Monday, September 8, 2008

Analysis of the Increase and Decrease Algorithms

This paper models congestion avoidance and control using control theory in which the system senses its state and feeds this back to the users who adjust their controls. This paper evaluates the various congestion avoidance schemes presented in prior work. Users receive feedback from the network using the “congestion bit” and accordingly adjust their loads. The paper does a very good job of defining the problem and mechanisms in easy-to-understand equations. The paper nicely shows that AIMD is the best strategy to use using allocations for two users. It demonstrates the oscillation of the AIMD strategy and how it converges to the optimal point between fairness and efficiency. The authors also propose nonlinear controls but are not too enthusiastic about it (rightly so), as it might be overly sensitive to the parameters. While this paper doesn’t include any measurement studies, I am convinced this should stay in the syllabus as it provides a solid grounded understanding of the topic using clear mathematical concepts.

Comments:

1. The notion of fairness is a little odd in this paper. It concentrates only on allocation but not on what are the relative requirements of the users!

2. I really want to read a paper that describes how transport works when nodes aren’t honest and are assumed to cheat. Any suggestions?

No comments: