Based On The Parameter Of Red Congestion Control Algorithm In The Study Of The Design And Implementation

With the rapid development of communication technology and the continuous appearance of various applications softwares, the number of users accessing computer network is increasing dramatically. As a result, Internet has become a global information infrastructure for people to communicate with others. However, a mass of data is sent into network, leading to greatly increase of network flow, which further aggravates network congestion. Congestion control has emerged as an important problem that restricts the development of network. In recent years, Intermediate nodes-based AQM (Active Queue Management) has been a novel research direction of network congestion, and RED (Random Early Detection) is recommended as its unique candidate scheme by IETF. RED gateways detect the current network congestion level, and send congestion signals to the sources before serious congestion to control congestion. RED not only can achieve high throughput and low queueing delay, but also it can be implemented easily in network devices. Nevertheless, a lot of research shows that RED is quite sensitive to both control parameters and the variation of network traffics. The undesired character of RED significantly affects its wide application in practice. Therefore, how to improve the robustness of RED has become a research hot spot, and is also the main research content of the paper.According to a discrete TCP-RED feedback network model, the paper studies the nonlinear instability and parametric sensitivity with RED. Based on the analysis, we discuss the relation between the maximum drop probability and network parameters. In addition, the upper bound of the exponential averaging moving weight is obtained by satisfying the linear stability condition.To sum up the results, we propose a novel improved RED algorithm. The main work in the paper is as follows.1. A discrete TCP-RED is introduced to analyze the impact of the maximum drop probability on queue length and average queue length. Based on the model, the study discusses the value of the maximum drop probability related with network parameters when the average queue length stabilizes at the target, and its property.2. According to the model, the paper also studies the impact of the weight on average queue length. Moreover, we compute the maximum weight when the average queue length stabilizes at the target by solving the linear stability condition.3. To sum up the theoretic analysis, a novel adaptive modified algorithm is proposed, which auto-configures the weight according to the target queue length and network scenarios.Moreover, the network constant K is estimated according to the history information of the average queue size for a period of time, and is used to dynamically compute an appropriate value of the maximum drop probability in order to maintain the average queue length on a given target value.