Duality-based Tcp Congestion Control with Error Analysis

We review optimization models for congestion control, focusing on the dual of a widely studied utility maximization problem. We show that many congestion control algorithms can be modeled within this framework, and conversely, standard techniques for iterative optimization can inspire practical congestion control algorithms. This previous work all assumes precise feedback of congestion information from network to traffic sources. We study the effect of error in feedback information within this duality model. Using standard inexact optimization techniques, we show that, provided relative error is bounded, the network behavior is attracted to a region that contains the optimal point if exact information were available.