Comparative performance of TCP variants on self-similar traffic

Network traffic of Internet will continue to expand in terms of both volume and users, and transmission control protocol (TCP) is accounting for more than 90% the Internet traffic. In addition to this, it is well known that Internet traffic exhibits self-similarity, which cannot be described by traditional Markovian models such as the Poisson process. In this paper we focus on experimental quantitative comparisons of four TCP variants namely Tahoe, NewReno, Vegas, and SACK running over self-similar traffic. With the aid of network simulator, NS2, aggregated network traffic of self-similarity behavior is simulated and will act as the background traffic. The impact of one TCP microflow will be observed, and data analysis such packets dropped, packets retransmitted, throughput, and the behavior of the data received over time is analyzed. In measuring the degree of self-similarity we employed optimization method to determine Hurst parameter, H. From our analysis, Vegas out performed other TCP variants at high traffic load and all TCP variants tend to demonstrate self-similar traffic flow even though initially the departure rate of TCP packets is constant.