FAST TCP Congestion Control Reasearsh In High Speed Network

In recent years, high-speed network whose bandwidth is more than 1Gbps has been put into practical use from the laboratory. In the data grid network and storage network, hosts usually are directly connected to the high-speed network through Gigabit network interface. In order to move the data, or backup the database, amount of transmission of data is caculated by the GB and even TB .In this case, if used to transfer a lot of data, the existing TCP implementation can not get enough throughputs to meet the demand of QoS. How to improve the throughout of high-bandwidth networks, fairness and bandwidth utilization is what we must consider. FAST TCP protocol has been put forward to solve the problems of low throughout of the traditional TCP. Although it is a high-speed network based on delay feedback protocol, it affects the stability of the network and fairness due to its own inaccurate estimation of RTT. So it is necessary to improve the performance of FAST TCP.In this paper, we do research on the FAST TCP congestion through both stability and fairness. In stability, we obtain the stability boundary conditions through the range of protocol parameterγand also improve the accurance of scope ofγ; As to fairness, we compare FAST TCP with TCP Reno using the new comparison method based on the management of queue; Finally, we optimize the estimation model of FAST TCP for the improvement. The main task is as follow:1. For the stability of FAST TCP congestion control algorithm, we construct a new link model for the comprehensive analysis of stable condition of FAST TCP data streams which share the link in a bottleneck .We also improve the congestion control scope of parameterγof FAST TCP.2. To get the fairness comparison between FAST TCP and TCP Reno, we put forward a new comparison method of two congestion protocols. The comparison result of these two protocols can give the guide to deploy the FAST TCP which will share one link with TCP Reno.3. In connection with inherent problem of FAST TCP: FAST TCP shows serious volatility of transmission if the estimatation of RTT is not accurate when the sharing flows are more than one data source at the same time. And this RTT fluctuations influence the queuing delay and the behavior of the congestion window fluctuations. A new algorithm of RTT calculation is designed to smooth the estimation of RTT based on combination of exponential weighting and ARMA model. Then we extend the improved protocol into the NS2 simulator to achieve the analysis of the new protocol.Two network scenarios are also designed to analyze the optimization result between before and after the optimization of FAST TCP according to rate, RTT, congestion, queue size and so on.