Research On A New TCP Congestion Control Algorithm

With computer networks being widely used in many fields, resources of the network are increasingly demanded by users that Internet traffic is constantly increasing at a high rate. The network systems often result in the status of congestion. For the IP networks have the characteristics, such as connectionless and packet switch, they can not avoid the congestion indeed. so the scheme of congestion control has also been the key factor of assuring the Internet's stability and robustness.Congestion in the network stems from the disequilibrium about the source and the flux in network, it wouldn't be eliminated because the improvement of the network's process ability this day, the congestion hasn't had a perfect solution scheme all the while. TCP congestion control is the primary congestion control mechanisms used in nowadays Internet. We analyzed the basic principle of the current TCP congestion control mechanism. Including slowstart,congestion avoidance,fast recovery and fast retransmit. TCP protocol has come through several improved versions. Every step makes possible higher bandwidth utilization rate. From RFC1122 to TCP Tahoe and to the most widely used TCP Reno, each modification has made an great enhancement in its performance.This paper investigates the current standard slow-start and congestion avoidance algorithm of TCP congestion control mechanism and its actual problem. A new variant congestion control mechanism which is based on the cosine function, and called"Pow-Reno"is presented. Finally, NS-2 simulation results demonstrate that it can significantly reduce both packet losses and traffic burstiness, and increase the bandwidth utilization ratio. It also provides a better granularity control over the congestion window, and still contributes to improving a system throughput.