} next-random{} next-substream{} Deﬁned in ﬁle ˜ns/tcl/lib/ns-random.tcl, the following instprocs generate random numbers with exponential distribution and uniform distribution:

300

12 Related Helper Classes

exponential{mu} Return a random number exponentially distributed with mean mu. uniform{min max} Return a random number uniformly distributed in [min,max]. integer{k} Return a random integer uniformly distributed in {0,1,...,k-1}.

C++ Functions In the C++ domain, the key functions of class RNG include (see the details in ﬁles ˜ns/tools/rng.cc,h): set_seed(n) If n = 0, set the the seed of the RNG to be current time and counter. Otherwise, set the seed to be n. seed() Return the seed of the RNG. next() Return a random int in {0,1,..., MAX_INT}. next_double() Return a random double in [0,1]. reset_start_substream() Move to the beginning of the current substream. reset_next_substream() Move to the beginning of the next substream. uniform(k) Return a random int number uniformly distributed in {0,1,...,k-1}. uniform(r) Return a random double number uniformly distributed in [0,r]. uniform(a,b) Return a random double number uniformly distributed in [a,b]. exponential(k) Return a random number exponentially distributed with mean k. normal(avg,std) Return a random number normally distributed with average avg and standard deviation std. lognormal(avg,std) Return a random number log-normally distributed with average avg and standard deviation std. 12.2.4 Randomness in Simulation Scenarios In most cases, a simulation falls into one of the following three scenarios. Deterministic Setting This type of simulation is usually used for debugging. Its purpose is to locate programming errors in the simulation codes or to understand complex behavior of a certain network. In both cases, it is convenient to run the program

12.2 Implementation of Random Numbers in NS2

301

under a deterministic setting and generate the same result repeatedly. By default, NS2 seeds the simulation with 1. The deterministic setting is therefore the default setting for NS2 simulation. Single-Stream Random Setting The simplest form of statistical analysis is to run a simulation for several times and compute statistical measures such as average and/or standard deviation. By default, NS2 always uses defaultRNG with seed “1” to generate random numbers. To statistically analyze a system, we need to generate several distinct sets of results. Therefore, we need to seed diﬀerent runs diﬀerently. In a single-stream random setting, we need only one RNG. Hence, we may simply introduce the diversity to each run by seeding diﬀerent runs with diﬀerent values (e.g., in Example 12.2, Line 18 seeds the default RNG with 101). $defaultRNG which seeds the default RNG with a number . Multiple-Stream Random Setting In some cases, we may need more than one independent random variable for a simulation. For example, we may need to generate random values of packet inter-arrival time as well as packet size. These two variables should be independent and should not share the same random stream. We can create two independent RNG using “new RNG”. Since NS2 seeds each RNG with diﬀerent random stream (see Section 12.2.2), the random processes with diﬀerent RNGs are independent of each other. Example 12.3. Suppose that the inter-arrival time and packet size are exponentially distributed with mean 5 and uniformly distributed within [100, 5000], respectively. Print out the ﬁrst 5 random values of inter-arrival time and packet size. Tcl simulation script: 1 $defaultRNG seed 101 2 set arrivalRNG [new RNG] 3 set sizeRNG [new RNG] 4 set arrival_ [new RandomVariable/Exponential] 5 $arrival_ set avg_ 5 6 $arrival_ use-rng $arrivalRNG 7 set size_ [new RandomVariable/Uniform] 8 $size_ set min_ 100

302

12 Related Helper Classes

9 $size_ set max_ 5000 10 $size_ use-rng $sizeRNG 11 puts "Inter-arrival time Packet size" 12 for {set j 0} {$j < 5} {incr j} { 13 puts [format "%-8.3f %-4d" [$arrival_ value] \ [expr round([$size_ value])]] 14 } Results on the Screen: Inter-arrival time 1.048 7.919 8.061 4.675 7.201 Packet size 1880 116 3635 2110 1590

The details of the above Tcl simulation script are as follows. Lines 4 and 7 create an exponentially random variable2 arrival_ and a uniformly random variable size_ whose parameters are deﬁned in Lines 5–6 and Lines 8–10, respectively. Lines 11–14 print out ﬁve random numbers generated by arrival_ and size_. In Section 12.2.5, we will see that the OTcl command “value” of class RandomVariable returns a random number and the OTcl command “use-rng” is used to specify an RNG for a random variable. By default, defaultRNG is used to generate random numbers for both arrival_ and size_. In this case, Lines 2 and 3 create two independent RNGs: arrivalRNG and sizeRNG. NS2 speciﬁes these two variables as RNGs for arrival_ and size_ by using an OTcl command use-rng in Lines 6 and 10, respectively. Since the created RNG objects are independent, random variable arrival_ and size_ are independent of each other. Exercise 12.4. From Example 12.3, (i) Change the seed to “999”. Re-run the script for a couple of times. Observe and explain the output. (ii) Change the seed to “0”. Re-run the script for a couple of times. Observe and explain the output. (iii) Print out values of arrival_ and size_ for (i) and (ii), and show that they are exponentially and uniformly distributed (Hint: Set the seed properly). (iv) Change the mean of arrival_ to 10 and the interval of size_ to [400, 2000], and repeat (iii). (v) Remove Line 6 and repeat (iii). Observe and explain the output. (vi) Remove Lines 6 an 10 and repeat (iii). Observe and explain the output.
2

We will discuss the details of random variables in the next section.

12.2 Implementation of Random Numbers in NS2

303

12.2.5 Random Variables In NS2, a random variable is a module which generates random values whose statistics follow a certain distribution. It employs an RNG to generate random numbers and transforms the generated numbers to values which conform to a given distribution. This implementation is carried out in C++ abstract class RandomVariable whose diagram and declaration are shown in Fig. 12.6 and Program 12.11, respectively.

Fig. 12.6. A schematic diagram of class RandomVariable.

Consider the declaration of class RandomVariable in Program 12.11. Class RandomVariable contains a pointer rng_ (Line 9) to an RNG object (used to generate random numbers), and two pure virtual interface functions: value() in Line 3 and avg() in Line 4. Function value() generates random numbers, transforms the generated numbers to values conforming to a given distribution, and returns the transformed values to the caller. Function avg() returns the average value of the underlying distribution. Since these two functions are pure virtual, they must be overridden by all derived instantiable classes of class RandomVariable. The list of key built-in instantiable C++ classes as well as their bound OTcl classes is given in Table 12.2. Program 12.11 Declaration of class RandomVariable.
//~/ns/tools/ranvar.h 1 class RandomVariable : public TclObject { 2 public: 3 virtual double value() = 0; 4 virtual double avg() = 0; 5 int command(int argc, const char*const* argv); 6 RandomVariable(); 7 int seed(char *); 8 protected: 9 RNG* rng_; 10 };

304

12 Related Helper Classes Table 12.2. Built-in C++ and OTcl random variable classes.

C++ class UniformRandomVariable ExponentialRandomVariable ParetoRandomVariable ParetoIIRandomVariable NormalRandomVariable LogNormalRandomVariable ConstantRandomVariable HyperExponentialRandomVariable WeibullRandomVariable EmpiricalRandomVariable

OTcl class RandomVariable/Uniform RandomVariable/Exponential RandomVariable/Pareto RandomVariable/ParetoII RandomVariable/Normal RandomVariable/LogNormal RandomVariable/Constant RandomVariable/HyperExponential RandomVariable/Weibull RandomVariable/Empirical

Random Number Generator A RandomVariable object utilizes variable rng_ to generate random numbers. By default, every random variable uses the defaultRNG as its RNG. As shown in Program 12.12, the constructor (Lines 1–4) of class RandomVariable stores the default RNG returned from function RNG::defaultrng() in variable rng_. To create multiple independent random variables, variable rng_ of each random variable must be independent of each other. From Example 12.3, this can be achieved by creating and binding a dedicated RNG to each random variable. As will be discussed in the next section, the process of binding an RNG to a random variable is carried out by using the OTcl command use-rng associated with a RandomVariable object. OTcl Commands Shown in Program 12.12, class RandomVariable deﬁnes the following two commands, which can be invoked from the OTcl domain: • value{}: Returns a random number by invoking function value() (Lines 9–12). • use-rng{rng}: Casts the input argument rng to type RNG*, and stores the cast object in variable rng_ (Lines 15–19). Note that an example use of the OTcl command use-rng{rng} is shown in Lines 6 and 10 in Example 12.3. Since class RandomVariable is abstract, it does not provide a shadow class in the OTcl domain. However, all its derived classes do have shadow classes in the OTcl domain. Table 12.2 lists 10 built-in C++ and OTcl random variable classes.

12.2 Implementation of Random Numbers in NS2

305

Program 12.12 The constructor, OTcl command value, and OTcl command use-rng of class RandomVariable.
1 2 3 4 //~/ns/tools/ranvar.cc RandomVariable::RandomVariable() { rng_ = RNG::defaultrng(); }

//~/ns/tools/ranvar.cc 5 int RandomVariable::command(int argc, const char*const* argv) 6 { 7 ... 8 if (argc == 2) { 9 if (strcmp(argv[1], "value") == 0) { 10 tcl.resultf("%6e", value()); 11 return(TCL_OK); 12 } 13 } 14 if (argc == 3) { 15 if (strcmp(argv[1], "use-rng") == 0) { 16 rng_ = (RNG*)TclObject::lookup(argv[2]); 17 ... 18 return(TCL_OK); 19 } 20 } 21 ... 22 }

Exponential Random Variable As an example, consider implementation of an exponentially random variable in Program 12.13. From Table 12.2, NS2 implements an exponentially random variable using the C++ class ExponentialRandomVariable and the OTcl class RandomVariable/Exponential. Since an exponential random variable is completely characterized by an average value, class ExponentialRandomVariable has only one member variable avg_ (Line 9), which stores the average value. At the construction (see Lines 18–20), class ExponentialRandomVariable binds variable avg_ to instvar avg_ in the OTcl domain. Functions avg() in Line 6 and avgp() in Line 5 return the value stored in avg_ and the address of avg_, respectively. Function setavg(d) in Line 7 stores the value in “d” into variable “avg_”. Function value() in Lines 21–23 returns a random number exponentially distributed with mean avg_. It invokes function exponential(avg_) of variable rng_, feeding variable avg_ as an input argument to obtain an exponentially distributed random number.

306

12 Related Helper Classes

Program 12.13 An implementation of class ExponentialRandomVariable.
//~/ns/tools/ranvar.h 1 class ExponentialRandomVariable : public RandomVariable { 2 public: 3 virtual double value(); 4 ExponentialRandomVariable(); 5 double* avgp() { return &avg_; }; 6 virtual inline double avg() { return avg_; }; 7 void setavg(double d) { avg_ = d; }; 8 private: 9 double avg_; 10 }; //~/ns/tools/ranvar.cc 11 static class ExponentialRandomVariableClass : public TclClass { 12 public: 13 ExponentialRandomVariableClass() : TclClass( "RandomVariable/Exponential") {} 14 TclObject* create(int, const char*const*) { 15 return(new ExponentialRandomVariable()); 16 } 17 } class_exponentialranvar; 18 ExponentialRandomVariable::ExponentialRandomVariable(){ 19 bind("avg_", &avg_); 20 } 21 double ExponentialRandomVariable::value(){ 22 return(rng_->exponential(avg_)); 23 }

Exercise 12.5. Write a simulation script which generates random numbers exponentially distributed with mean 1.0. To verify the script, plot the probability density function. Exercise 12.6. Write a simulation script which generates a random number normally distributed with mean 1.0 and standard deviation 0.05. To verify the script, plot the probability density function. Exercise 12.7. Develop a new class for a discrete random variable whose probability mass function is (0.1, 0.3, 0.3, 0.2, 0.1). Test the code by generating random numbers and verify the probability mass function. 12.2.6 Guidelines for Random Number Generation in NS2 We conclude this section, by providing the following guidelines for implementing randomness numbers in NS2:

12.3 Built-in Error Models

307

• Determine the type of simulation: deterministic setting, single-stream random setting, or multi-stream random setting. • Create RNG(s) according to the simulation type. • If needed, create a random variable – Deﬁne the inheritance structure: C++, OTcl, and mapping classes. – Deﬁne function avg() which returns the average value of the distribution to the caller. – Deﬁne function value() which returns a random number conforming to the speciﬁed distribution. • Specify an RNG for each random variable by using an OTcl command use-rng of class RandomVariable.

12.3 Built-in Error Models
An error model is an NS2 module which imposes error on packet transmission. Derived from class Connector, it can be inserted between two NsObjects. An error model simulates packet error upon receiving a packet. If the packet is simulated to be in error, the error model will either drop the packet or mark the packet with an error ﬂag. If the packet is simulated not to be in error, on the other hand, the error model will forward the packet to its downstream object. An error model can be used for both wired and wireless networks. However, this section discusses the details of an error model through a wired class SimpleLink only. Program 12.14 Class ErrorModelClass which binds C++ and OTcl classes ErrorModel.
1 2 3 4 5 6 7 //~/ns/queue/errmodel.cc static class ErrorModelClass : public TclClass { public: ErrorModelClass() : TclClass("ErrorModel") {} TclObject* create(int, const char*const*) { return (new ErrorModel); } } class_errormodel;

NS2 implements error models using a C++ class ErrorModel which is bound to an OTcl class with the same name (see Program 12.14). Class ErrorModel simulates Bernoulli error, where transmission is simulated to be either in error or not in error. NS2 also provides ErrorModel classes with more functionalities such as two-state error model. Tables 12.3 and 12.4 show NS2 built-in error models whose implementation is in the C++ and OTcl domain, respectively.

308

12 Related Helper Classes

Table 12.3. Built-in error models which contain C++ and OTcl implementation. C++ class TwoStateErrorModel ComplexTwoState MarkovModel MultiStateErrorModel TraceErrorModel PeriodicErrorModel ListErrorModel SelectErrorModel SRMErrorModel MrouteErrorModel ErrorModule OTcl class ErrorModel/TwoState ErrorModel/Complex TwoStateMarkov ErrorModel/MultiState ErrorModel/Trace ErrorModel/Periodic ErrorModel/List SelectErrorModel SRMErrorModel ErrorModel/Trace/Mroute ErrorModule Description Error-free and error-prone states Contain two objects of class TwoStateErrorModel Error model with more than two states Impose error based on a trace ﬁle Drop packets once every n packets Specify the a list of packets to be dropped Selective packet drop Error model for SRM Error model for multicast routing Send packets to classiﬁer rather than target_ Error model for PGM Error model for LMS

PGMErrorModel LMSErrorModel

PGMErrorModel LMSErrorModel

Table 12.4. Built-in OTcl error models deﬁned in ﬁle ˜ns/tcl/lib/ns-errmodel.tcl. OTcl class ErrorModel/Uniform ErrorModel/Expo Base class ErrorModel ErrorModel/TwoState Description Uniform error model Two state error model; Each state is represented by an exponential random variable. Two state error model; Each state is represented by an empirical random variable. ErrorModel/Expo model where the state residence time is exponential

ErrorModel/Empirical

ErrorModel/TwoState

ErrorModel/TwoStateMarkov

ErrorModel/Expo

12.3 Built-in Error Models

309

12.3.1 OTcl Implementation: Error Model Conﬁguration In common with those of most objects, conﬁguration interfaces of an error model are deﬁned in the OTcl domain. Such a conﬁguration includes parameter conﬁguration and network conﬁguration. Parameter Conﬁguration There are two ways to conﬁgure an error model object: through bound variables, and through OTcl commands. Class ErrorModel binds the following C++ variables to OTcl instvars with the same name: enabled_ Set to 1 if this error model is active, and set to 0 otherwise. rate_ Error probability delay_pkt_ If set to true, the error model will delay (rather than drop) the transmission of corrupted packets. delay_ Delay time in case that delay_pkt_ is set to true. bandwidth_ Used to compute packet transmission time markecn_ If set to true, the error model will mark error ﬂag (rather than drop) in ﬂag header of the corrupted packet. The second conﬁguration method is through the following OTcl commands whose input arguments are stored in args: unit{arg} ranvar{arg} FECstrength{arg} datapktsize{arg} cntrlpktsize{arg} eventtrace{arg} Store Store Store Store Store Store arg arg arg arg arg arg in in in in in in C++ C++ C++ C++ C++ C++ variable variable variable variable variable variable unit_. ranvar_. FECstrength_. datapktsize_. cntrlpktsize_. et_.

