Research On The Issue Of Congestion Control In Networks

The swift and violent development of Internet made the traffic increase quickly and all kinds of real time business demonding high network QoS. The rapid increase of traffic makes congestion of Backbone more and more grave. Therefore, congestion control became an important issue in netwok study.For congestion control problem, V.Jacobson proposed the window-based end-to-end congestion control of TCP. But not all users and applications have the window-based end-to-end congestion control with the network scale broadening qulickly and netwok application increasing, especially, the broad application of multimedia business. It is difficulty to keep the high effect and impartial operation of network by using end-to-end congestion control only. The centre node of netwok must participate in congestion control, which is IP layer congestion control. And it has become an important issue of network studing in these years. Random Early Detection algorithm, which is one of IP congestion control algorithms, has been recommened by IETF to using in router broadly.In this thesis, Random Early Detection algorithm study is emphasized on the base of introducing the principles of TCP and IP layer congestion control. For Random Early Detection algorithm, the drop probability linearly increases with average queue length, which makes the drop probability relative high, when the average queue length is close to minimal or maximal limit of queue length, and link utilization degrade. So an improved RED algorithm is proposed which amended the computation of drop probability. Index nonlinear relation is used to decrease drop packet and improve the link utilization in RED algorithm.In addition, the fairness of RED is analyzed. RED algorithm cannot guarantee that each flow has fair bandwidth relatively, when many flows using limited bandwidth. To improve the fairness of RED, an improved algorithm is proposed. The relationship of the arrive speed of each flow and the average arrive speed of router is used to estimate whether this flow is high bandwidth flow or not when the drop packets are more than the limt. For high bandwidth flow, drop some packets randomly accoding to certain drop probability, then put the other packets into the RED output queue. In this way, the bandwidth using by high bandwidth is limited. The validity of algorithm is proved by emulational experiment.