Comparison of TCP implementations in simulated environment

Abstract

Behaviors of various TCPs have been analyzed over wired and wireless links, how they initialize their congestion window during slow start and congestion avoidance phase and what is the effect in the performance in wired and wireless networks. Here the study of TCP version called TCP Westwood which adjusts the congestion window is included. It controls the congestion window by using end-to-end bandwidth estimation. TCP Westwood continuously estimates the end to end available bandwidth on the basis of ACK arrival rate that is used to reset the window to match the availability of bandwidth. This makes TCP Westwood highly robust to the sporadic losses because of wireless channel problems. In this research paper performance of TCP Westwood has been evaluated in the presence of older versions of TCP over wired and wireless networks using a network simulator tool, NS-2.35. Some of the issues with TCP Westwood are addressed in this paper they are, firstly TCP Westwood is unable to distinguish the losses when timeout occurs in wireless links. It is unable to distinguish them due to congestion or link failure. Secondly in slow start phase, TCP Westwood increases congestion window exponentially according to the predefine value of ssthresh. Then switches to the congestion avoidance phase when cwnd value becomes greater than or equal to slow start threshold. Due to this Westwood cannot increase cwnd efficiently during the availability of suddenly increased bandwidth because of network conditions. More elaborately, though available bandwidth is more utilized in slow start phase it cannot enhance the slow start phase and as a result it cannot efficiently access the available bandwidth.