Among the above OTcl commands, commands unit{}, ranvar{}, and FECstrength{}, when taking no input argument, return values stored in unit_, ranvar_, and FECstrength_, respectively. Network Conﬁguration As a Connector object, an error model can be inserted into a network to simulate packet errors. OTcl deﬁnes two pairs of instprocs to insert an error model into a SimpleLink object (see Section 7.1). Each pair consists of one instproc from class SimpleLink and one instproc from class Simulator as shown below (see Fig. 12.7): • SimpleLink::errormodule{em}: Inserts an error model “em” right after instvar head_ of a SimpleLink object.

310

12 Related Helper Classes

Fig. 12.7. Instprocs errormodule and insert-linkloss of class SimpleLink.

• Simulator::lossmodel{lossobj from to}: Executes “errormodule” from within the SimpleLink object which connects node “from” to node “to”. • SimpleLink::insert-linkloss{em}: Inserts an error model “em” right after instvar queue_ (or instvar deqT_ if it exists) of the SimpleLink object. • Simulator::link-lossmodel{lossobj from to}: Executes the instproc “insert-linkloss{...}” from within the SimpleLink object which connects node “from” to node “to”. Program 12.15 shows the details of instproc errormodule{em} of class SimpleLink, which inserts the input error model (e.g., em) immediately after the link’s head. Lines 6–7 store the input error model (i.e., em) in instvar Program 12.15 Instproc errormodule of class SimpleLink, and instproc add-to-head of class Link.
//~/ns/tcl/lib/ns-link.tcl 1 SimpleLink instproc errormodule args { 2 $self instvar errmodule_ queue_ drophead_ 3 if { $args == "" } { 4 return $errmodule_ 5 } 6 set em [lindex $args 0] 7 set errmodule_ $em 8 $self add-to-head $em 9 $em drop-target $drophead_ 10 } 11 Link instproc add-to-head { connector } { 12 $self instvar head_ 13 $connector target [$head_ target] 14 $head_ target $connector 15 }

12.3 Built-in Error Models

311

errmodule_. Line 8 inserts the input error model next to the link’s head by invoking instproc add-to-head{em}, and Line 9 sets the drop target of the input error model em to drophead_. In Lines 11–15 of Program 12.15, instproc add-to-head{connector} inserts the input argument connector between link’s head (i.e., the instvar head_) and target of the link’s head (see Lines 13–14). Program 12.16 shows the details of instproc insert-linkloss{em}, which inserts the input error model after instvar queue_ or instvar deqT_. Line 6 stores the input error model in variable em. Lines 7–9 delete instvar link_errmodule_ if it exists. Then Line 10 stores variable em in instvar link_errmodule_. If instvar deqT_ exists (i.e., trace is enabled), Lines 12–13 insert the input variable em immediately after instvar deqT_. Otherwise, Lines 15-16 insert the input variable em immediately after instvar queue_. Finally, Line 18 sets the drop target of the input variable em to instvar drophead_. Program 12.16 An instproc insert-linkloss of class SimpleLink.
//~/ns/tcl/lib/ns-link.tcl 1 SimpleLink instproc insert-linkloss args { 2 $self instvar link_errmodule_ queue_ drophead_ deqT_ 3 if { $args == "" } { 4 return $link_errmodule_ 5 } 6 set em [lindex $args 0] 7 if [info exists link_errmodule_] { 8 delete link_errmodule_ 9 } 10 set link_errmodule_ $em 11 if [info exists deqT_] { 12 $em target [$deqT_ target] 13 $deqT_ target $em 14 } else { 15 $em target [$queue_ target] 16 $queue_ target $em 17 } 18 $em drop-target $drophead_ 19 }

In most cases, a SimpleLink object is inaccessible from a Tcl simulation script. Therefore, class Simulator provides interface instprocs lossmodel{...} and link-lossmodel{...} to invoke instprocs errormodule{em} and insertlinkloss{em}, respectively, of class SimpleLink. The details of both the instproc lossmodel{lossobj from to} and the instproc link-lossmodel{lossobj from to} of class Simulator are shown in Program 12.17, where they insert an error model “lossobj” into the link

312

12 Related Helper Classes

Program 12.17 Instprocs lossmodel, link-lossmodel, and link of class Simulator.
1 2 3 4 5 6 7 8 //~/ns/tcl/lib/ns-lib.tcl Simulator instproc lossmodel {lossobj from to} { set link [$self link $from $to] $link errormodule $lossobj } Simulator instproc link-lossmodel {lossobj from to} { set link [$self link $from $to] $link insert-linkloss $lossobj }

9 Simulator instproc link { n1 n2 } { 10 $self instvar Node_ link_ 11 if { ![catch "$n1 info class Node"] } { 12 set n1 [$n1 id] 13 } 14 if { ![catch "$n2 info class Node"] } { 15 set n2 [$n2 id] 16 } 17 if [info exists link_($n1:$n2)] { 18 return $link_($n1:$n2) 19 } 20 return "" 21 }

which connect a node “from” to a node “to”. Lines 2 and 6 invoke instproc link{from to} of class Simulator. In Line 18, this instproc returns the Link object which connects a node “from” to a node “to”. Lines 3 and 7 then insert an error model into the returned Link object, by executing errormodule{em} and insert-linkloss{em}, respectively. 12.3.2 C++ Implementation: Error Model Simulation The internal mechanism of an error model is speciﬁed in the C++ domain. As shown in Program 12.18, C++ class ErrorModel derives from class Connector. It employs packet forwarding/dropping capabilities (e.g., a variable target_ and a function recv(p,h)) inherited from class Connector, and deﬁne error simulation mechanism. Variables Key variables of class ErrorModel are given below:

12.3 Built-in Error Models

313

Program 12.18 Declaration of class ErrorModel.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 //~/ns/queue/errmodel.h enum ErrorUnit { EU_TIME=0, EU_BYTE, EU_PKT, EU_BIT }; class ErrorModel : public Connector { public: ErrorModel(); virtual void recv(Packet*, Handler*); virtual void reset(); virtual int corrupt(Packet*); inline double rate() { return rate_; } inline ErrorUnit unit() { return unit_; } protected: int enable_; ErrorUnit unit_; double rate_; double delay_; double bandwidth_; RandomVariable *ranvar_; int FECstrength_; int datapktsize_; int cntrlpktsize_; double *cntrlprb_; double *dataprb_; Event intr_; virtual int command(int argc, const char*const* argv); int CorruptPkt(Packet*); int CorruptByte(Packet*); int CorruptBit(Packet*); double PktLength(Packet*); double* ComputeBitErrProb(int); }; //~/ns/queue/errmodel.cc ErrorModel::ErrorModel() : firstTime_(1), unit_(EU_PKT), ranvar_(0), FECstrength_(1) { bind("enable_", &enable_); bind("rate_", &rate_); bind("delay_", &delay_); }

30 31 32 33 34 35

enable_ rate_ delay_ bandwidth_

Set to 1 if this error model is active, and set to 0 otherwise. Error probability Time used to delay (rather than dropping) a corrupted packet Transmission bandwidth used to compute packet transmission time

314

12 Related Helper Classes

unit_ Error unit (EU_TIME, EU_BYTE(default), EU_PKT, or EU_BIT) ranvar_ Random variable which simulates error FECstrength_ Number of bits in a packet which can be corrected datapktsize_ Number of bytes in data payload cntrlpktsize_ Number of bytes in packet header dataprb_ An array whose ith entry is the probability of having at most i corrupted data bits cntrlprb_ An array whose ith entry is the probability of having at most most i corrupted control bits firstTime_ Indicate whether an error has occurred. intr_ A queue callback object (see Section 7.3.3). Variable rate_ speciﬁes the error probability, while the variable unit_ indicates the unit of rate_. If unit_ is packets (i.e., EU_PKT), rate_ will represent packet error probability. If unit_ is bytes (i.e., EU_BYTE) or bits (i.e., EU_BIT), rate_ will represent byte error probability or bit error probability, respectively. Functions Key functions of class ErrorModel are given below: rate() Return the error probability stored in variable rate_. unit() Return the error unit stored in variable unit_. PktLength(p) Return the length (in error units) of the packet p. reset() Set the variable firstTime_ to 1. recv(p,h) Receive a packet p and a handler h. corrupt(p) Return 1/0 if the transmission is in error/not in error. CorruptPkt(p) Return 1/0 if the transmission is in error/not in error. CorruptByte(p) Return 1/0 if the transmission is in error/not in error. CorruptBit(p) Return the number of corrupted bits in error. ComputeBitErrProb(size) Computes the cumulative distribution of having i= {0, · · · , FECstrength_} error bits. Main Mechanism The main mechanism of an ErrorModel object lies within the packet reception function recv(p,h) shown in Program 12.19. When receiving a packet, an ErrorModel object simulates packet error (by invoking function corrupt(p)

12.3 Built-in Error Models

315

Program 12.19 Function recv of class ErrorModel.
//~/ns/queue/errmodel.cc 1 void ErrorModel::recv(Packet* p, Handler* h) 2 { 3 hdr_cmn* ch = hdr_cmn::access(p); 4 int error = corrupt(p); 5 if (h && ((error && drop_) || !target_)) { 6 double delay = Random::uniform(8.0*ch->size()/bandwidth_); 7 if (intr_.uid_ < 0) 8 Scheduler::instance().schedule(h, &intr_, delay); 9 } 10 if (error) { 11 ch->error() |= error; 12 if (drop_) { 13 drop_->recv(p); 14 return; 15 } 16 } 17 if (target_) { 18 target_->recv(p, h); 19 } 20 }

in Line 4 of Program 12.19), and reacts to the error based on the underlying conﬁguration. If an error occurs, Line 11 will mark an error ﬂag in the common packet header. Then if drop_ exists, Lines 13 and 14 will drop the packet and terminate the function. If the packet is not in error, on the other hand, function recv(p,h) will skip Lines 11–15, and will forward the packet to target_ if it exists. A cautionary note: since a corrupted packet will also be forwarded to target_ if drop_ does not exist. NS2 will not show any error but the simulation results might not be correct! Lines 6–8 in Program 12.19 are related to NS2 callback mechanism discussed in Section 7.3.3. Callback mechanism is an NS2 technique to have a downstream object invoke an upstream object along a downstream path. For example, after transmitting a packet, a queue needs to wait until the packet leaves the queue (i.e., wait for a callback signal to release the queue for the blocked state), before commencing another packet transmission. From Section 7.2, a LinkDelay object employs the Scheduler to inform the queue of packet departure (i.e., send a release signal) at the packet departure time. A callback process is implemented by passing the handler (h) of an upstream object (e.g., the queue) along with packet (p) to a downstream object through function recv(p,h). Upon receiving the handler, an NsObject reacts by either (1) passing the handler to its downstream object and hoping that the handler will be dealt with somewhere along the downstream path, or (2) immediately scheduling a callback event at a certain time.

316

12 Related Helper Classes

According to Line 5 in Program 12.19, the ErrorModel object chooses to call back when both of the following conditions are satisﬁed: (i) Handler “h” exists (i.e., non-zero), and (ii) Either (a) Packet is in error and variable drop_ exists, and/or (b) Variable target_ does not exist. Condition (i) occurs when an upstream object passes down the handler “h”, and is waiting for a callback signal. Condition (ii) indicates the case where the ErrorModel object is responsible for sending a callback signal.3 Condition (ii) consists of two following subconditions. One is the case where the packet will be dropped. Another is when target_ does not exist. In these cases, the ErrorModel will be the last object in a downstream path which can deal with the packet, and is therefore, responsible for the callback mechanism. When choosing to callback, Line 8 schedules a callback event after a delay time of “delay” seconds. NS2 assumes that an error can occur in any place in a packet with equal probability. Correspondingly, the time at which an error is materialized is uniformly distributed in [0, txt], where txt is the packet transmission time (Line 6). Simulating Transmission Errors In the previous section, we have discussed how class ErrorModel forwards or drops (or marks with an error ﬂag) packets based on the simulated error. In this section, we will discuss the details of function corrupt(p) which simulates transmission error. Taking a packet pointer p as an input argument, function corrupt(p) returns zero and one if the transmission is simulated not to be and to be in error, respectively. Program 12.20 shows the details of function corrupt(p). The function corrupt(p) always returns zero if the ErrorModel object is disabled (i.e., enable_=0; see Lines 4–5). Given that the ErrorModel object is enabled, function corrupt(p) will return a logic value (i.e., true or false) depending on whether the value returned from functions CorruptPkt(p) in Line 16, CorruptByte(p) in Line 10, CorruptBit(p) in Lines 13-14, and CorruptTime(p) in Line 8 is zero, when unit_ is equal to EU_PKT, EU_BYTE, EU_BIT, and EU_TIME, respectively. Similar to function corrupt(p), these functions return a zero and a non-zero value if the packet is not in error and is in error, respectively. In some cases, the packet error process in a communication link can be modeled as having Bernoulli distribution. Suppose that ranvar_ (Line 16 in Program 12.18) is a random variable which generates uniformly distributed
3

If not, the ErrorModel object will assign the responsibility to its downstream object. In this case, handler “h” should be passed to the downstream object, by invoking target ->recv(p,h).

12.3 Built-in Error Models

317

Program 12.20 Functions corrupt CorruptPkt, CorruptByte, and PktLength of class ErrorModel.
//~/ns/queue/errmodel.cc 1 int ErrorModel::corrupt(Packet* p) 2 { 3 hdr_cmn* ch = HDR_CMN(p); 4 if (enable_ == 0) 5 return 0; 6 switch (unit_) { 7 case EU_TIME: 8 return (CorruptTime(p) != 0); 9 case EU_BYTE: 10 return (CorruptByte(p) != 0); 11 case EU_BIT: 12 ch = hdr_cmn::access(p); 13 ch->errbitcnt() = CorruptBit(p); 14 return (ch->errbitcnt() != 0); 15 default: 16 return (CorruptPkt(p) != 0); 17 } 18 return 0; 19 } 20 int ErrorModel::CorruptPkt(Packet*) 21 { 22 double u = ranvar_ ? ranvar_->value() : Random::uniform(); 23 return (u < rate_); 24 } 25 int ErrorModel::CorruptByte(Packet* p) 26 { 27 double per = 1 - pow(1.0 - rate_, PktLength(p)); 28 double u = ranvar_ ? ranvar_->value() : Random::uniform(); 29 return (u < per); 30 } 31 double ErrorModel::PktLength(Packet* p) 32 { 33 if (unit_ == EU_PKT) 34 return 1; 35 int byte = hdr_cmn::access(p)->size(); 36 if (unit_ == EU_BYTE) 37 return byte; 38 if (unit_ == EU_BIT) 39 return 8.0 * byte; 40 return 8.0 * byte / bandwidth_; 41 }

318

12 Related Helper Classes

Fig. 12.8. Transforming uniform distribution to Bernoulli distribution.

random numbers “u” in the range [0,1]. From Fig. 12.8, “u” could be any point “×” in [0,1] with equal probability. Given a threshold rate_, “u” will be in [0,rate_) with probability rate_. In other words, to have probability of rate_ for an event (e.g., packet error), we need to generate a uniformly distributed random number “u”, and assume the occurrence of the event if and only if u < rate_. Lines 20–41 of Program 12.20 show the details of functions CorruptPkt(p), CorruptByte(p), and pktLength(p). Function CorruptPkt(p) in Lines 20-24 employs the above method (see Fig. 12.8) to simulate packet error. In other words, it generates uniformly distributed random numbers “u” and assumes that a packet is in error if and only if u < rate_. For function CorruptByte(p), variable rate_ represents byte error probability. Line 27 translates byte error probability to packet error probability (per)4 and simulates packet error in the same way as function CorruptPkt(p) does. Function PktLength(p) in Lines 31–40 of Program 12.20 computes the length of a packet in the corresponding unit_. In particular, if unit_ is • EU_PKT, function PktLength(p) will return 1 (see Line 34). • EU_BYTE, function PktLength(p) will return the number of bytes in the packet stored in ﬁeld size_ of common header (see Lines 35-37). • EU_BITS, function PktLength(p) will return the number of bits in the packet (see Line 39). • EU_TIME (if none of the above matches), function PktLength(p) will return the transmission time of the packet (see Line 40). Program 12.21 shows the details of function CorruptBit(p) of class ErrorModel. When this function is called for the ﬁrst time (i.e., firstTime_ is 1), Lines 5 and 6 precompute error probabilities for control header and data payload and store the probabilities in cntrlprb_ and dataprb_, respectively. The computation is achieved via function ComputeBitErrProb(size) which takes the size of control header (i.e., size=cntrlpktsize_) or data payload (i.e., size=datapktsize_) as its input argument. The values stored
4

Packet error probability is 1 − (1−rate )n , where rate is byte error probability and n = PktLength(p) is number of bytes in a packet.

