Analysis of the Increase and Decrease Algorithms for Congestion Avoidance in Computer Networks
Reviewer: Robert Dugas
This paper addresses the problem of finding an optimal congestion avoidance algorithm in
terms of fairness and efficiency.
The primary contribution is the mathematical insight and exploration of congestion avoidance
schemes and their predicted behaviors.
We want to stay as close to the "knee" load/throughput level as possible
Binary feedback provides end-to-end nodes with adequate control feedback
Additive increase, multiplicative decrease meets performance criteria
It is not clear to me what the degree of novelty of this paper is due to my
lack of background knowledge of the times. However, it seems that the mathematical
analysis of the available congestion control mechanisms proved exceedingly useful
This paper was largely an abstract, theoretical survey and thus the methodology did not
include simulations or tests. Instead, the authors provided mathematical arguments supplemented
by graphs and explanations.
One obvious limitation of the paper is the lack of real-world data. In addition, the
authors suggest that future work in the realm of feedback timing and structure might
improve performace. Finally, not a lot of attention is focused on reducing oscilation on the
way to the fair and efficient optimum.
The basica message of this paper seems to be that additive increase and multiplicative
decrease is the optimal control mechanism.