A Duality Model of TCP Congestion Controls

In this paper we propose a uniied optimization model of TCP ow control based on duality theory. We show that TCP Reno and TCP Vegas, with or without random marking, can be regarded as carrying out a dual method to maximize aggregate source utility. Diierent schemes correspond to diierent choices of source utility functions, and we derive these utility functions. This implies in particular that TCP Reno, with or without random marking, attempts to equalize window, and discriminates against sources with large propagation delays as observed in many previous studies. This feature is used to validate our model with preliminary simulation results.