12.3 Built-in Error Models

319

in cntrlprb_[i] and dataprb_[i] denote the probability that at most i bits are in error. Line 7 then sets firstTime_ to zero so that function CorruptBit will skip Lines 5–7 when it is invoked again. Function CorruptBit(p) computes packet error probability based on either dataprb_ or cntrlprb_, not on the packet size speciﬁed in common header. In Line 10, its uses cntrlprb_ and dataprb_ as packet error probability, if the packet size speciﬁed in common header is not less than and less than datapktsize_, respectively. Since the value stored in dptr[i] is the probability that at most i bits are in error, Lines 11–12 increment i until the probability exceeds u and returns i to the caller. In this case, variable i is the number of corrupted bits. The details of function ComputeBitErrProb(size) are shown in Program 12.21. This function takes the packet size as an input argument and returns an array dptr of double whose ith entry contains the probability Program 12.21 Functions CorruptBit and ComputeBitErrProb of class ErrorModel.
//~/ns/queue/errmodel.cc 1 int ErrorModel::CorruptBit(Packet* p) 2 { 3 double u, *dptr; int i; 4 if (firstTime_ && FECstrength_) { 5 cntrlprb_ = ComputeBitErrProb(cntrlpktsize_); 6 dataprb_ = ComputeBitErrProb(datapktsize_); 7 firstTime_ = 0; 8 } 9 u = ranvar_ ? ranvar_->value() : Random::uniform(); 10 dptr = (hdr_cmn::access(p)->size() >= datapktsize_) ? dataprb_ : cntrlprb_; 11 for (i = 0; i < (FECstrength_ + 2); i++) 12 if (dptr[i] > u) break; 13 return(i); 14 } 15 double* ErrorModel::ComputeBitErrProb(int size) 16 { 17 double *dptr; int i; 18 dptr = (double *)calloc((FECstrength_ + 2), sizeof(double)); 19 for (i = 0; i < (FECstrength_ + 1) ; i++) dptr[i] = comb(size, i) * pow(rate_, 20 (double)i) * pow(1.0 - rate_, (double)(size - i)); 21 for (i = 0; i < FECstrength_ ; i++) 22 dptr[i + 1] += dptr[i]; 23 dptr[FECstrength_ + 1] = 1.0; 24 return dptr; 25 }

320

12 Related Helper Classes

of having at most i corrupted bits. Given packet size size, the probability of having exactly i corrupted bits is size (rate )i (1 − rate )size-i, i as shown in Line 20, where rate_ is the bit error probability. Lines 21– 23 compute the cumulative summation of dprt. Note that Line 23 sets dptr[FECstrength_ + 1] to 1.0 since a packet is considered to be in error if the number of corrupted bits is greater than FECstrength_. 12.3.3 Guidelines for Implementing a New Error Model in NS2 In order to implement a new error model in NS2, we need to follow the three steps below: (i) Design and create an error model class in OTcl, C++, or both domains. (ii) Conﬁgure the parameters of the error model object such as error probability (rate_), error unit (unit_), random variable (ranvar_). (iii) Insert an error model into the network (e.g., by using instproc lossmodel{lossobj from to} or instproc link-lossmodel{lossobj from to} of class Simulator). Example 12.8. Consider the simulation script in Program 9.1, which creates a network as shown Fig. 9.3. Include an error model with packet error probability 0.1 for the link connecting nodes n1 and n3. Tcl Simulation Script: 1 2 3 4 5 6 7 set set set set $ns $ns $ns ns [new Simulator] n1 [$ns node] n2 [$ns node] n3 [$ns node] duplex-link $n1 $n2 5Mb 2ms DropTail duplex-link $n2 $n3 5Mb 2ms DropTail duplex-link $n1 $n2 5Mb 2ms DropTail

8 set em [new ErrorModel] 9 $em set rate_ 0.1 10 $em unit pkt 11 $em ranvar [new RandomVariable/Uniform] 12 $em drop-target [new Agent/Null] 13 $ns link-lossmodel $em $n1 $n3 14 15 16 17 18 set set set $ns $ns udp [new Agent/UDP] null [new Agent/Null] cbr [new Application/Traffic/CBR] attach-agent $n1 $udp attach-agent $n3 $null

12.4 Bit Operations in NS2

321

19 20 21 22 23

$cbr attach-agent $udp $ns connect $udp $null $ns at 1.0 "$cbr start" $ns at 100.0 "$cbr stop" $ns run

where Lines 8–13 are included (into the simulation script in Program 9.1) in order to impose error on packet transmission. Note that the OTcl command unit{u} sets variable unit_ to the value corresponding to the input argument u. The possible values of u include “time”, “byte”, “pkt”, and “bit”. Exercise 12.9. From Example 12.8, collect statistics for packets which are in error and not in error. Verify that the packet error probability is 0.1. Adjust simulation time if necessary. How long must your simulation be to ensure the convergence of 0.1 error probability ? • Initially set link bandwidth to be 5 Mbps. • Change the bandwidth to be 500 Kbps. What happen to the measured convergence time ? Explain why. Exercise 12.10. Consider a two state error model, which consists of good and bad states. Packet transmission in a good state is always error free, while packet transmitted in a bad state is always corrupted. The time that an error model stays in good and bad states is exponentially distributed with means tgood and tbad , respectively. Write a simulation script for the above two state error model with tgood = 10 sand tbad = 1 s. Verify the results and show the convergence time.

12.4 Bit Operations in NS2
12.4.1 Bit Masking Bit masking is a bit transformation technique, which can be used for various purposes. Given a mask, a bit masking process transforms an original value to a masked value (see Fig. 12.9). In this section, we will show two examples of bit masking: subnet masking and modulo masking.

Fig. 12.9. Bit masking.

322

12 Related Helper Classes

Subnet Masking A 4-byte IP address can be divided into host address and network address. While a host address identiﬁes a host (e.g., a computer), a network address characterizes a group of hosts. A host is given a host IP address as its identiﬁcation and a 4-byte subnet mask which identiﬁes its network. A subnet mask consists of all-one upper bits and all-zero lower bits (i.e., of format “1 · · · 10 · · · 0”). For a given host IP address and a subnet mask, the network IP address can be determined as follows: Network IP Address = Host IP Address & Subnet Mask where & is a bitwise “AND” operator. Example 12.11. A class-C (i.e., subnet mask = 255.255.255.0) host IP address 10.1.2.3 has the network IP address of (10.1.2.3)&(255.255.255.0) = (10&255).(1&255).(2&255).(3&0) = 10.1.2.0 (12.2) In fact, all class-C IP addresses whose ﬁrst three bytes are 10.1.2 have the same network address. Correspondingly, a class-C network address corresponds to 256 IP addresses. From the above example, the original value (i.e., host IP address) 10.1.2.3 is masked (by using bitwise “and”) with a mask 255.255.255.0 (i.e., class C subnet mask) such that the masked value (i.e., network IP address) is 10.1.2.0. Modulo Masking Modulo is a remainder computation process. Suppose a = b × c + d. Then a%c = d, where % is a modulo operator. Bit masking can also be used as a modulo operator with c = 2n where n is a positive integer. To implement a modulo masking, the upper and lower bits of a modulo mask are set to contiguous zeros and contiguous ones, respectively (i.e., of format “0 · · · 01 · · · 1”), and the masking operation is a bitwise “AND” operation. Suppose, an original value is of format xx..xx, where x can be zero or one. The modulo masking applies bitwise “AND” to an original value and the modulo mask, and obtains the masked value as follows: original value = x · · · xx · · · x upper-bound mask = 0 · · · 01 · · · 1 masked value = 0 · · · 0x · · · x. Suppose the number of one-bits of a modulo mask is n. The bits whose positions are greater than n are removed during a masking process, and the (12.1)

12.4 Bit Operations in NS2

323

masked value is bounded by 2n −1. On the other hand, the bits whose positions are not greater than n are kept unchanged. These lower order bits in fact represent the remainder when the original value is divided by 2n . Modulo masking is therefore equivalent to a modulo operation. Exercise 12.12. Let a modulo mask be 64. Show that the modulo masking and modulo operation are equivalent for the following original values: 63, 64, 65, 127, 128, and 129. Exercise 12.13. Consider a ball color-number matching experiment, where balls are fed one-by-one to an observer. Each ball is masked with a color and a number. The color can be either black or white, while the unique number is increased one-by-one as the balls are fed to the observer. From time to time, the observer is given a number and is asked to identify the color of one of the 64 most recently observed balls. Design a memory-friendly approach for the observation. We summarize the masking components of subnet masking and modulo masking in Table 12.5. Note that, both subnet masking and modulo masking use a bitwise “AND” as their mask operation. Since their masks are diﬀerent, the implications for their masked value are diﬀerent. 12.4.2 Bit Shifting and Decimal Multiplication Another important bit operation is bit shifting which is equivalent to decimal multiplication. If a binary value is shifted to the left by n bits, the corresponding decimal value will increase by 2n times. Similarly, a binary number right shifted by n bits returns the quotient of the decimal value divided by 2n . To prove the above statement, consider an arbitrary value y = M xm 2m , m=0 where xm ∈ {0, 1}, m = {0, · · · , M }. Let y round(Scheduler::instance().clock()), s, d, name, th->size(), flags, iph->flowid(), src_nodeaddr, src_portaddr, dst_nodeaddr, dst_portaddr, seqno, th->uid(), tcph->ackno(), tcph->flags(), tcph->hlen(), tcph->sa_length() ); delete [] src_nodeaddr; delete [] src_portaddr; delete [] dst_nodeaddr; delete [] dst_portaddr; }

26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50

15). While channel_ is an interface to a Tcl channel, wrk_ is a buﬀer which stores a trace string to be written to the Tcl channel. At the construction, the Tcl channel channel_ is set to Null, and the trace string wrk_ is allocated with memory space which can hold upto 1026 characters. Key functions of class BaseTrace include channel(...), buffer(), flush (channel), and dump(). The operations of the ﬁrst three functions are fairly straightforward, and are omitted for brevity. Function dump() shown in Lines 28–37 of Program 13.17 is responsible for dumping a trace string stored in wrk_ to the Tcl channel. Here, Line 30 retrieves the length of the string wrk and stores the length in a local variable “n”. Line 32 attaches an end-ofline character to wrk_. Line 33 attaches zero to wrk_ indicating the end of the string. Line 34 writes wrk_ to the Tcl channel channel_ using function Tcl_Write(...). Finally, Line 35 clears the value stored in wrk_. In common with class Trace, class BaseTrace has three main OTcl commands: flush{}, detach{}, and attach{file}. These three commands per-

352

13 Processing an NS2 Simulation

Program 13.16 Function command of class Trace.
//~/ns/trace/trace.cc 1 int Trace::command(int argc, const char*const* argv) 2 { 3 Tcl& tcl = Tcl::instance(); 4 if (argc == 2) { 5 if (strcmp(argv[1], "flush") == 0) { 6 Tcl_Channel ch = pt_->channel(); 7 if (ch != 0) 8 pt_->flush(ch); 9 return (TCL_OK); 10 } 11 if (strcmp(argv[1], "detach") == 0) { 12 pt_->channel(0) ; 13 return (TCL_OK); 14 } 15 } else if (argc == 3) { 16 if (strcmp(argv[1], "attach") == 0) { 17 int mode; 18 const char* id = argv[2]; 19 Tcl_Channel ch = Tcl_GetChannel(tcl.interp(), (char*)id,&mode); 20 pt_->channel(ch); 21 if (pt_->channel() == 0) { 22 tcl.resultf("trace: can’t attach %s for writing", id); 23 return (TCL_ERROR); 24 } 25 return (TCL_OK); 26 } 27 } 28 return (Connector::command(argc, argv)); 29 }

form the same action as those in class Trace. We will omit the details of these three OTcl commands for brevity. 13.3.4 Various Types of Packet Tracing Objects NS2 employs diﬀerent types of packet tracing objects to trace packets at diﬀerent places. For example, a Trace/Enque object is placed immediately before a queue to trace packets which enter the queue. The type (i.e., variable type_) of a Trace/Enque object is “+”, which is equivalent to 43 in decimal. When a packet passes through a Trace/Enque object, a line beginning with “+” is appended to the Tcl Channel.

13.3 Packet Tracing

353

Program 13.17 Declaration, an OTcl binding class, the constructor of class BaseTrace, and function dump of class BaseTrace.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 //~/ns/trace/basetrace.h class BaseTrace : public TclObject { public: BaseTrace(); ~BaseTrace(); virtual int command(int argc, const char*const* argv); virtual void dump(); inline Tcl_Channel channel() { return channel_; } inline void channel(Tcl_Channel ch) {channel_ = ch; } inline char* buffer() { return wrk_ ; } void flush(Tcl_Channel channel) { Tcl_Flush(channel); } #define PRECISION 1.0E+6 #define TIME_FORMAT "%.15g" protected: Tcl_Channel channel_; char *wrk_; };

//~/ns/trace/basetrace.cc 17 class BaseTraceClass : public TclClass { 18 public: 19 BaseTraceClass() : TclClass("BaseTrace") { } 20 TclObject* create(int argc, const char*const* argv) { 21 return (new BaseTrace()); 22 } 23 } basetrace_class; 24 BaseTrace::BaseTrace() : channel_(0), 25 { 26 wrk_ = new char[1026]; 27 } 28 void BaseTrace::dump() 29 { 30 int n = strlen(wrk_); 31 if ((n > 0) && (channel_ != 0)) { 32 wrk_[n] = ’\n’; 33 wrk_[n + 1] = 0; 34 (void)Tcl_Write(channel_, wrk_, n + 1); 35 wrk_[n] = 0; 36 } 37 }

354

13 Processing an NS2 Simulation

Among all built-in OTcl packet tracing classes, the most common ones include • • • • Trace/Enque (“+”): Trace packet arrival (usually at a queue) Trace/Deque (“-”): Trace packet departure (usually at a queue) Trace/Drop (“d”): Trace packet drop (delivered to a drop-target) Trace/Recv (“r”): Trace packet reception at a certain node

where the characters in the parentheses are attributed to each packet tracing object class. Program 13.18 Constructors of classes Trace, Trace/Enque, and Trace/Deque.
1 2 3 4 5 6 7 //~/ns/tcl/lib/ns-trace.tcl Class Trace/Enque -superclass Trace Trace/Enque instproc init {} { $self next "+" } Trace/Deque instproc init {} { $self next "-" }

//~/ns/trace/trace.h 8 static class DequeTraceClass : public TclClass { 9 public: 10 DequeTraceClass() : TclClass("Trace/Deque") { } 11 TclObject* create(int args, const char*const* argv) { 12 if (args >= 5) 13 return (new DequeTrace(*argv[4])); 14 return NULL; 15 } 16 } dequetrace_class;

Among these four classes, only class Trace/Deque has an implementation in the C++ domain. Other three classes are not bound to the OTcl domain. The main diﬀerence among the above four packet tracing objects lie in their constructors. As shown in Program 13.18, OTcl class Trace/Enque derives from the OTcl class Trace (Line 1), while class OTcl Trace/Deque is mapped to the C++ OTcl class DequeTrace (Lines 8–16). Lines 3 and 6 show that classes Trace/Enque and Trace/Deque are constructed with characters “+” and “-”, respectively. In Line 24 of Program 13.12, this character is stored in the variable type_ of the packet tracing object. As an example, consider the process of creating a Trace/Enque object in Fig. 13.6. The process starts when a statement “new Trace/Enque” is executed. From within an OTcl constructor, the type “+” is repeatedly fed to the

13.3 Packet Tracing

355

Fig. 13.6. Construction of a Trace/Enque object.

constructor up the hierarchy by the statement “$self next "+"”. When class SplitObject is reached, instproc create-shadow{...} is invoked with an input argument “+”. Instproc create-shadow{...} invokes function create() of class TraceClass in the C++ domain. From Line 24 in Program 13.12, the constructor of class Trace is invoked, and type “+” is fed as an input argument. Since the constructor takes an integer as an input argument, the ascii code “+” is converted into a decimal value “43”. Finally, the constructor stores the input argument (i.e., “43” in this case) in the variable type_. 13.3.5 Packet Trace Format Packet trace format is deﬁned in function format(...) (Programs 13.1413.15). In a normal case, each line of a trace ﬁle follows the format in Fig. 13.7. There are 12 ﬁelds in each line of a trace ﬁle (i.e., a Tcl channel):

Fig. 13.7. Packet tracing ﬁle format

• Type Identiﬁer: depends on the type (i.e., variable type_) of packet tracing object which generates the string. Most widely used type identiﬁers are shown below. The complete list of type identiﬁers is given in ﬁle ˜ns/tcl/lib/ns-trace.tcl.

