Nonlinear Adaptive Red Algorithm-based Network Congestion Control

With the development of computer network, the scale of the Internet grows explosively. More and more new applications are launched to satisfy the needs of people, which may greatly aggravate the traffic load of the Internet, thus, increase the loss rate of the packets, enlarge the delay of the end-to-end system, worsen the performance of the whole Internet system. Therefore, taking effective measures to control the network congestion will be very meaningful to enhance the performance of the network.In order to get more details about the network congestion control, this thesis firstly took a view to some basic ideas about the congestion control, analyzed the reasons of network congestion, and then introduced the congestion control schemes based on TCP layer and IP layer respectively. As to the TCP-based congestion control strategies, we mainly discussed their 4 basic phases and some classic algorithms. And as to the one based on IP layer, we gave a brief introduction to the class algorithms of queue management, carried an in-depth study to the RED algorithm, and analyzed the performance of RED by NS simulations, and then introduced several related algorithm according to the defects of RED, which is one of the study points of this thesis.On the basis of simulations, we found that the queue fluctuation of the routers has some thing to do with the changing rate of the drop probability, that is, the smoother of the changing rate of the drop probability, the stabler of the queue fluctuation. Based on this point and the ARED's feature to stabilize the queue length in a target range, a new improve approach of ARED, called Nonlinear ARED, is proposed, which is the other study points of this thesis. Nonlinear ARED algorithm takes a non-linear modification to the original drop-probability function adopted in ARED algorithm, which makes sure that the changing rate of the drop probability changes in a smooth manner, thus, leads to a stabler queue. The simulations show that the Nonlinear ARED algorithm can not only cut down the queue fluctuation, but also reduce the drop probability of packets in most scenarios, compared with ARED algorithm.