Reviewer: Manfred Georg
Date: 10-13-2005
How would you rate this paper, relative to others we have read? top 25%, but not top 10%
How would you rate your knowledge of the topic of this paper? expert
What problem or issue does the paper address? Why is it important?
Detecting receiver misbehavior in Congestion Control algorithms. This is important since the Internet is no longer a trusted environment and receivers of network traffic have both the motive and opportunity to misbehave to increase their rate unfairly.
What are the main contributions of the paper and why are they important?
This paper presents an analysis of several different approaches for detecting receiver misbehavior. The authors focus on the receiver centric protocol, RCP and analyze different ways of detecting receiver misbehavior. The approaches used include techniques in the network core, and in the network endpoints. They further analyze detection when dealing with optimized protocol stacks and with short term misbehavior.
How significant are these contributions relative to previous work?
Although this work builds on previous work on misbehavior, there has not been a strong work on different generic methods for detecting such misbehavior. The techniques outlined in this paper can be easily applied to any network protocol. This provides a significant contribution to the subject.
Give detailed comments justifying your view of the paper.
Most of the paper hinges on the definition of fair. The question should be raised, is it fair to behave like TCP? Is it fair to have all flows through a bottleneck link send at the same rate? The authors seem to take it as given that fairness means behaving like TCP. Even if this point is conceded, problems later arise when dealing with optimized protocol stacks. Which TCP are we taking as our golden standard? What is fairness?
In general, the paper is well written and raises interesting points. However, I fear that the general assumptions are both understated, and debatable. The problems with regard to optimized protocol stacks spring directly from their lack of a good definition of fairness. This causes some cleverness, in the form of approved optimizations, to be deemed allowable, while punishing other cleverness, in the form of parameter tweaking. We must thank the authors more for bringing up topics that are of importance, than for actually solving anything.