356

13 Processing an NS2 Simulation

– – – – • • • • •

• •

• •

“+” which represents a packet enque event, “-” which represents a packet deque event, “r” which represents a packet reception event, “d” which represents a packet drop (e.g., sent to dropHead_) event, and – “c” which represents a packet collision at the MAC level. Time: at which the packet tracing string is created. Source Node and Destination Node: denote the IDs of the source and the destination nodes of the tracing object. Packet Name: Name of the packet type (as speciﬁed in Program 8.9). Packet Size: Size of the packet in bytes. Flags: A 7-digit ﬂag string is deﬁned in Lines 9–21 of Program 13.14. Each ﬂag digit is set to “-” if the corresponding ﬂag is disabled. Otherwise, it will be set as follows. The ﬁrst is set to “E” if an ECN (Explicit Congestion Notiﬁcation) echo is enabled. The second is set to “P” if the priority in the IP header is enabled. The fourth is set to “A” if the corresponding TCP takes an action on a congestion (e.g., closes the congestion window). The ﬁfth is set to “E” if the congestion has occurred. The sixth is set to “F” if the TCP fast start is used. Finally, the seventh is set to “N”, when the transport layer protocol is capable of using Explicit Congestion Notiﬁcation (ECN). Flow ID: Flow ID speciﬁed in the ﬁeld fid_ of an IP packet header. Source Address and Destination Address: The source and destination addresses of a packet speciﬁed in an IP packet header. For a ﬂat addressing scheme, the format of these two ﬁelds is “a.b”, where “a” is the address and “b” is the port. Sequence Number: The sequence number speciﬁed in packet header. Speciﬁed by a transport layes protocol. Packet Unique ID: A unique ID stored in a common packet header.

13.4 Compilation of Simulation Results
One of the main objectives of network simulation is to study network performance. Compilation of simulation results refers to a process of collecting information from simulation and compute performance measures under consideration. There are three main approaches to collect simulation results in NS2: through C++ codes, through Tcl codes, and through a trace ﬁle. • Through C++ codes: This refers to an approach which inserts C++ codes into the original NS2 codes. As mentioned earlier in this book, the modiﬁcation of C++ code results in a quick simulation. However, programmers require a fair amount of knowledge in the C++ architecture to collect results from the simulation.

13.4 Compilation of Simulation Results

357

• Through Tcl codes: This method is perhaps the most convenient way to collect the results. The programmers do not need to know the details of the C++ architecture. They only need to know the variable binding structure of classes under consideration. • Through trace ﬁle: This method consists of two main steps. In the ﬁrst step, a trace ﬁle is created during simulation. The second step is to retrieve the relevant information from the trace ﬁle. In most cases, a scripting language (e.g., AWK) can be used to extract the necessary information from a trace ﬁle (see Appendix A). Although this approach is widely demonstrated in the NS2 tutorial in the internet, advanced users are not encouraged to use this approach due to the following reasons. First, the OTcl command “trace-all” consumes a signiﬁcant amount of resources (e.g., memory, simulation time), and dramatically slows down the simulation. Secondly, a generated trace ﬁle usually contains too much information. In most cases, an NS2 user need to learn another scripting language (e.g., AWK) to extract relevant information from a trace ﬁle. Finally, the trace ﬁle may not contain the required information. For example, information on instantaneous buﬀer occupancy is not available in a trace ﬁle. Example 13.5. Consider Example 10.1 which creates the network in Fig. 9.3. Insert an error model with error probability 0.05 into the link connecting Node 1 and Node 3. Suppose the maximum TCP transmission window size is set to 20. • Through C++ result codes: Find out the number of times TCP transmission window is reduced. • Through Tcl codes: Plot the dynamic variation of TCP transmission window. • Through trace ﬁle: Compute the average interval between two TCP packets entering the link layer buﬀer. Constructing a Network An error model can be inserted into the network by inserting the following OTcl codes immediately after Line 7 of Example 10.1: set $em $em $em $em $ns em [new ErrorModel] set rate_ 0.005 unit pkt ranvar [new RandomVariable/Uniform] drop-target [new Agent/Null] lossmodel $em $n1 $n3

The maximum TCP transmission window is set to 20 by the following statement after Line 10 in Example 10.1: “ $tcp set window_ 20”.

358

13 Processing an NS2 Simulation

Collection of Results Through C++ Codes TCP shrinks its transmission window when function slowdown(how) of class TcpAgent is invoked. Therefore, we may declare a variable num_slowdowns_ of class TcpAgent in ﬁle ˜ns/tcp/tcp.h, initialize it to zero in the constructor, and add the two following lines in function slowdown(how): num_slowdowns_++; printf("Total number of TCP window reduction is %d \n", now, num_slowdowns_); After recompiling NS2, we run the script “tcp.tcl” and obtain the following results: >> ns Total Total Total ... Total tcp.tcl number of TCP window reduction is 1 number of TCP window reduction is 2 number of TCP window reduction is 3 number of TCP window reduction is 36

In this simulation, TCP shrinks its transmission window 36 times. Collection of Results Through Tcl Codes Transmission window size of a TCP connection is the minimum of instvars cwnd_ and window_ of a Agent/TCP object. Since these two variables are available in the OTcl domain, we may collect samples of TCP window size by inserting the following Tcl script after Line 14 in Example 10.1. 1 2 3 4 5 6 7 8 9 10 11 set f_cwnd [open cwnd.tr w] proc plot_tcp { } { global f_cwnd tcp ns if { [$tcp set cwnd_] < [$tcp set window_] } { puts $f_cwnd "[$ns now] [$tcp set cwnd_]" } else { puts $f_cwnd "[$ns now] [$tcp set window_]" } $ns at [expr [$ns now] + 0.2] plot_tcp } $ns at 0.01 "plot_tcp"

The above statements put time and TCP transmission window size in ﬁle “cwnd.tr” every 0.2 seconds. Line 1 above creates a Tcl channel f_cwnd which is bound to the ﬁle cwnd.tr. Lines 2–10 deﬁne a procedure plot_tcp{}. Lines 11 invokes procedure plot_tcp at 0.01 second. Within the procedure plot_tcp{}, Lines 5 and 7 print instvars cwnd_ and window_, whichever is less, on the Tcl channel f_cwnd. Line 9 schedules an invocation of procedure

13.4 Compilation of Simulation Results
25

359

Transmission window size

20

15

10

5

0

0

2

4

6 Time

8

10

Fig. 13.8. Dynamics of TCP transmission window for Example 13.5.

plot_tcp at 0.2 seconds in future. This invocation continuously prints out simulation time and TCP transmission window size to the Tcl channel until the simulation terminates. After running the above Tcl simulation script, the ﬁle cwnd.tr is created. The ﬁrst and the second columns of ﬁle cwnd.tr are the time and the corresponding TCP transmission window, respectively. We now plot Fig. 13.8, using the ﬁrst and second columns as X axis and Y axis, respectively. Since we set instvar window_ to be 20, TCP transmission window can never exceed 20. We can also observe frequent decreases in TCP transmission window size due to packet losses. Collection of Results Through Trace File The ﬁrst step in this process is to enable tracing in the Tcl simulation script. Again, this step can be carried out by inserting the following lines after Line 4 in Example 10.1. set f_trace [open trace.tr w] $ns trace-all $f_trace The second step is to process the trace ﬁle. In this case, there is only one TCP ﬂow in the simulation and we can measure the interval between two TCP packets entering a queue, which connect Node 1 (with ID 0) to Node 3 (with ID 1), using the AWK script ﬁle avg.awk in Program 13.19. By executing the AWK script, we will see the following result on the screen: >> awk -f avg.awk trace.tr Average TCP packet inter-arrival time is 0.001703

360

13 Processing an NS2 Simulation

Program 13.19 An AWK script which computes average interval between two TCP packets entering a link layer buﬀer of Node 1.
//avg.awk 1 BEGIN{ started = 0 } 2 /^+/ { time = $2; 3 if (started == 1) { 4 if ($3==0 && $4==2 && $5 == "tcp") { 5 interval = time-old_time; 6 old_time = time; 7 cum_interval += interval; 8 total_samples ++; 9 } 10 } else { 11 started = 1; old_time = time; 12 } 13 } 14 END { avg_interval = cum_interval/total_samples; 15 printf("Average TCP packet inter-arrival time is %f\n", avg_interval); 16 }

Line 1 in Program 13.19 initializes variable started to zero. Lines 2–13 collect samples of the inter-arrival time of TCP packets. Line 2 indicates the actions to be executed for all the lines beginning with “+” in the subsequent curly braces. From Line 4, the samples are collected only for the source node 0, the destination node 2, and protocol tcp. Finally, Lines 14–16 compute and print the average TCP packet inter-arrival time on the screen.

13.5 Chapter Summary
Two of the most important aspects in a network simulation are debugging and compilation of simulation results. Debugging refers to a procession of removing compilation and run-time errors in both C++ and OTcl domains. This chapter provides guidelines and necessary commands for debugging. Although originally designed to facilitate the understanding of network dynamics, NS2 tracing could also be useful in the debugging process. NS2 supports two types of tracing. Variable tracing records the changes in value of a variable (in most cases in a ﬁle), while packet tracing stores the details of packets passing through network checkpoints (again in most cases in a ﬁle). There are three major ways to collect simulation results. First, collecting simulation results through C++ codes is a quick and easy way. However, the users may require a fair amount of knowledge on the C++ architecture. Also, since this method involves the modiﬁcation of C++ code, it could mess up the original NS2 source codes. The upside of this approach is that it gives users

13.5 Chapter Summary

361

a access to most NS2 components. At runtime, the simulation proceeds very fast since the modiﬁcation is carried out using the C++ compiler. Secondly, collecting simulation results through Tcl codes allows the programmer to collect the results from the OTcl domain in a simple way. In this case, the users do not need to understand the entire architecture of NS2, but they need to know how the variables in C++ and OTcl domains are bound together. Since NS2 is written mostly in C++, some variables are inaccessible from the OTcl domain. This approach may not be able to collect all required performance measures. Despite its necessity, this approach does not provide an access to NS2 internal variables (which might be needed in some case). Proceeding by the interpreter, this approach can take long runtime compared to the ﬁrst approach. Finally, collecting simulation results through a trace ﬁle consists of two steps: 1) running simulation to create a trace ﬁle and 2) processing the created trace ﬁle. Despite its prevalance in the on-line tutuorial, this method is not recommended since it takes too much simulation resource and might not give the users required information. In fact, the recommended method is the ﬁrst one (using C++).

14 Developing New Modules for NS2

So far, we have explained the details of the basic components of NS2 including their functionalities, internal mechanisms, and conﬁguration methods. In this ﬁnal chapter, we demonstrate how new NS2 modules are created, conﬁgured, and incorporated through two following examples. One is an Automatic Repeat reQuest (ARQ) protocol, which is a mechanism to improve transmission reliability of a communication link by means of packet retransmission. Another is a packet scheduler which arranges the transmission sequence of packets from multiple incoming data ﬂows.

14.1 Automatic Repeat reQuest (ARQ)
Automatic Repeat reQuest (ARQ) is a method of handling communication errors by packet retransmission. An ARQ transmitter (i.e., a transmitting node which implements an ARQ protocol) is responsible for transmitting data packets and retransmitting the lost packets. An ARQ receiver (i.e., a receiving node which implements an ARQ protocol), on the other hand, is responsible for receiving packets and (implicitly or explicitly) informing the transmitter of the transmission result. It returns an ACK (acknowledgement) message or a NACK (negative acknowledgement) message to the transmitter if a packet is successfully or unsuccessfully (respectively) received. Based on the received ACK/NACK pattern, the ARQ transmitter decides whether to retransmit the lost packet or to transmit a new packet. This section focuses on a limited-persistence stop-and-wait ARQ protocol. This type of ARQ protocols is characterized by two following properties. With limited-persistence, an ARQ transmitter gives up the retransmission if the transmission fails consecutively for a certain number of times. Another property is “stop-and-wait”. Here, an ARQ transmitter transmits a packets and waits for an acknowledgement from the corresponding ARQ receiver before commencing another (lost or new) packet transmission.

T. Issariyakul, E. Hossain, Introduction to Network Simulator NS2, DOI: 10.1007/978-0-387-71760-9 14, c Springer Science+Business Media, LLC 2009

364

14 Developing New Modules for NS2

In the following, we ﬁrst design NS2 modules for a limited-persistence stop-and-wait ARQ protocol with an error-free and delay-free (i.e., immediate) feedback channel in Section 14.1.1. Sections 14.1.2 and 14.1.3 demonstrate C++ and OTcl implementations, respectively. Finally, Section 14.1.4, we extend the ARQ model for an error-free feedback channel with non-zero processing and propagation delay. Implementation of an ARQ protocol with an error prone feedback channel is left as an exercise for the readers. 14.1.1 The Design Figure 14.1 shows an architecture of a link with an ARQ protocol. Here, the feedback channel is assumed to be error-free and the feedbacks are assumed to be immediate. The link is constructed by inserting an error module and an ARQ module into a SimpleLink object. From Fig. 7.1, a SimpleLink object consists of four main instvars: queue_ which models the packet buﬀering, link_ which models the service time of the queue and the link propagation delay, ttl_ which models time-to-live of a packet, and drophead_ which acts as a common dropping point for a SimpleLink object. An error model link_errmodule_ is inserted into a SimpleLink object by an OTcl command link-lossmodel{...} of class Simulator. Based on this basic SimpleLink object with an error model, we incorporate the three following ARQ components (i.e., instvars) to implement an ARQ module. The ﬁrst component is an ARQ transmitter (instvar tARQ_), which transmits, retransmits, and drops the packet based on the underlying ARQ protocol. The second and third components (instvars acker_ and nacker_) are responsible for transmitting ACK and NACK messages, respectively, to the ARQ transmitter tARQ_. NS2 employs a queue blocking and callback mechanism to model packet forwarding in a SimpleLink object. The process starts when a Queue object queue_ receives and forwards a packet as well as its queue handler (whose class is QueueHandler) to its downstream object, and blocks itself. Here, the Queue object stops transmitting packets until the head-of-the-line packet is completely transmitted (when it is unblocked by the downstream object).

Fig. 14.1. Architecture of a SimpleLink object with an ARQ module.

