From: T Scott Saponas (ssaponas@cs.washington.edu)
Date: Mon Oct 18 2004 - 07:20:33 PDT
Review by T. Scott Saponas
"Congestion Avoidance and Control" describes a scheme to avoid and recover from congestion in TCP. The author suggests this can be accomplished with a combination of Slow-start, good RTT estimation, multiplicative increase, and additive increase. It also suggests a threshold value be used for how long to use slow-start before switching to additive increase (this is to avoid wild oscillation). The main contribution of this paper is the presentation of these algorithms. The presentation of these algorithms is well backed up by convincing arguments plus experimental graphs that these methods will actually prevent and recover from congestion well.
One downside of this method of congestion control is its dependence on their observation that in most networks a packet loss is much more likely to be due to congestion than to errors introduced by the connection medium. This suggests such an implementation would not work well on a lossy high bandwidth connection because TCP would then be trying to backoff from some phantom congestion.
This paper is quite relevant to the Internet today. Their methods for congestion control are essentially what we use now and they do a good job of avoiding massive network collapses, unlike previous methods. Besides it limited applicability to lossy high bandwidth lines, this paper's contribution is relevant and useful.
This archive was generated by hypermail 2.1.6 : Mon Oct 18 2004 - 07:20:34 PDT