RTT Estimation And Enhanced TCP Vegas

Internet has been developing sharply since the 1990s, and the problem of Internet congestion becomes more and more severe. As TCP is a main transport protocol in Internet, TCP congestion control algorithm is broadly researched.This thesis summarizes researches of TCP congestion control algorithms and pays attention to improving for Vegas. The contributions of this thesis are as follows:Firstly, TCP congestion control algorithm is introduced. Four-fold mechanisms are included: Slow Start, Congestion Avoidance, Fast Retransmit and Fast Recovery, which are the basis of congestion control algorithm. Reno which is the most widely-used and well-developed algorithm is composed of the mechanisms. Reno which uses packet loss as the measure of congestion has some disadvantages. So Vegas is proposed to avoid packet loss. In contrast to Reno, Vegas uses delay as the measure of congestion and anticipates the onset of congestion by monitoring the difference between the actual throughput and the expected throughput.Vegas can avoid packet loss, and its throughput and stability are better than Reno. Because of its conservative congestion avoidance, its throughput declines sharply coexiting with Reno. There are some problems in Vegas too, which arise from RTT estimation precision. Because RTT is important for Vegas, two-order adaptive Kalman filter is used to estimate RTT. Old RTT is replaced with estimated RTT to improve performance of Vegas. And Slow Start and Congestion Avoidance of Vegas are enhanced: Fast response of the initial stage of Slow Start and deducing packet loss are reached; Adaptive Congestion Avoidance is used to increase aggressiveness of Vegas. And modified Vegas based on RTT estimation is proposed.Finally, the analysis is validated with NS simulation. The emphasis is the study of congestion window, throughput, RTT and packet loss of Reno, Vegas and proposed Vegas in different simulated scenarios. Simulations prove that the proposed algorithm performs well in many aspects. The biggest practical problem of TCP Vegas: fairness is enhanced in a certain extent, which hampers popular application of Vegas in Internet. But there is also something to be improved, for example: finding better RTT estimation method and online tuning for the parameters of filter, etc.