14.1 Automatic Repeat reQuest (ARQ)

365

In the absence of instvar link_errmodule_ and the ARQ-related instvars, instvar link_ (of class LinkDelay) is responsible for unblocking the Queue object. It does so by placing the input queue handler on the simulation time line at the time where the packet is completely transmitted (i.e., has left the Queue object). At the ﬁring time, the queue handler is dispatched, and the Queue object is unblocked. At this moment, the Queue object is allowed to transmit another packet. Instvar link_errmodule_ is of C++ class ErrorModel (see Section 12.3) and is responsible for simulating packet errors. A packet will be forwarded to the variables drop_ and target_, respectively, depending on whether it is in error or not. We now incorporate the instvars tARQ_, acker_, and nacker_ into the packet forwarding mechanism as follows: • ACK/NACK message passing: The key components of the ACK/NACK message passing mechanism are instvars acker_ and nacker_, which are responsible for creating and forwarding ACK and NACK messages (respectively) to an ARQ transmitter. From Fig. 14.1, these two components are attached to variables target_ and drop_, respectively, of instvar link_errmodule_. A packet will be forwarded to the instvars acker_ and nacker_, respectively, depending on whether the packet is in error or not in error, respectively. Instvar acker_ informs the ARQ transmitter of a transmission success, and forwards the received packet to the instvar link_, while nacker_ drops the corrupted packet, and informs the ARQ transmitter of transmission failure. • A callback mechanism: In case of a SimpleLink object, instprocs link_ and link_errmodule_ are responsible for the callback mechanism. When inserting ARQ components, the callback mechanism is modiﬁed as follows. Instvars link_ and link_errmodule_ call back to an ARQ transmitter (i.e., tARQ_) which in turns calls back to a Queue object (i.e., queue_). Upon receiving a packet and a queue handler from the Queue object, the ARQ transmitter stores the queue handler in its member variable, and transmits the received packet as well as its handler to the downstream object. Depending on whether the packet is in error or not in error, the link_errmodule_ and link_ (respectively) will place a callback event on the simulation timeline. At the same time, nacker_ and acker_ will inform the ARQ transmitter of the transmission result. At the ﬁring time (when the packet is completely transmitted), the ARQ transmitter determines whether the packet was successfully transmitted or not. Then, it decides whether to retransmit the lost packet or to fetch another packet from the upstream Queue object based on the received ACK/NACK messages. 14.1.2 C++ Implementation In Fig. 14.1, the ARQ-related instvars include an ARQ transmitter tARQ_, an ACK message transmitter acker_, and a NACK message transmitter

366

14 Developing New Modules for NS2

nacker_. These instvars are implemented in the C++ classes ARQTx, ARQAcker, and ARQNacker, respectively, which are bound to OTcl classes with the same name. Implementations of these three classes are shown in Programs 14.1–14.4. Program 14.1 Declaration of classes ARQTx and ARQHandler
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 //arq.h class ARQTx; enum ARQStatus {IDLE,SENT,ACKED,RTX,DROPPED}; class ARQHandler : public Handler { public: ARQHandler(ARQTx& arq) : arq_tx_(arq) {}; void handle(Event*); private: ARQTx& arq_tx_; }; class ARQTx : public Connector { public: ARQTx(); void recv(Packet*, Handler*); void nack(Packet*); void ack(); void resume(); protected: ARQHandler arqh_; Handler* qh_; Packet* pkt_; ARQStatus status_; int blocked_; int retry_limit_; int num_rtxs_; };

Class ARQTx Class ARQTx derives from class Connector, and can be used to connect two NsObjects.1 The main C++ variables of class ARQTx are shown below: num_rtxs_ Current number of packet retransmission; It is increased by one for every transmission failure, and is reset to zero when a new packet arrives (e.g., due to a packet drop or a transmission success).
1

In Fig. 14.1, we use an ARQTx object tARQ to connect a Queue object queue with an ErrorModule object link errmodule .

14.1 Automatic Repeat reQuest (ARQ)

367

Program 14.2 Functions of classes ARQTx and ARQHandler
26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 //arq.cc void ARQHandler::handle(Event*){arq_tx_.resume();} ARQTx::ARQTx() : arqh_(*this) { num_rtxs_ = 0; retry_limit_ = 0; qh_ = 0;pkt_ = 0; status_ = IDLE; blocked_ = 0; bind("retry_limit_", &retry_limit_); } void ARQTx::recv(Packet* p, Handler* h) { qh_ = h; status_ = SENT; blocked_ = 1; send(p,&arqh_); } void ARQTx::ack() { num_rtxs_ = 0; status_ = ACKED; } void ARQTx::nack(Packet* p) { num_rtxs_++; if( num_rtxs_ handle(0); } else if ( status_ == RTX ) { status_ = SENT; blocked_ = 1; send(pkt_,&arqh_); } else if ( status_ == DROPPED ) { status_ = IDLE; drop(pkt_); qh_->handle(0); } }

368

14 Developing New Modules for NS2

retry_limit_ The retry limit; The ARQ protocol will retransmit the lost packet as long as num_rtxs_handle(0) (Line 57 in Program 14.2). If status_ is DROPPED, it will drop the packet stored in pkt_, set status_ to be IDLE and fetch another packet from the upstream Queue object by invoking qh_->handle(0) (Lines 62–63 in Program 14.2). Finally, if status_ is RTX, function resume() will block the ARQ transmitter and forward packet *pkt_ as well as its handler *arqh_ to the downstream object (Lines 59–60 in Program 14.2). Classes ARQRx, ARQAcker, and ARQNacker Another part of ARQ implementation is an ARQ receiver, which is responsible for reacting to packet transmission from the ARQ transmitter. Here, ARQ receivers are represented by a C++ class ARQRx, which contains a pointer arq_tx_ (see Line 72 in Fig. 14.3) to an ARQ transmitter (i.e., an ARQTx object). This pointer is initialized to zero at the object construction, and is associated with an ARQ transmitter by OTcl command attach-ARQTx (Lines 104–107 in Program 14.4). Class ARQRx declares function recv(p,h) as pure virtual (see Line 70 in Program 14.3) to force its derived classes to implement this function. There are two classes derived from class ARQRx: classes ARQAcker and ARQNacker. These two classes are responsible for sending ACK and NACK messages, respectively, to the associated ARQ transmitter. Upon receiving a packet, class ARQAcker (see Lines 110–114 in Program 14.4) informs the associated ARQ transmitter of successful packet delivery by invoking function ack() associated with the pointer arq_tx_ (see the detail of function ack()

370

14 Developing New Modules for NS2

Program 14.3 Declaration of classes ARQRx, ARQAcker, and ARQNacker, and their OTcl classes
66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 //arq.h class ARQRx : public Connector { public: ARQRx(); int command(int argc, const char*const* argv); virtual void recv(Packet*, Handler*)=0; protected: ARQTx* arq_tx_; }; class ARQAcker : public ARQRx { public: ARQAcker() {}; virtual void recv(Packet*, Handler*); }; class ARQNacker : public ARQRx { public: ARQNacker() {}; virtual void recv(Packet*, Handler*); }; //arq.cc static class ARQAckerClass: public TclClass { public: ARQAckerClass() : TclClass("ARQAcker") {} TclObject* create(int, const char*const*) { return (new ARQAcker); } } class_arq_acker; static class ARQNackerClass: public TclClass { public: ARQNackerClass() : TclClass("ARQNacker") {} TclObject* create(int, const char*const*) { return (new ARQNacker); } } class_arq_nacker;

84 85 86 87 88 89 90 91 92 93 94 95 96 97 98

14.1 Automatic Repeat reQuest (ARQ)

371

Program 14.4 Functions of classes ARQRx, ARQAcker, and ARQNacker
99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 //arq.cc ARQRx::ARQRx() { arq_tx_ = 0; } int ARQRx::command(int argc, const char*const* argv) { Tcl& tcl = Tcl::instance(); if (argc == 3) { if (strcmp(argv[1], "attach-ARQTx") == 0) { arq_tx_ = (ARQTx*)TclObject::lookup(argv[2]); return(TCL_OK); } } return Connector::command(argc, argv); } void ARQAcker::recv(Packet* p, Handler* h) { arq_tx_->ack(); send(p,h); } void ARQNacker::recv(Packet* p, Handler* h) { arq_tx_->nack(p); }

in Lines 40–43 of Program 14.2). Then, it sends out the received packet to its downstream NsObject. Similarly, class ARQNacker (see Lines 115–118 in Program 14.4) informs the associated ARQ transmitter of transmission failure by invoking function nack(p) associated with the pointer arq_tx_ (see function nack(p) in Lines 44–52 of Program 14.12). 14.1.3 OTcl Implementation In the OTcl domain, we need to create ARQTx, ARQAcker, and ARQNack objects– tARQ_, acker_, and nacker_, respectively, and insert them into a SimpleLink object as shown in Fig. 14.1. Program 14.5 shows two OTcl instprocs developed for this purpose. • SimpleLink::link-arq{limit}: This instproc creates the ARQ-related instances and conﬁgures the SimpleLink object as shown in Fig. 14.1. Lines 4–6 create instvars tARQ_, acker_, and nacker_. Line 7 stores the input argument “limit” in variable retry_limit_ of tARQ_. Lines 8 and 9 associate acker_ and nacker_, respectively, with tARQ_. Finally, Lines 10–15 conﬁgure the rest of components as shown in Fig. 14.1. • Simulator::link-arq{limit from to}: This instproc is an interface instproc which creates and conﬁgures ARQ modules of the link connecting Node “from” to Node “to”. The input argument limit here is used as the retry limit of the ARQ module.

372

14 Developing New Modules for NS2

Program 14.5 OTcl Instprocs for an ARQ Module
//~ns/tcl/lib/ns-link.tcl 1 SimpleLink instproc link-arq { limit } { 2 $self instvar link_ link_errmodule_ queue_ drophead_ 3 $self instvar tARQ_ acker_ nacker_ 4 set tARQ_ [new ARQTx] 5 set acker_ [new ARQAcker] 6 set nacker_ [new ARQNacker] 7 $tARQ_ set retry_limit_ $limit 8 $acker_ attach-ARQTx $tARQ_ 9 $nacker_ attach-ARQTx $tARQ_ 10 $queue_ target $tARQ_ 11 $tARQ_ target $link_errmodule_ 12 $link_errmodule_ target $acker_ 13 $acker_ target $link_ 14 $tARQ_ drop-target $drophead_ 15 $link_errmodule_ drop-target $nacker_ 16 } //~ns/tcl/lib/ns-lib.tcl 17 Simulator instproc link-arq {limit from to} { 18 set link [$self link $from $to] 19 $link link-arq $limit 20 }

Example 14.1. We now setup an experiment to show the impact of retry limit of a limited-persistence stop-and-wait ARQ protocol on TCP throughput. Our experiment is based on Example 10.1. We insert an error module with 0.3 error probability in the link connecting Node n1 to Node n3, and implement a limited-persistence ARQ over this lossy link, and plot TCP throughput versus the retry limit. Tcl Simulation Script We insert the following codes in the Tcl simulation script ﬁle “tcp.tcl” in Example 10.1: 1 2 3 4 5 6 7 //tcp.tcl set em [new ErrorModel] $em set rate_ 0.3 $em unit pkt $em ranvar [new RandomVariable/Uniform] $em drop-target [new Agent/Null] $ns link-lossmodel $em $n1 $n3 $ns link-arq 3 $n1 $n3

14.1 Automatic Repeat reQuest (ARQ)

373

8 proc show_tcp_seqno {} { 9 global tcp 10 puts "The final tcp sequence number is [$tcp set t_seqno_]" 11 } 12 13 14 15 $ns $ns $ns $ns at 0.0 "$ftp start" at 100.0 "show_tcp_seqno" at 100.1 "$ns halt" run

Here, Lines 1–6 create an error module with packet error probability 0.3, and insert the created error module immediately after instvar queue_ of the link connecting Node n1 and Node n3. Line 7 creates and conﬁgures ARQrelated components with retry limit of 3. We run the simulation for 50.1 seconds and collect the results when the simulation time is 50.0. After running the script ﬁle “tcp.tcl” above, the following result appear on the screen: >> ns tcp.tcl >> The final tcp sequence number is 37587 TCP throughput in packets per second is computed as the ﬁnal TCP sequence number divided by the simulation time. We vary the retry limit (in Line 7 above) to {0, 1, 2, 3}, and plot TCP throughput in Fig. 14.4. Clearly, increasing retry limit increases link reliability, and therefore, increases TCP throughput.
400 TCP throughput (packets seconds) 350 300 250 200 150 100 50 0 0 1 Retry limit 2 3

Fig. 14.4. Impact of retry limit of a limited persistent ARQ protocol on TCP throughput.

374

14 Developing New Modules for NS2

14.1.4 ARQ Under a Delayed (Error-Free) Feedback Channel We have developed an NS2 module for an ARQ protocol with an immediate and error-free feedback. In practice, a feedback channel would be nonimmediate and/or error prone. This section extends the modules developed earlier for a non-immediate error-free feedback channel. The extension for a non-immediate and error-prone feedback channel is left for the reader as an exercise (Exercise 14.4). Program 14.6 shows the details of class ARQRx modiﬁed to support a delayed feedback channel. The idea is to defer the generation of ACK/NACK message for delay_ seconds, where the variable delay_ is bound to an instProgram 14.6 Modiﬁcation of class ARQRx for a limited-persistence ARQ protocol with a delayed feedback channel.
//arq.h 1 class ARQRx : public Connector { 2 public: 3 virtual void recv(Packet*, Handler*); 4 virtual void handle(Event*); 5 virtual void resume()=0; 6 protected: 7 ARQTx* arq_tx_; 8 Packet *pkt_; 9 Handler *handler_; 10 double delay_; 11 Event event_; 12 }; //arq.cc ARQRx::ARQRx() { arq_tx_ = 0; pkt_ = 0; handler_ = 0; bind("delay_", &delay_); } void ARQRx::handle(Event *e) {resume();} void ARQRx::recv(Packet* p, Handler* h) { pkt_ = p; handler_ = h; Scheduler::instance().schedule(this, &event_, delay_); } void ARQAcker::resume() { arq_tx_->ack(); send(pkt_,handler_); } void ARQNacker::resume() {arq_tx_->nack(pkt_);}

13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29

14.1 Automatic Repeat reQuest (ARQ)

375

var with the same name in the OTcl domain (see Line 16). Deﬁned in Lines 19–23, function recv(p,h) invokes function schedule(this,&event_,delay_) of class Scheduler to defer the generation of ACK/NACK message. At the ﬁring time, the Scheduler dispatches the scheduled event by invoking function handle(e) of the ARQRx object. In Line 18, function handle(...) invokes the pure virtual function resume() to resume the pending actions. Deﬁned in classes ARQAcker (Lines 24–28) and ARQNacker (Line 29), function resume() simply sends either an ACK message or a NACK message, respectively, to the attached ARQTx object. Note that class ARQRx also deﬁnes a variable event_ of class Event which is used as an ACK/NACK reception dummy event (see also Section 4.3.6). In the OTcl domain, we only need to include the two following lines into instproc link-arq{limit} of class SimpleLink (e.g., after Line 6 in Program 14.5): $acker_ set delay_ [$self delay] $nacker_ set delay_ [$self delay] Here, the link delay in the forward direction (returned from $self delay) is used as the ARQ feedback delay for both ACK and NACK generators (i.e., acker_ and nacker_, respectively). Example 14.2. Compare the TCP throughputs for the cases with an immediate feedback channel and a delayed feedback channel in the link layer ARQ protocols. Here, we use the results in Example 14.1 as a benchmark. When rerunning the Tcl simulation script in Example 14.1 under the ARQ protocol with a delayed feedback channel, the following result should appear on the screen: >> ns tcp.tcl >> The final tcp sequence number is 20596 which is less than 37587 in Example 14.1. The readers are encourage to experiment with diﬀerent input parameters (e.g., feedback delay or retry limit) to gain more insights into the impact of link layer ARQ protocols on TCP performance. Exercise 14.3. Why class TimerHandler was not used to implement the delayed feedback channel? Exercise 14.4. Based on Examples 14.1 and 14.2, modify the ARQ protocol as follows. (i) Remove the variable nacker_, and use a timer-based retransmission mechanism: A packet is assumed to be lost unless an ACK message is received within a timeout period. (ii) Develop the codes for an ARQ protocol with an error prone delayed feedback channel.

376

14 Developing New Modules for NS2

14.2 Packet Scheduling for Multi-Flow Data Transmission
Packet scheduling is a mechanism to arrange transmission sequence of incoming packets in a node. For example, a round-robin (RR) packet scheduler transmits packets from diﬀerent ﬂows in sequence. This section shows the implementation of a round-robin packet scheduler in NS2. 14.2.1 The Design Figure 14.5 shows the architecture of a packet scheduler in NS2. Here, the packet scheduler is implemented in an OTcl class LinkSch, which is modiﬁed from class SimpleLink.

