Research On The Congestion Control In Computer Networks Based On Active Queue Management

Internet has been developed into the basic foundation of information society. As the rapid development of Internet, network congestion has become an important issue. Internet primarily depends on TCP end-to-end congestion control to avoid network congestion. But TCP could not meet the various demands of every application on the complex network. So the network itself should take part in the congestion avoidance and control. Recent research has indicated that it is more effective for detecting and preventing congestion if the routers perform queue management schemes. The congestion control strategies in the routers- Active Queue Management has been the hotspot direction. By evaluating the state of network and foretelling the congestion, AQM can drop the packet purposefully, so that the sender side can be informed of the state of network and then adjust its sending rate. But these previously proposed algorithms are heuristic and lack of systematic theory and method to analyze and design, and they are not perfect in terms of response time, stability and robustness. Control theory is a quite mature system theory, and some of its methods may profit the congestion control, and enhance its performance.For the purpose of improving the performance of congestion control algorithm, the main congestion control mechanisms are studied in this paper. After studing the AQM algorithms that proposed previously, and combined with the control theory, the new AQM algorithms are designed. The main research points in this dissertation are as follows:1. The fundamental reasons of congestion and congestion control mechanism that used more recently are analyzed. The four phases of source end algorithm and some typical algorithms of the link algorithm are studied, and also the performances of every algorithm are analyzed. 2. On the deeply study of TCP/AQM flow model, the achievement processes of RED, P, PI, PID, etc. are analyzed from the angle of control theory, and the performances of every algorithm are analyzed.3. For the parameter of RED is difficult to be regulated, a new incompletion differential PID-RED has been proposed to solve the problem, and its performance of robustness has been verified to be better than RED and PI algorithms.4. Though the large delay has impact the performance of the network, almost all the existed algorithms neglect the large delay, for example RED,PI algorithms all neglect this impact. In this study, a new AQM algorithm-SNGM-PID is proposed which is based on the combination of neural network and gray prediction. In this new algorithm, the self-study and self-adaptive characteristic of neural network is used to tune the parameter of PID controller, and the gray predictor is used to compensate the impact produced by the delay. Finally the performances in queue length stability and robustness were simulated by the simulator, and the results have been studied and analyzed, and the new algorithm is verified to be better than the RED and PI algorithms.