A Game Theory Perspective on TCP Congestion Control Evaluation

Abstract

A dynamic game model with incomplete information is proposed in this paper for the analysis of TCP congestion control. Based on this model, the essence of non-equilibrium in the TCP congestion control game is deduced by probabilistic analysis, that is, the influences of players unilaterally changing congestion control mechanisms on their transmission rates are greater than those on the network congestion probability. Experiment results show that the TCP protocol cannot guarantee the congestion control game reaching equilibrium in a certain period, and players can increase their revenues by 43%, but reduce the global revenue by 6% through unilaterally changing their congestion control algorithms. With the extension of transmission time, the impact of players unilaterally changing strategies on themselves gradually decreased, and the global revenue rate fluctuated around 0.