Primal Congestion Control Algorithms for Multimedia Streams

We investigate in this paper congestion control algorithms that are TCP-friendly and media-friendly at the same time, so they are more suitable for multimedia streaming than TCP's AIMD. We use in this investigation the optimization framework developed by Kelly et al. in. Kelly et al. derived two gradient-like algorithms, a primal algorithm and a dual one, to control the congestion in a computer network in an end-to-end way and to obtain optimal bandwidth allocation among competing sources. Starting with Kelly et. al's paper, several other papers proposed different variants of the original primal and dual algorithms, but none of these considers the specific characteristics of multimedia streams. The aim of this paper is to build window-based, primal congestion control algorithms for optimizing the perceived quality of multimedia streams in best-effort networks. We build these media-friendly primal congestion control algorithms using the following recipe: 1) first we construct a window-based congestion control algorithm that is TCP-friendly, but has a much more stable throughput and then 2) we shape this stable throughput according to media characteristics (e.g. bitrate). We present two such primal congestion control algorithms and prove their stability around the equilibrium.