Fig. 14.5. Architecture of a LinkSch object.

Key Diﬀerences between Class LinkSch and Class SimpleLink Class LinkSch deﬁnes two addition components–a ﬂow classiﬁer flow_clsfr_ and a packet scheduler sch_, and modiﬁes one component of class SimpleLink– it is instvar queue_. • Flow classiﬁer flow_clsfr_ examines packet header and forwards packets with the same ﬂow ID to the same forwarding NsObject.2 • Packet scheduler sch_ takes a packet from one of the attached upstream data ﬂows and forwards the packet to its downstream object. It complies with an underlying packet scheduling protocol to take a packet from a certain ﬂow.
2

Flow classiﬁers are implemented in a C++ class FidHashClassifier. However, we do not use the built-in C++ class, since we would like to show how to implement a new C++ class.

14.2 Packet Scheduling for Multi-Flow Data Transmission

377

• Modiﬁed instvar queue_: Instvar queue_ is scalar in class SimpleLink. However, it is used as an associative array in class LinkSch. Its index and value are, respectively, the ﬂow ID and the Queue object which stores packets of the corresponding ﬂow ID. Since each element of queue_ contains packets with the same ﬂow ID, we use the terms “ﬂow ID” and “queue ID” interchangeably. Packet Flow Mechanism When a packet enters a LinkSch object, it is sorted by a ﬂow classiﬁer, flow_clsfr_, which forwards packets from the same ﬂow to the same queue. In particular, the ﬂow classiﬁer forwards packets with ﬂow ID “i” to instvar “queue_[i]”, as shown in Fig. 14.5. Each element of queue_ forwards packets to the packet scheduler sch_ according to the underlying mechanism deﬁned in the C++ domain (e.g., class Queue). Based on an underlying scheduling mechanism, the packet scheduler sch_ takes a packet from one of these queues and forwards it to a LinkDelay object, link_. Callback Mechanism The packet scheduler breaks a callback connection between a Queue object queue_ and a LinkDelay object link_ into two connections. One is between instvar queue_ and instvar sch_ and another is between instvar sch_ and instvar link_. Instead of calling back to instvar queue_, instvar link_ reports (i.e., calls back) to the packet scheduler sch_ to indicate that it is ready to receive another packet. Upon receiving a call back message, the packet scheduler selects the next transmission ﬂow based on its underlying scheduling discipline, and fetches another packet from the selected ﬂow (i.e., an element of queue_). Every element of queue_ deactivates the queue blocking mechanism, since they do not need to wait before sending a packet to a packet scheduler. Rather, such a blocking (i.e., waiting) mechanism is implemented in the packet scheduler. Under this call back mechanism, instvar link_ calls back to the packet scheduler (rather than a Queue object) to indicate the completion of packet transmission. 14.2.2 C++ Implementation In the C++ domain, we deﬁne two new NS2 components – ﬂow classiﬁers and packet schedulers in C++ classes FlowClassifier and PktScheduler, respectively. Class FlowClassifier deﬁnes how a ﬂow classiﬁer forwards packets with the same ﬂow ID to the same NsObject. Class PktScheduler is a base class from which more speciﬁc packet scheduler classes derive. As an example, we develop a C++ class RRScheduler which is a derived class of class PktScheduler to represent round-robin packet schedulers.

378

14 Developing New Modules for NS2

Program 14.7 C++ Implementation of class FlowClassifier
1 2 3 4 //classifier-flow.h class FlowClassifier : public Classifier { protected: int classify(Packet *p); }; //classifier-flow.cc static class FlowClassifierClass : public TclClass { public: FlowClassifierClass() : TclClass("Classifier/Flow") {} TclObject* create(int, const char*const*) { return (new FlowClassifier()); } } class_flow_classifier; int FlowClassifier::classify(Packet *p) { return hdr_ip::access(p)->flowid(); }

5 6 7 8 9 10 11 12 13 14 15

Here, we bind the C++ class FlowClassifier to an OTcl class Classifier/Flow, but do not bind the C++ class PktScheduler. However, we bind a C++ class RRScheduler, a child class of class PktScheduler, to an OTcl class PktScheduler/RR. Flow Classiﬁers A ﬂow classiﬁer is represented by a C++ class FlowClassifier implementation of which is shown in Program 14.7. Class FlowClassifier is bound to an OTcl class Classifier/Flow (see Lines 5–10). Derived from class Classifier, class FlowClassifier overrides function classify(p) by returning the ﬂow ID speciﬁed in the header of packet p* (Line 14). Packet Schedulers The main responsibility of a packet scheduler is to determine transmission sequence of the attached upstream Queue objects. In this section, we assume that each Queue object holds packets of the same ﬂow ID and the packet scheduler determines the transmission sequence based on the ﬂow ID only. Packet schedulers are implemented using a C++ class PktScheduler, declaration and implementation of which are shown in Programs 14.8 and 14.9, respectively. From Program 14.8, class PktScheduler has one constant and four key variables:

14.2 Packet Scheduling for Multi-Flow Data Transmission

379

Program 14.8 Declaration of a C++ class PktScheduler
16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 //pkt-sched.h #define MAX_FLOWS 10 class PktScheduler : public Connector { public: PktScheduler(); virtual void handle(Event*); virtual void recv(Packet*, Handler*); protected: void send(int fid, Handler* h); virtual void resume(); int getFlowID(Packet* p) {return hdr_ip::access(p)->flowid();}; virtual int nextID() = 0; Handler* qh_[MAX_FLOWS]; Packet* pkt_[MAX_FLOWS]; int blocked_; int active_flow_id_; };

MAX_FLOW The maximum number of queues which can be attached to the packet scheduler. blocked_ Set to “1” if the packet scheduler is in the “blocked” state, and set to “0” otherwise. active_flow_id_ The ﬂow ID of the packet being transmitted pkt_[i] The HOL packet of the queue corresponding to ﬂow “i” qh_[i] The QueueHandler object of the queue corresponding to ﬂow “i” Class PktScheduler has two main tasks. One is to determine the transmission sequence for all attached data ﬂow. Another is to insert itself in the middle of a callback connection between a Queue object and a LinkDelay object. While the ﬁrst task is implemented in function nextID(), the second task is attributed to functions recv(p,h) and resume(). Taking no input argument, function nextID() returns the next transmitting ﬂow ID based on the underlying scheduling discipline. Class PktScheduler declares this function as pure virtual, and leaves the detailed implementation to its derived classes (Line 26 in Program 14.8). As an example, we will show how a round-robin packet scheduler implements this function later in this section. The details of functions recv(p,h) and resume() are shown in Lines 41–50 and 57–67 of Program 14.9. Function recv(p,h) is the main packet reception function. Function recv(p,h) ﬁrst determines the ﬂow ID of packet *p by invoking function getFlow(p) in Line 43. Line 44 stores the input packet *p and the input handler *h in variable pkt_[fid] and qh_[fid], respectively, where fid is the ID of the packet *p. If the PktScheduler object is not blocked, Line 46 will send the head of the line packet stored in pkt_[fid] to

380

14 Developing New Modules for NS2

Program 14.9 Functions of a C++ class PktScheduler
32 33 34 35 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 //pkt-sched.cc PktScheduler::PktScheduler() { int i; for (i=0;ihandle(0); PktScheduler::PktScheduler() { int i; for (i=0;ihandle(0); int index = nextID(); blocked_ = 0; if (index >= 0) { send(index,this); blocked_ = 1; active_flow_id_ = index; } }

14.2 Packet Scheduling for Multi-Flow Data Transmission

381

the downstream object, and reset pkt_[fid] to zero (see Lines 51–55). Note that the PktScheduler object passes its address (i.e., this) rather than the input handler to the downstream object. Line 47 blocks the PktScheduler object and Line 48 stores the ﬂow ID of the packet under transmission in variable active_flow_id_. Upon receiving a packet *p and a handler *h, a LinkDelay object schedules a packet departure event. Since the received handler belongs to the above PktScheduler object, at the ﬁring time, function handle(e) of the PktScheduler object is invoked. In Line 56, function handle(e) simply invokes function resume(), the details of which are shown in Lines 57–67. Line 59 ﬁrst fetches a packet (for which transmission has just been ﬁnished) from the ﬂow by invoking function handle(e) of queue_[active_flow_id_]. Line 60 determines the next transmitting ﬂow based on the underlying scheduling discipline. Finally, Lines 61–66 forward the selected packet to the downstream object (similar to Lines 45–49). As an example, consider an implementation of round-robin schedulers, which transmit packets from each ﬂow sequentially. We implement this type of schedulers using a C++ class RRScheduler which is bound to an OTcl class Scheduler/RR. The declaration and implementation of the C++ class RRScheduler are shown in Program 14.10. Class RRScheduler has one variable current_id_ and one function NextID(). The variable current_id_ records the most recently selected ﬂow ID. Based on the round-robin scheduling principle, class RRScheduler overrides function nextID() by returning the next ID, whose corresponding Queue object contains at least one packet. If the Queue objects do not contain any packet, this function will return −1 (see Lines 95–108). 14.2.3 OTcl Implementation In the OTcl domain, we put together the components of a link with a scheduler, as shown in Fig. 14.5. Again, the major diﬀerences of class LinkSch and class SimpleLink lie in the instvars flow_clsfr_, queue_, and sch_ of class LinkSch. Instvar flow_clsfr_ is a ﬂow classiﬁer (whose OTcl class is Classifier/Flow). It forwards incoming packets with ﬂow ID “i” to the NsObject stored in the slot number “i”. Instvar queue_ is an array of Queue objects. Here, queue_[fid] is installed in the slot corresponding to ﬂow ID “fid”. Finally, instvar sch_ is a round-robin packet scheduler instantiated from an OTcl class PktScheduler/RR. Programs 14.11–14.12 show the OTcl implementation of a link with a scheduler. The implementation involves two OTcl classes: LinkSch and Simulator. Similar to class SimpleLink, class LinkSch derives from class Link. In addition to those deﬁned in class SimpleLink, the following instvars are deﬁned in class LinkSch:

382

14 Developing New Modules for NS2

Program 14.10 C++ implementation of c Class RRScheduler
68 69 70 71 72 73 74 //pkt-sched.h class RRScheduler : public PktScheduler { public: RRScheduler() ; private: virtual int nextID(); int current_id_; }; //pkt-sched.cc RRScheduler::RRScheduler() { current_id_ = -1; } static class RRSchedulerClass: public TclClass { public: RRSchedulerClass() : TclClass("PktScheduler/RR") {} TclObject* create(int, const char*const*) { return (new RRScheduler()); } } class_rr_scheduler; int RRScheduler::nextID() { int count = 0; current_id_++;current_id_ %= MAX_FLOWS; while((pkt_[current_id_] == 0)&&(count> ns rr.tcl 1 The final tcp(0) sequence number is 60110 >> ns rr.tcl 3 The final tcp(0) sequence number is 20052 The final tcp(1) sequence number is 20051 The final tcp(2) sequence number is 20051 The TCP throughput is computed by the ﬁnal sequence number divided by the simulation time. Since the simulation time here is 100 seconds (see

386

14 Developing New Modules for NS2

Line 28), the throughput of TCP ﬂow “0” is 610.1 packets/sec and 200.52 packets/sec when the number of TCP ﬂows is 1 and 3, respectively. With a round-robin scheduler, each element of the array queue_ has equal chance to transmit packets. In principle, each TCP ﬂow should have the same throughput performance (as shown above). Also, the throughput in case of n TCP ﬂows should be approximately n times less than that in case of single TCP ﬂow. From the above result, the per-ﬂow TCP throughput in case of 3 TCP ﬂows is almost the same as each other, and is approximately one third of TCP throughput in case of single TCP ﬂow (i.e., (60110/100)/3 = 601.10/3 = 200.37). Next, we run the above Tcl simulation script for 1 to 10 TCP ﬂows. We compare the average TCP throughput and the fair share TCP throughput in Fig. 14.6. Here, we deﬁne the fair share TCP throughput for n TCP ﬂows as γ/n, where γ is the TCP throughput in case of single TCP ﬂow. We observe that both average and fair share throughput are almost inline with each other. We also observe that TCP throughput for each ﬂow is very similar to each other. These two observations validate the round-robin operation, which treats every TCP ﬂow equally. Exercise 14.6. A Weighted Fair Queue (WFQ) packet scheduler gives fair access to every data ﬂow. Under a WFQ packet scheduler, each data ﬂow gains channel access in proportion to its weight. The algorithm for WFQbased packet scheduling can be found in [25]. Develop a module for a WFQ packet scheduler. Validate the module by plotting the results in a graph.
700 Per−flow average TCP throughput (packets per second) 600 500 400 300 200 100 0 Fair share Average throughput

1

2

3

4

5

6

7

8

9

10

Number of TCP flows

Fig. 14.6. Impact of number of TCP ﬂows on per-ﬂow throughput under roundrobin packet scheduling.

14.3 Chapter Summary

387

14.3 Chapter Summary
This ﬁnal chapter has demonstrated how new modules are created, conﬁgured, and incorporated into NS2. Two examples provided here include: an Automatic Repeat reQuest (ARQ)-based error recovery module and a packet scheduling module. In most of the cases, we need to develop NS2 codes in both C++ and OTcl domains. In the C++ domain, the main task is to deﬁne the internal mechanisms of the new NS2 components. The main task in the OTcl domain, on the other hand, are to integrate the developed NS2 components into the existing NS2 modules, and to instantiate and conﬁgure the newly developed modules from a Tcl simulation script.

A Programming Essentials

This appendix covers the basic elements of the programming languages, which are essential for developing NS2 simulation programs. These include Tcl/OTcl which is the basic building block of NS2 and AWK which can be used for post simulation analysis.

A.1 Tcl Programming
Tcl is a general purpose scripting language. While it can do anything other languages could possibly do, its integration with other languages has proven even more powerful. Tcl runs on most of the platforms such as Unix, Windows, and Mac. The strength of Tcl is its simplicity. It is not necessary to declare a data type for variable prior to the usage. At runtime, Tcl interprets the codes line by line and converts the string into appropriate data type (e.g., integer) on the ﬂy. A.1.1 Program Invocation Tcl can be invoked from a shell command prompt with the following syntax: tclsh [ ...] where tclsh is mandatory. Other input arguments are optional. When the above command is invoked without input argument, the shell enters Tcl environment where it waits for the Tcl statements line by line. If is speciﬁed, Tcl will interpret the text speciﬁed in the ﬁle whose name is line by line. In addition, if ... are speciﬁed, they will be placed in a list variable (see Section A.1.3) argv. In the main program, can be retrieved by executing “lindex $argv $i”.

