A Novel Approach to Distributed Network Utility Maximization for a Low-Loss Low-Delay Rate Control

This paper proposes an optimization approach to rate control using a novel estimation framework where the objective is to maximize the aggregate utility of sources over their transmission rates. The link and sources are viewed as processors of distributed computation and the control mechanism is derived as Sequential Quadratic Programming (SQP) algorithm to solve the dual problem. The uniqueness of our approach is that it allows sources to estimate link bandwidth prices and thereby maximize their own benefits rather than depending on the price feedback from network links. This technique has the advantage over exiting approaches in that it allows reduced computational complexity at routers. Moreover, the fast convergence of our algorithm in turn improves on stability and responsiveness of rate control and enables reduced buffer occupancy over congested links, hence allowing low packet loss and low delays. We present simulation results in comparison with another dual algorithm.