Modeling TCP loss recovery latency with the number of retransmissions

Abstract

So far, several analytic models have been proposed to predict the performance of transmission control protocol (TCP) such as steadystate throughput. However, for more detailed performance analysis of various TCP implementations, the fast recovery latency during which packet losses are retransmitted should be considered based on the relevant strategy. In this paper, we derive the loss recovery latency of three TCP implementations including TCP Reno, NewReno, and when selective acknowledgement (SACK) option is used. Specifically, the number of lost packets each TCP sender detects and retransmits during fast recovery is considered. Thereby, the proposed model can differentiate the loss recovery latency of TCP using SACK option from TCP NewReno. By numerical results verified by simulations, we evaluate that the proposed model can capture the precise latency of TCP loss recovery period. © Springer-Verlag 2004.