390

A Programming Essentials

A.1.2 A Simple Example To get a feeling about the language, we look at Example A.1 below: Example A.1. The following Tcl script, “convert.tcl”, converts temperatures from Fahrenheit to Celsius. The conversion starts at 0 degree (Fahrenheit), proceeds with a step of 25 degrees (Fahrenheit), and stops when the temperature exceeds 140 degrees (Fahrenheit). The program prints out the converted temperature in Celsius as long as the temperature in Fahrenheit does not exceed 140 degrees. # convert.tcl # Fahrenheit to Celsius Conversion 1 proc tempconv {} { 2 set lower 0 3 set upper 140 4 set step 25 5 set fahr $lower 6 while {fahr < $upper} { 7 set celsius [expr 5*($fahr - 32)/9] 8 puts "Fahrenheit / Celsius : $fahr / $celsius" 9 set fahr [expr $fahr + $step] 10 } 11 } The details of the above example are as follows. The symbol # here denotes the beginning of a line comment. The reserved word proc in Line 1 declares a procedure tempconv{} which takes no input argument. The procedure also deﬁnes four local variables (i.e., lower, upper, step, and fahr) and assigns values to them using the reserved word set followed by the name and its assigned value (Lines 2–5). Note here that, to refer to the value of a variable, the reserved character $ is used in front of the variable (e.g., set fahr $lower). The keyword expr in Line 9 informs the Tcl interpreter to interpret the following string as a mathematical expression. The while loop in Lines 6–10 controls the iteration of the procedure through the test expression enclosed in a double quotation mark. The command puts in Line 8 prints out the string contained within the quotation mark. If the name of the script is convert.tcl, the script can be executed by typing the following on a shell prompt: >>tclsh convert.tcl Fahrenheit / Celsius Fahrenheit / Celsius Fahrenheit / Celsius Fahrenheit / Celsius Fahrenheit / Celsius Fahrenheit / Celsius : : : : : : 0 / -17.778 25 / -3.889 50 / 10 75 / 23.889 100 / 37.778 125 / 51.667

A.1 Tcl Programming

391

Alternatively, since NS2 is written in Tcl, the following invocation would lead to the same result. >>ns convert.tcl A.1.3 Variables and Data Types Data Types As an interpreter, Tcl does not need to deﬁne data type of variables. Instead, it stores everything in string and interprets them based on the context. Example A.2. Consider the following Tcl codes: # 1 2 3 4 5 6 7 8 vars.tcl set a "10+1" set b "5" set c $a$b set d [expr $a$b] puts $c puts $d unset c puts $c

After executing the Tcl script “vars.tcl”, the following result should appear on the screen: >>tclsh vars.tcl 10+15 25 Here, variable c is simply a string “10+15”, whereas variable d is 25 obtained by numerically evaluating the string “10+15” stored in variable c. Therefore, we may conclude that everything is treated as a string unless speciﬁed otherwise [26]. Variable Assignment and Retrieval Tcl stores a value in a variable using the reserved word “set”. The value stored in a variable can be retrieved by placing a character “$” in front of a variable name. In addition, a reserved word “unset” is used to clear the value stored in a variable. Example A.3. Insert the following two lines into the end of the codes in Example A.2. 7 unset c 8 puts $c

392

A Programming Essentials

After executing the Tcl script “vars.tcl”, the following result should appear on the screen: >>tclsh vars.tcl 10+15 25 can’t read "c": no such variable while executing "puts c" (file "var.tcl" line 8) Clearly after being unset, variable c stores nothing. Printing the variable would result in a runtime error. Bracketing There are four type of bracketing in Tcl. These are used to group a series of strings. Tcl interprets strings inside diﬀerent types of bracket diﬀerently. Suppose a variable $var stores a value 10. Tcl interprets a statement “expr $var + 1” with four diﬀerent bracketing diﬀerently. • Curly braces ({expr $var + 1}): Tcl interprets this statement as it is. • Quotation marks ("expr $var + 1"): Tcl interpolates the variable var in the string. This statement would be interpreted as “expr 10 + 1”. • Square brackets ([expr $var + 1]): Tcl regards a square bracket in the same way that C++ regards a parenthesis. It interprets the string in a square bracket before interpreting the entire line. This statement would be interpreted as “11”. • Parentheses ((expr $var + 1)): Tcl uses a parentheses for indexing an array and for invoking built-in mathematical function. Example A.4. Insert the following two lines into the end of the codes in Example A.2. 7 puts -nonewline {{}: } 8 puts {expr $c} 9 puts -nonewline {"": } 10 puts "expr $c" 11 puts -nonewline {[]: } 12 puts [expr $c] After executing the Tcl script “vars.tcl”, the following result should appear on the screen: >>tclsh vars.tcl 10+15 25

A.1 Tcl Programming

393

{}: expr $c "": expr 10+15 []: 25 When bracketing with “{}”, Tcl interprets the string as it is; The result in this case is “expr $c”. The string $c is replaced with its value when bracketing with “""”. The result in this case is “expr 10+15”. Finally, “[]” identiﬁes the sequence of execution. The string “expr $c” is executed ﬁrst. The result in this case is “25”. Global Variables In Example A.1, we brieﬂy mentioned about local variables. But what was missing there is the notion of global variables. Global variables are common and used extensively throughout a program. These variables can be called upon by any procedure in the program. Example A.5 shows an example use of global variables. Example A.5 (Global variables). set PI 3.1415926536 proc perimeter {radius} { global PI expr 2*$PI*$radius } Since “PI” is deﬁned outside of the procedure perimeter, the keyword global is used here to make “PI” global and available within the procedure. When called upon, this procedure simply calculates the perimeter of a circle based on the supplied input radius. Finally, we note here that no default values are automatically assigned to variables. Any attempt to call an uninitialized variable would lead to a runtime error. Array An array is a special variable which can be used to store a collection of items. An array stores both the indexes and the values as strings. For example, index “0” is not a number, but a numeric string. By default, an array in Tcl is an associative array. Example A.6 below shows various ways of string manipulation. Example A.6 (Array assignment). # Numeric indexing set arr(0) 1 set arr(1) 3 set arr(1) 5

394

A Programming Essentials

# String indexing set wlan(datarate) 54000000 set wlan(protocol) "tcp"

Lists A list is an ordered collection of elements such as numbers, strings or even lists themselves. The key list manipulations are shown below: • List creation: A list can be created in various ways as shown in Example A.7 below. Example A.7. The following two statement are equivalent (i) set mylist "1 2 3" (ii) set mylist {1 2 3} From the above, a list can be created in three ways. First, it can be created by the reserved word list which takes list members as input arguments. Alternatively, it can be created by embracing the members within a pair of curly braces or a pair of quotation marks. • Member retrieval: The following command returns the nth (= {0, 1, · · ·}) element in a list mylist: lindex $mylist $n • Member setting: The following command sets the value of the nth element in a list mylist to be : lset $mylist $n $value • Group retrieval: The following command returns a list whose members are the nth member through the mth member of a list mylist: lrange $mylist $n $m • Appending the list: The following command attaches a list alist to the end of a list mylist: lappend $mylist $alist A.1.4 Input/Output Tcl employs a so-called Tcl channel to receive an input using a command gets or to send an output using a command puts.

A.1 Tcl Programming

395

Tcl Channels A Tcl channel refers to an interface used to interact to the outside world. Two main types of Tcl channels include standard reading/writing channels and ﬁle channels. The former are classiﬁed into stdin for reading, stdout for writing, and stderr for error reporting. The latter needs to be attached to a ﬁle before it is usable. The syntax for attaching a ﬁle to a Tcl ﬁle channel is shown below: open [] This command returns a Tcl channel attached to a ﬁle with the name . The optional input argument could be “r” for reading, “w” for writing to a new ﬁle, or “a” for appending an existing ﬁle. When a Tcl channel is no longer in use, it can be closed by using the command close whose syntax is as follows: close where is the Tcl channel which need to be closed. The Commands gets and puts The command puts and gets reads and writes, respectively, a message to a speciﬁed Tcl channel. In particular, the command “gets” reads a line from a Tcl channel, and passes every character in the line except the end-of-line character to the Tcl running environment. The Tcl channel could be a standard channel or a ﬁle channel. The syntax of the command gets is as follows: gets Here, all the characters in the current line from the channel channel will be stored in the variable . The command “puts” writes a string followed by an end-ofline character to a Tcl channel . If is not speciﬁed, the stdout will be used as a default channel. The syntax of the command puts is as follows: puts [-nonewline] ][ where the nonewline option above speciﬁes not to write an end-of-line character to the end of the string. Normally, the command puts does not output immediately onto a Tcl channel. Instead, it puts the input argument (i.e., string) in its buﬀer, and releases the stored string either when the buﬀer is full or when the channel is closed. To force the immediate outputting, flush is used. Note that while a standard channel is opened and closed on the ﬂy (i.e., upon an invocation of “puts), a ﬁle channel needs to be closed explicitly using the command close.

396

A Programming Essentials

Example A.8. Consider the following Tcl codes: puts "Press any key to continue..." gets stdin set ch_in [open "input.txt" "r}] set ch_out [open "output.txt" "a"] set line_no 0 while {[gets $ch_in line] >= 0} { puts $ch_out "[incr line_no] $line" } close $ch_in close $ch_out In this example, the content of ﬁle input.txt is copied to ﬁle output.txt line by line. In addition, the line number is preﬁxed at the beginning of each new line. A.1.5 Mathematical Expressions Tcl implements mathematical expressions through mathematical operators and mathematical functions. A mathematical expression of either type must be preceded by a reserved word “expr”. Otherwise, Tcl will recognize the operator as a character (see Lines 1 and 4 in Example A.2). A mathematical expression consists of an operator and operands. A list of most widely used operators is given in Table A.1. An operand can be either ﬂoating-point, octal or hexadecimal numbers. To be evaluated as octal and hexadecimal numbers, the numbers must be preceded by 0 and 0x, respectively. As another means to implement mathematical operations, mathematics functions can be placed after the reserved word “expr”. The built-in mathematical functions are shown below, where the input argument of a function is enclosed by parentheses.
Table A.1. Tcl mathematical operators. Operators −+ ∼! ∗/% +− > = & ∧ | && || x?y : z Usage Unary minus, unary plus, bit-wise negation, logical negation Multiplication, division, remainder Addition, subtraction Bit shift left, right Less than, greater than, less than or equal, greater than or equal Bit-wise AND Bit-wise exclusive OR Bit-wise OR Logical AND Logical exclusive OR If x is non-zero, then y. Otherwise, z.

A.1 Tcl Programming

397

abs(x) acos(x) asin(x) atan(x) atan2(x) ceil(x) cos(x)

cosh(x) double(x) exp(x) floor(x) fmod(x) hypot(x,y) int(x)

log(x) log10(x) pow(x,y) rand(x) round(x) sin(x) sinh(x)

sqrt(x) srand(x) tan(x) tanh(x) wide(x)

The detail of all the above functions is given in [27] Example A.9. Examples of invocation of mathematical functions log10(x) and abs(x) are shown below. >>tclsh >>expr log10(10) 1.0 >>expr abs(-10) 10 >>expr 1+2 3 A.1.6 Control Structure Tcl control structure deﬁnes how the program proceeds. This is carried out using the commands if/else/elseif, switch, for, while, foreach, and break/continue if/else/elseif An if/else/elseif command provides a program with a selective choice. A general form of this command is shown below: if {} { } elseif {} { } . . . else { } Here, the command ﬁrst checks whether condition1 in the if statement is true. If so, it will take actions_1. Otherwise, it will check whether condition2 in the elseif statement is true. If so, it will take actions_2. If not, the process continues for every elseif statement. If nothing matches, actions_n deﬁned under the else condition will be taken.

398

A Programming Essentials

switch The switch command is a good substitute of a long series of a if/else/elseif command. It checks the value of a variable against given patterns, and takes actions associated with a matched pattern. The structure of a switch command is shown below: switch { { } { } . . . default { } } In this case action_i will be taken if matches with where i= { 1,2,...,n-1}. If none of the predeﬁned patterns matches with the value, the default actions (i.e., actions_n) will be taken. while/for/foreach The commands while, for, and foreach are used when actions need to be repeated for several times. The command while repeats actions until a predeﬁned condition is no longer true. The command for repeats the actions for a given number of times. The command foreach repeats the actions for every item in a given list. The syntax of these three commands are as follows: while {} { } for {} {} {} { } foreach {} {} { }

A.1 Tcl Programming

399

The while command repeats the as long as the is true. The for command begins with an initialization statement init. After taking , it executes the Tcl statement and checks whether the is true. If so, it will repeat the . Otherwise, the command will terminate. The command foreach repeats for every member in the . In each repetition, the member is stored in variable and can be used for various purposes. break/continue Commands break and continue are used in looping structures while, for, and foreach. They are used to prematurely stop the looping mechanism. Their key diﬀerence is that while the command break immediately exits the loop, the command continue simply restarts the loop. Example A.10. set var 0 while {$var < 100} { puts $var set var [expr $var+5] if {$var == 20} break } puts $var In this example, the loop continues as long as $var < 100. However, the command break terminates the looping mechanism if $var == 20. Therefore, the above program will print out 20. If the reserved word break is replaced with a reserved word continue, the loop will restart after being stopped. In this case the program will print out 100. A.1.7 Procedures A procedure is usually used in place of a series of Tcl statements to tidy up the program. The syntax of a procedure is shown below: proc { ... } { [return ] } The deﬁnition of a procedures begins with a reserved word proc. The procedure name is placed after the word proc. The input arguments are placed within a curly braces, located immediately after the procedure name. Embracing with a curly braces, the main body placed next to the input argument.

400

A Programming Essentials

Here, the actions for the procedures are deﬁned. Optionally, the procedure may return a , using a reserved word return. After deﬁning a procedure, one may invoke the procedure by executing the following statement: set var [ ] where var is set to the value returned from the procedure , and the values are fed as input arguments of the procedure.

A.2 Objected Oriented Tcl (OTcl) Programming
OTcl is an object-oriented version of Tcl, just like C++ is an object-oriented version of C [28]. The basic architecture and syntax in OTcl are much the same as those in Tcl. The diﬀerence, however, is the philosophy behind each of them. In OTcl, the concepts of classes and objects are of great importance. A class is a representation of a group of objects which share the same behavior(s) or trait(s). Such a behavior can be passed down to child classes. In this respect, the donor and the receiver of the behaviors are called a superclass (or a parent class) and a subclass (or a child class), respectively. Apart from inheriting behaviors from a parent class, a class deﬁnes its own functionalities to make itself more speciﬁc. This inheritance is the very main concept for any OOP including OTcl. A.2.1 Class and Inheritance In OTcl, a class can be declared using the following syntax: Class [-superclass ] If the optional argument in the square bracket is present, OTcl will recognize class as a child class of class . Alternatively, if the option is absent, class can be also declared as a child class of class by executing superclass Note that, class inherits the functionalities (including procedures and variables) of class . In OTcl, the top-level class is class Object, which provides basic procedures and variables, from which every user-deﬁned class inherits. Example A.11. Consider a general network node. When equipped with mobility, this node becomes a mobile node. Declaration of a class Node and its child class Mobile is shown below. This declaration allows class Mobile to inherit capabilities of class Node (e.g., receiving packets) and to include more capabilities (e.g., moving) to itself.

A.2 Objected Oriented Tcl (OTcl) Programming

401

1 Class Node 2 Class Mobile -superclass Node A.2.2 Class Member Procedures and Variables A class can be associated with procedures and variables. In OTcl, a procedure and a variable associated with a class are referred to as instance procedure (i.e., instproc) and an instance variable (i.e., instvar), respectively. Instance Procedures The syntax which declaver an instproc is shown below: instproc [{args}] { } where instproc “instproc” is deﬁned in the top-level class Object. Here, the name of the instproc is . The detail (i.e., ) of the instproc is embraced within curly braces. The input arguments of the instproc are given in . Each input argument is separated by a white space. OTcl supports assignment of each input argument with a default value. That is, the input argument will be assigned with the default value if the value is not given at the invocation. Denote an input argument and its default value by and , respectively. The argument declaration is as follows: { def} For example, let an instproc has two input arguments: and . The ﬁrst input argument is not given a default value. The default value for the second input argument is given by . To declare this instproc, we replace [args] above with “ {}”. Once declared, an instproc is usually invoke through an object (whose class is ) using the following syntax. [{args}] Instance Variables Unlike instprocs, instvars are not declared with the class name. Instead, they can be declared anywhere in the ﬁle. The syntax for the declaration is as follows: $self instvar [ ...]

402

A Programming Essentials

where instproc “instvar” and an instvar “self” (which represents the object itself) are deﬁned in the top-level class Object. More than one instvar can be declared within an OTcl statement. Syntactically, we simply put the names of all instvars (each seperated by a white space) after “$self instvar”. After the declaration, an instvar can be manipulated by using a command set with the following syntax set [] set [] When presented, the input argument will be stored in the instvar associated with the object or the class . In absence of the argument the above statements return the value stored in the associated instvar . Example A.12. Based on Example A.11, the followings deﬁne a packet reception instproc for class Node and a moving instproc for class Mobile. 3 4 5 6 7 Node instproc recv {pkt} { $self instvar state set state 1 # $self process-pkt $pkt }

8 Mobile instproc move {x y} { 9 $self instvar location 10 set location[0] $x 11 set location[1] $y 12 } Upon receiving a packet pkt, a Node sets its state to be active (i.e., 1), and invokes instproc process-pkt to process the packet pkt. As a derived class of class Node, class Mobile inherits this instproc. It also deﬁnes an instproc move to move to a new coordinate (x,y). This instproc simply sets the new coordiate to be as speciﬁed in the input argument (Lines 10–11). A.2.3 Object Construction and the Constructor An object can be created (i.e., instantiated) from a declared class by using the following syntax: In the object construction process, instprocs alloc and init of class Object is invoked to initialize the object. Instproc alloc allocates memory space to stored the initiated object. Usually, referred as a constructor, instproc init deﬁnes necessary object initialization. This instproc is usually overridden by the derived classes.

