Improved TCP congestion control algorithm for wireless networks

Abstract

Transmission Control Protocol (TCP) congestion control algorithm deals with reducing the network load during congestion for achieving beller throughput. It rcduces the transmission window size after a rctransmission timeout or reception of consccutivc three duplicate acknowledgements (3-dupacks). It works well for the wired networks where most of the timeouts and 3-dupacks are the result of congestion. But we like to argue that in the wireless networks, where random segment loss due to bit errors is a dominant concern, the arrival of duplicate acknowledgements or even the retransmission timeouts do not necessarily denote congestion. In those cases, throttling transmission rate is not necessary. As the basic TCP congestion control algorithm cannot distinguish between congestion event and bit crror event, it fails to perform well in wireless networks. In this thesis, we propose some modifications to the basic TCP congestion control algorithm so that its performance is enhanced in wireless networks. In particular, our algorithm refines the multiplicative decrease algorithm of TCP NewReno. We are using some statistical counters to track the frequencies of the occurrences of timeouts and 3-dupacks. Different ratios of these counter values are then used to differentiate a congestion event from a non-congestion event. We 'are also tracking the time difference between two consecutive timeouts to figure out whether timeouts are caused by network congestion or random bit errors. We tested our proposed algorithm using the Network Simulator version 2 (ns-2) and found that it shows better performance than any other TCP variants in wired-cum-wireless networks. Moreover, our algorithm is end-to-end in nature and modifies only TCP sender's algorithm.