Congest Control Research Of High Speed Network

With the rapid development of Internet-based long distance communication, the mode of human information intercourse has been changed greatly. The influence of computer network have infiltrated into economy, politics and culture. When more and more modern communication technologies such as DWDM and ASIC have been used in the bottom of networks, the computer networks continuous develop to high speed. At the same time, Moore's Law, Metcalfe Law and Matthews Effect are keeping on impacting the information era by their magical function, so the throughputs of networks have been grow awfully. But current TCP/IP networks only provide best effort services and their resources allocations are mainly done end hosts, and all the packets are treated similarly without effective management. So it is not easy to solve that congestion often results in the decline of Quality of service (QoS).Congestion control is the basic part of QoS, and the precondition of basic QoS is the use of congestion control. Therefore, it is very important to solve the problem of congestion of networks. Combining the projects supported by National Natural Science Foundation of China (60074026) and Guangdong Province Science Foundation (000409), this dissertation employs control theory to study the mechanism of TCP/IP congestion control. Our main contributions in link arithmetic and source arithmetic include:Firstly, we propose an integral-separated PI algorithm that is developed from general PI algorithm. Because the loads of network often fluctuate, general PI algorithm need more time to converge. So we propose integral-separated PI algorithm. When bias between the control object and the goal is large, disable the action of integral. When the control object is close the goal, integral start to action. Compared with general PI, Simulation results show that integral-separated PI can not only eliminate the queue static error and keep the queue stabilized, but also it is able to speed up the queue convergence.Secondly, we propose a differential forward PI control algorithm to solve the problem of network delay. When the delay of connection is large, the static quality and the dynamic quality of general PI algorithm become worse. Our researches prove that differential forward PI control has better performance than general PI algorithm.Thirdly, we propose a fuzzy adaptive PI link algorithm. The complexity of...