Modeling general AIMD congestion control throughput with slow start

摘要

General Additive Increase and Multiplicative Decrease (General AIMD or GAIMD) congestion control generalizes the standard TCP congestion control mechanism. In this paper, we present GAIMD-SS, an enhanced model for predicting the long-term steady-state mean throughput of the GAIMD congestion control, towards more accurate results than previous GAIMD model. We develop a three-state Markov chain for analyzing the behaviors of the GAIMD, and enhance previous works by taking into account the slow start phase and receiver's maximum window limitation. Our experiment and simulation results show that the GAIMD-SS model predicts the sending rate of GAIMD congestion control more accurately than the previous works in a wider range of packet loss rate.