Performance Analysis of Wireless TCP

Abstract

TCP performs reasonably well over Internet where packet losses are mainly due to network congestion. However, TCP originally targeted on the wire networks is inapplicable for the wireless networks where packet losses are mainly caused by link errors. TCP treats a packet loss as an indication of network congestion and misleads the packet loss caused by the link error in the wireless networks. This confusion makes the over-reduction of window size in TCP, resulting in poor performance. Wireless TCP (WTCP) was proposed to promote the performance of TCP in the wireless networks [1]. WTCP splits a TCP connection between a fixed host and a mobile host into two segments with end-to-end semantics. It is installed into the base station to more correctly estimate the retransmission timeout and speed up the packet retransmission in the wireless link. In this paper, we analyzed the WTCP based on the analysis of TCP Reno [2]. WTCP treats a timeout or a duplicate ACK as an indication of packet loss. We first analyzed the throughput under only considering the loss caused by a duplicate ACK, and then expanded to the case with timeouts. The simulation results show that WTCP outperforms TCP in the wireless environment where link error rate is high, while has the similar throughput in the wire environment with low link error rate.