A Model for Steady State Throughput of TCP CUBIC

Abstract

For transmission control protocol (TCP), CUBIC is a TCP-friendly high-speed variant, in which the window size is a cubic function of time since the last loss event. TCP CUBIC is implemented in Linux operating systems and performs well in wired networks with large bandwidth-delay product. Most of the evaluations of TCP CUBIC are conducted via simulations or experiments. Analytical models for TCP CUBIC are few. In this paper, we propose a Markovian model to determine the steady state throughput of TCP CUBIC in wireless environment. The proposed model considers both congestion loss and random packet loss due to fading. We derive the stationary distribution of the Markov chain and obtain the average throughput based on the stationary distribution. Simulations are carried out to validate the analytical model. Results show that the simulated stationary distribution and the average throughput are both very close to our analytical results. Furthermore, we analyze the throughput performance of TCP CUBIC. Results show that random packet loss reduces the normalized average throughput more for end-to-end flow with large bandwidth-delay product. We propose an improvement to increase the throughput performance of TCP CUBIC by moderately increasing the window growth factor and the multiplicative decrease factor.