Experimental Evaluation Of TCP Protocols Over High Speed Long Delay Networks

The regular TCP congestion control algorithm has proved remarkably successful. However, in recent years, it is recognized its performance is unsatisfactory in networks with high bandwidth-delay product (BDP) paths. The main reason for this is the too conservative behavior of its congestion window adjusting. A number of high-speed TCP variants have been proposed to remedy the aforementioned problem:Highspeed, Scalable TCP, BIC, CUBIC, FAST, Hamilton TCP, Africa, Compound, Illinois, YeAH and so on. While the design of high-speed TCPs has made great progress, far less attention has been paid to the performance evaluation of these protocols, especially based on an experimental test bed. Since deployment of high-speed TCPs has become a reality, comprehensive and complex evaluation of these protocols is very meaningful.The current experimental evaluation of high-speed TCPs based on a test bed is very insufficient, either too little protocols included or too little metrics evaluated. This dissertation is intended to make a comprehensive performance evaluation of high-speed TCPs on a test bed, including relatively complete protocols and metrics.This dissertation setup a network test bed based on Dummynet and designed various experiments to evaluate the performance of Highspeed, CUBIC, Compound, Hamilton TCP, Illinois, YeAH and Reno. A lot of metrics were used:throughput, fairness, friendliness, convergence and response. By changing the key parameters of the network, such as bottleneck bandwidth, prorogation delay and buffer size, to analyze the impact on these metrics. Based on the interpretation of experimental results, the performance and effects of these high-speed TCPs are well understood and it brings real meaning for protocols selection and further design.