Random Early Detection Gateways For Congestion Avoidance

Random Early Detection Gateways for Congestion Avoidance
Sally Floyd and Van Jacobson

Lawrence Berkeley Laboratory
University of California [e-mail] [e-mail]
To appear in the August 1993 IEEE/ACM Transactions on Networking

Abstract

of a delay-bandwidth product) that were full much of the time; this would significantly increase the average delay
This paper presents Random Early Detection (RED) gate- in the network. Therefore, with increasingly high-speed ways for congestion avoidance in packet-switched net- networks, it is increasingly important to have mechanisms works. The gateway detects incipient congestion by com- that keep throughput high but average queue sizes low. puting the average queue size. The gateway could notify
In the absence of explicit feedback from the gateway, connections of congestion either by dropping packets ar- there are a number of mechanisms that have been proriving at the gateway or by setting a bit in packet headers. posed for transport-layer protocols to maintain high throughWhen the average queue size exceeds a preset threshold, put and low delay in the network. Some of these proposed the gateway drops or marks each arriving packet with a mechanisms are designed to work with current gateways certain probability, where the exact probability is a func- [15, 23, 31, 33, 34], while other mechanisms are coution of the average queue size. pled with gateway scheduling algorithms that require perRED gateways keep the average queue size low while connection state in the gateway [20, 22]. In the absence of allowing occasional bursts of packets in the queue. During explicit feedback from the gateway, transport-layer protocongestion, the probability that the gateway notifies a par- cols could infer congestion from the estimated bottleneck ticular connection to reduce its window is roughly propor- service time, from changes in throughput, from changes tional to that connection’s share of the bandwidth through in end-to-end

References: Tool”, Proceedings of Winter 1988 USENIX Conference, 1988, pp. 17-78. [2] Bala, K., Cidon, I., and Sohraby, K., “Congestion Control for High Speed Packet Switched Networks”, INFOCOM ’90, pp. 520-526, 1990. March, 1992. [9] Hansen, A Table of Series and Products, Prentice Hall, Englewood Cliffs, NJ, 1975. [10] Harvey, A., Forecasting, structural time series models and the Kalman filter, Cambridge University Press, 1989. Association Journal, Vol. 58, March 1963, p. 1330. [13] Hofri, M., Probabilistic Analysis of Algorithms, Springer-Verlag, 1987. [19] Keshav, S., “REAL: a Network Simulator”, Report 88/472, Computer Science Department, University of California at Berkeley, Berkeley, California, 1988. [28] Ramakrishnan, K.K., Chiu, D., and Jain, R., “Congestion Avoidance in Computer Networks with a Connectionless Network Layer, Part IV: A Selective Binary Feedback Scheme for General Topologies”, DEC-TR-510, November, 1987. Computer Networks”, ACM Transactions on Computer Systems, V. 8, N. 2, pp. 158-181, 1990. 1-3, 1993. gateway congestion control policies”, unpublished, 1989. [35] Young, P., Recursive Estimation and Time-Series Analysis, Springer-Verlag, 1984, pp