Random Early Detection Gateways for Congestion Avoidance

Summary

This paper explains how Random Early Detection can be used in gateways for congestion avoidance. RED is based on the idea that gateways are the best places to detect congestion. It has the best view and can detect the difference between propagation delay and persistent queueing delay.

RED uses average queue size to detect congestion just as it begins to appear. The RED gateway would then notify the sender of this congestion by marking the packet (either by dropping the packet or setting a bit in the header). The gateway decides which packet to randomly mark based on a certain probability. With this strategy, RED can ensure that a sender is told to reduce its window size proportional to its share of the bandwidth. Additional features of RED are that it is able to keep the average queue size low, accommodate the occasional burst of traffic and avoid global synchronization.

When evaluating RED gateways, the authors used 7 criteria. These included congestion avoidance, appropriate time scales, no global synchronization, simplicity, fairness and appropriate for a range of environments. They explained how RED met each of those criteria.

Criticism and Questions

The authors mention that fairness is inherently built into the RED protocol. It cannot ensure that all users get the same amount of bandwidth. It is mentioned that it is possible to add in more mechanisms that implement fairness. I think it would be interesting to see the result of adding in those mechanisms. Fairness has been one of the major goals of the network and would help determine if a protocol could be used in practice. Therefore, it would nice to see RED's performance once the fairness mechanisms were added in.

Feedback

I enjoyed reading this paper. It laid out the basics of the RED protocol clearly and had some great simulations. I would say keep this paper in the syllabus as it shows another approach to congestion avoidance.