Reviewer: Oleg Elkhunovich
The paper discusses a problem of Congestion Avoidance. The key metrics are considered in analysis: efficiency, fairness, convergence time, and size of oscillations.
The concentration of the paper is on analysis of increase/decrase algorithm that is in heart of congestion avoidence scheme. It derives the optimal amounts by for increase/decrease that are used in actual implementation.
Heterogeneity of networks has resulted in mismatch of arrival and service rates in the intermediate nodes in the network causing increased queuing and congestion
Congestion avoidance scheme aims to keep the network operating at the knee. Users are encouraged to increase their traffic load as long as this does not significantly affect the response time, and are required to decrease it when it happens.
Authors derive that decrease should be muliplicative and increase should be additive for the optimal policy.
This paper is significant because it thouroughly analyzes the problem of the increase/decrease algorithm and gives the optimal way to design that algorithm, which is at the heart of network avoidence scheme.
The paper is very well organized. Authors present the problem and analyze each of the four possible solutions based on four metrics. They use linear models to derive an optimal solution.
A number of practical considerations is left open. Impact of asynchronous operations is brought up but not discussed.
Setting up the problem and than methologicly evaluating the possible solutions can lead to an elegant analysis.