A.2 Objected Oriented Tcl (OTcl) Programming

403

Example A.13. The constructors of classes Node and Mobile in Example A.11 are deﬁned below. 13 Node instproc init {} { 14 $self instvar state 15 set state 0 16 } 17 Mobile instproc init {} { 18 $self next 19 $self instvar location 20 set location[0] 0 21 set location[1] 0 22 } At the constuction, class Node sets its variable state to 0 (i.e., inactive). Class Mobile ﬁrst invokes the constructor of class Node in Line 18 (see the details of function next in Section A.2.4). Then, Lines 20–21 set the location of the mobile node to be (0,0). A.2.4 Related Instprocs Class Object also deﬁnes the following instprocs. Instproc next Invoked from within an instproc, next searches up the hierarchy (in parent classes) for an instproc with the same name, and invokes the instproc belonging to the closest parent class.

Instproc info This instproc returns related information based on the input argument. It can be invoked using one of the two following ways: info info The upper and lower invocations return the information about the object and the class, respectively. The choice of the input argument for these two invocations are shown in Tables A.2 and A.3, respectively. Example A.14. Include the following code to the above deﬁnition of classes Node and Mobile, and save the code in a ﬁle “node.tcl”.

404

A Programming Essentials Table A.2. Options of the info instproc for objects.

Options class procs commands

Functions

Returns the class of the object. Return the list of all local methods. Return the list of both Tcl and C local methods deﬁned on the object. vars Return the list of instance variables deﬁned on the object. args Return the list of arguments of the instproc deﬁned on the object. body Returns the body of the instproc deﬁned on the object. default ... Returns 1 if the default value of the argument of the instproc is , and returns 0 otherwise.

Table A.3. Options of the info instproc for classes. Options superclass subclass heritage instances instprocs instcommands instargs instbody instdefault ... Functions Return the superclass of the current class. Return the list of all subclasses down the heirachy. Return the inheritance precedence list. Return the list of instances of the class. Return the list of instprocs deﬁned on the class. Return the list of instprocs and OTcl commands deﬁned on the class. Return the list of arguments of the instproc deﬁned on the class. Return the body of the instproc deﬁned on the class. Return 1 if the default value of the argument of the instproc is , and return 0 otherwise.

23 Node n 24 puts "The instance of class Node is [Node info instances]" 25 puts "The class of n is [n info class]" By executing the ﬁle “node.tcl”, the following result should appear on the screen. >>ns node.tcl n The instance of class Node is n The class of n is Node Exercise A.15. Write OTcl codes which make use of the above options for instproc info in Tables A.2–A.3.

A.3 AWK Programming

405

A.3 AWK Programming
AWK is a general-purpose programming language designed for processing of text ﬁles [29]. AWK refers to each line in a ﬁle as a record. Each record consists of ﬁelds, each of which is separated by one or more spaces or tabs. Generally, AWK reads data from a ﬁle consisting of ﬁelds of records, processes those ﬁelds with certain arithmetic or string operations, and outputs the results to a ﬁle as a formatted report. To process an input ﬁle, AWK follows an instruction speciﬁed in an AWK script. An AWK script can be speciﬁed at the command prompt or in a ﬁle. While the strength of the former is the simplicity (in invocation), that of the latter is the functionality. In the latter, the programming functionalities such as variables, loop, and conditions can be included into an AWK script to perform desired actions. In what follows we give a brief introduction to the AWK language. The details of AWK programming can be found in [30]. A.3.1 Program Invocation AWK can be invoked from a command prompt in two ways based on the following syntax: >>awk [ -F ] {} [ ] [ ] >>awk [ -F ] { -f } [ ] [ ] where {} and [] contain mandatory and optional arguments, respectively. The bracket contains a variable which should be replaced with actual values at the invocation. These variables include ch pgm pgm_file vars data_file Field separator An AWK script A ﬁle containing an AWK script (i.e., an AWK ﬁle) Variables used in an AWK ﬁle An input text ﬁle

By default, AWK separates records by using a white space (i.e., one or more spaces or tabs). However, if the option “-F is present, AWK will use as a ﬁeld separator.1 The upper invocation takes an AWK script as an input argument, while the lower one takes an AWK ﬁle as an input argument. In both cases, variables and input text ﬁle can be optionally provided. If an input text ﬁle is not provided, AWK will wait for input argument from the standard input (e.g., keyboard) line by line. Example A.16. Deﬁnes an input text ﬁle “infile.txt” in the following. We shall use this input ﬁle for most of the examples in this section.
1

For example, awk -F: uses a colon “:” as a ﬁeld separator.

406

A Programming Essentials

#infile.txt Rcv 0.162 FromNode EnQ 0.164 FromNode DeQ 0.164 FromNode Rcv 0.170 FromNode EnQ 0.170 FromNode DeQ 0.170 FromNode Rcv 0.171 FromNode EnQ 0.172 FromNode DeQ 0.172 FromNode Rcv 0.178 FromNode EnQ 0.178 FromNode DeQ 0.178 FromNode

2 1 1 1 2 2 2 1 1 1 2 2

ToNode ToNode ToNode ToNode ToNode ToNode ToNode ToNode ToNode ToNode ToNode ToNode

3 2 2 2 3 3 3 2 2 2 3 3

cbr cbr cbr cbr cbr cbr cbr cbr cbr cbr cbr cbr

PktSize= PktSize= PktSize= PktSize= PktSize= PktSize= PktSize= PktSize= PktSize= PktSize= PktSize= PktSize=

500 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000 1000

UID= UID= UID= UID= UID= UID= UID= UID= UID= UID= UID= UID=

3 8 8 7 7 7 4 9 9 8 8 8

Note that in AWK, “#” marks the beginning of a comment line. At the command prompt, we may run an AWK script to show the lines which contains “EnQ” as follows: >>awk /EnQ/ infile.txt EnQ 0.164 FromNode 1 ToNode EnQ 0.170 FromNode 2 ToNode EnQ 0.172 FromNode 1 ToNode EnQ 0.178 FromNode 2 ToNode 2 3 2 3 cbr cbr cbr cbr PktSize= PktSize= PktSize= PktSize= 1000 1000 1000 1000 UID= UID= UID= UID= 8 7 9 8

Here, the is speciﬁed as /EnQ/ and the is speciﬁes as infile.txt. An AWK script /EnQ/ looks for a line which contains a text EnQ and display the line on the screen. A.3.2 An AWK Script An AWK script contains an instruction for what AWK will perform. It asks AWK to look for a pattern in a record, and performs actions on a matched pattern. The syntax of an AWK script is as follows: {} A could be a logical expression or a regular expression.2 An speciﬁes actions for the matched pattern. Each actions in the curly braces is separated by a semi-colon (“;”). As will be discussed later in this section, AWK provides a wide variety of .

2

While a logical expression is usually implemented by an if statement, a regular expression returns true when ﬁnding a matched pattern. The formal deﬁnition of a regular expression can be found in [31].

A.3 AWK Programming

407

A.3.3 AWK Programming Structure The general form of an AWK program is shown below: BEGIN {} {} {} . . . END {} Prior to procession an input text ﬁle, AWK performs speciﬁed in the curly braces located after the reserved word BEGIN. Then, for each record, it performs actions if the records match with the corresponding pattern. After processing the entire ﬁle, it performs speciﬁed in the curly braces located after the reserved word END. A.3.4 Pattern Matching The ﬁrst part of an AWK script is a pattern as speciﬁed in . The pattern can be a logical or a regular expression. If this part evaluates to true, the corresponding action will be taken.

Logical Expressions For a logical expression, the following operators could be necessary: < >= (less than) (less than or equal) (greater than) (greater than or equal) = != || && (equal) (Not Equal) (OR) (AND)

Regular Expressions A regular expression provides a concise and ﬂexible means to represent a text of interest. It is used extensively in programming language such as AWK, Tcl, Perl, etc. Syntactically, a regular expression is enclosed within a pair of forward slashes (“/”, e.g., /EnQ/). It supports much more functionalities in searching for a pattern as shown in Table A.4: Exercise A.17. Write an input string which matches with each of the following regular expressions. The input string should not match with other regular expressions.

408

A Programming Essentials Table A.4. Special characters used in regular expressions.

Character // ^ $ [] [a-z] [A-Z] [0-9] [a-zA-Z0-9] . * .* ? +

Description Contain a regular expression (e.g., /text/) Match the beginning of a record only (e.g., /^text/) Match the end of a record only (e.g., /text$/) Match any character inside (e.g., [text]) Match any lower-case alphabet Match any upper-case alphabet Match any number Match any alphabet or number Match any character (e.g., /tex./) Match zero or more character in front of it (e.g., /tex*/) Match any string of characters Match zero or more regular expression in front of it (e.g., /[a-z]?/) Match one or more regular expression in front of it (e.g., /[a-z]+/)

(i) /^Node/ (ii) /Node$/ (iii) /[Nn]ode/ (iv) /Node./ (v) /Node*/ (vi) /Nod[Ee]?/ (vii) /Nod[Ee]+/ By default, a regular expression is matched against the entire record (i.e., line). To match a certain regular expression againt a given variable var, we use the following syntax: $var ~ // $var !~ // While the upper command searches for a line which matches with , the lower command searches for a line which does not match with . A.3.5 Basic Actions: Operators and Output The key operators in AWK are shown below. + * / (addition) (subtraction) (multiplication) (division) ++ == = % (increment) (decrement) (assignment) (modulo)

Like in C++, a combination of arithmatic operators and an assignment operator is also possible. For example, “a += b” is equivalent to “a = a+b”. The combined operator in AWK include “+=”, “-=”, “*=”, “/=”, and “%=”. AWK outputs a variable or a string to a screen using either print or printf, whose syntax are as follows:

A.3 AWK Programming

409

print ... printf(,,,...) where , , and so on can be either variables or strings, is the format of the output. Using print, a string needs to be enclosed within a quotation mark (""), while a variable could be indicated as it is. Example A.18. Deﬁne an AWK ﬁle “myscript.awk” as shown below. # myscript.awk BEGIN{} /EnQ/ {var = 10; print "No Quotation: " var;} /DeQ/ {var = 10; print "In Quotation: " "var";} END{} Run this script for the input text ﬁle infile.txt deﬁned in Example A.16. The following result should appear on the screen. >>awk -f myscript.awk infile.txt No Quotation: 10 In Quotation: var No Quotation: 10 In Quotation: var No Quotation: 10 In Quotation: var No Quotation: 10 In Quotation: var The above AWK script prints out two versions of variable var. The upper line prints out the value (i.e., 10) stored in variable var. In the lower line, variable var is enclosed within a quotation mark. Therefore, string var will be printed instead. The command printf provides more printing functionality. It is very similar to function printf in C++. In particular, it speciﬁes the printing format as the ﬁrst input argument. The subsequent arguments simply provide the value for the place-holders in the ﬁrst input argument. The readers are encouraged to ﬁnd the detail of the printing format in any C++ book (e.g., [14]) or in [30]. AWK does not have a direct command for ﬁle printing. Rather, output redirection can be used in conjunction with print and printf. In a Unixlike system (e.g., Linux or Cygwin) a character “>” and “>>” can be used to redirect the output to a ﬁle. The syntax of the output redirection is shown below. print > print >>

410

A Programming Essentials

Note that the command print can be replaced with the command printf. The diﬀerence between the above two lines is that while “>” redirects the output to a new ﬁle, “>>” appends the output to an existing ﬁle. If exists, the upper line will delete and recreate the ﬁle whose name is , while the lower line will append the output to the ﬁle without destroying the existing ﬁle. Exercise A.19. Repeat Example A.18, but print the result in a ﬁle “outfile .txt”. Show the diﬀerence when using “>” and “>>”. A.3.6 Variables As an interpreter, AWK does not need to declare data type for variables. It can simply assign a value to a variable using an assignment operator (“=”). To avoid ambiguity, AWK diﬀerentiates a variable from a string by quotation marks (“""”). For example, var is a variable while "var" is a string (see Example A.18).3 AWK also support arrays. Arrays in AWK can have only one dimension. Identiﬁed by a square bracket ([]), indexes of an array can be both numeric (i.e., a regular array) or string (i.e., an associative array). Example of arrays are node[1], node[2], link["1:2"], etc. Apart from the above user-deﬁned variables, AWK also provides several useful built-in variables as shown in Table A.5.
Table A.5. Built-in variables. Variables $0 $1,$2,... FILENAME FS RS NF NR OFMT OFS ORS Descriptions The current record The 1st, 2nd,... ﬁeld of the record Name of the input text ﬁle (Input) Field separator (a white space by default) (Input) Record separator (a newline by default) Number of ﬁelds in a current record Total number of records Format for numeric output (%6g be default) Output ﬁeld separator (a space by default) Output record separator (a newline by default)

Exercise A.20. Based on the input ﬁle in Example A.16, develop an AWK script to show
3

Unlike Tcl, AWK retrieves the value of a variable without a preﬁx (not like “$” in Tcl).

A.3 AWK Programming

411

(i) Total number of “EnQ” events, (ii) The number of packets that Node 3 receives, and (iii) Total number of bytes that Node 3 receives. A.3.7 Control Structure In common with Tcl, AWK support three major types of control structures: if/else, while, and for (see Section A.1.6). The syntaxes of these control structures are as follows: if() [else ] while() for(;;) Again, when the actions contain more than one statement, these statements must be embraced by a curly braces. AWK also contains four unconditional control commands: break contine next exit Exit the loop Restart the loop Process the next record Exit the program by executing the END operation

B A Review of the Polymorphism Concept in OOP

B.1 Fundamentals of Polymorphism
As one of the main OOP concepts, polymorphism refers to the ability to invoke the same function with diﬀerent implementation under diﬀerent context. This concept should be simple to understand, since it occurs in our daily life.
Receptionist

FriendlyReceptionist

MoodyReceptionist

RudeReceptionist

Fig. B.1. A polymorphism example: Receptionist class hierarchy.

Example B.1. Consider receptionists and how they greet customers. Friendly, moody, and rude receptionists greet customers by saying “Good morning. How can I help you today?”, “What do you want?”, and “What do you want? I’m busy. Come back later!!”, respectively. We design a class hierarchy for receptionists as shown in Fig. B.1. The base class of the hierarchy is class Receptionist. Based on the personality, we derive classes Friendly Receptionist and MoodyReceptionist directly from class Receptionist. Also, we derive another class RudeReceptionist from class Moody Receptionist. The C++ code which represents these four classes is given below: 1 //receptionist.cc #include "iostream.h"

414

B A Review of the Polymorphism Concept in OOP

2 3 4 5

class Receptionist { public: void greet() {cout

Ns2 Introduction to Network Simulator

You May Also Find These Documents Helpful

Network Communications Infrastructure

Network Communications Infrastructure

International Business Negotiation Plan to Commercialize Scooters in Brazil

International Business Negotiation Plan to Commercialize Scooters in Brazil

American Fur Trading Company and John Jacob Astor

American Fur Trading Company and John Jacob Astor

A Case Study: Deals-R-Us Brokers

A Case Study: Deals-R-Us Brokers

Ccna 2 Router and Routing Protocols Study Guide

Ccna 2 Router and Routing Protocols Study Guide

Home Depot Case

Home Depot Case

Scientific Management

Scientific Management

The Effects of the Internet on Channel Strategy

The Effects of the Internet on Channel Strategy

Progress Report of My Educational and Career Goals

Progress Report of My Educational and Career Goals

Epidemiology Pocket Guide

Epidemiology Pocket Guide

Defining Team Roles: the Missing Link in Creating Winning Teams in Corporate Teamwork

Defining Team Roles: the Missing Link in Creating Winning Teams in Corporate Teamwork

Ghgh

Ghgh

Do What You Love

Do What You Love

Gilt Groupes Ceo on Building a Team of a Players

Gilt Groupes Ceo on Building a Team of a Players

Sqoop

Sqoop

Related Topics