Network Congestion Control Improved Red Algorithm

The self-similar model tends to be more suitable than the traditional Markov model in the research of modern high-speed technology with existing of the quality of self-similarity in today's network business, as the former can present the characteristic of the actual network business more accurately. And the quality of self-similarity will bring some kinds of unexpected influence to the business of the queue system and the performance of the network, whereas the conclusions and network design strategies based on the traditional model can not be equally applicable to the self-similar high-speed network. With the rapid development of network, self-similar network design, traffic control, analysis and management have put forward a more serious challenge. Therefore, self-similar traffic, the congestion control strategy to carry out research according to the development of Internet network with great significance.Floyd proposed the first Random Early Detection(RED) algorithm, it is based on the traditional Poisson model. And its basic idea is that discarding the data group according to a certain probability mechanism during the period of congestion occurs of the early group arrival, in order to avoid congestion. The key point is how to detect congestion, and how to calculate the packet drop probability. But it is based on the traditional Poisson model, so it is not suited for network traffic with self-similarity characteristics.This paper summarizes the research results have been based on the relevant issues and careful analysis of how to model self-similar traffic queue on the router buffer management, improved RED algorithm, to better avoid the congestion occurs and improve network performance, and then presents a self-similar traffic RED Algorithm. The primary studies are list bellow.1. Research the RED algorithm and improved algorithm.2. Research the causes of self-similar characteristics, properties, estimation, modeling, prediction. Using the ON/OFF model based on Pareto distribution under the NS2, generated self-similar traffic. This model can truly reflect the actual network environment, generate a self-similar nature network traffic. 3. Analysis of self-similarity on network performance, combined with AQM performance evaluation criteria, to improve on the classic RED algorithm is proposed at any time, each input of the largest self-similar traffic, the minimum threshold and the current buffer is not used proportional to the size of the buffer zone, and the Hurst coefficient is inversely proportional. Dynamic adjustment of RED algorithm to adapt self-similar network traffic.4. Finally, Compare the parameter of the network such as the average queue length, network throughout, network delay .etc through the simulation based on the NS2 simulation platform. The results show that the proposed self-similar traffic of new RED algorithm can improve network utilization, stability the average queue length network throughput and some other network performance.