Study On Congestion Control Base On Middle Nodes

With the rapid growth of network users, network congestion problems become more serious. Congestion control mechanism provides the transmission speed of data flow effective matched with available bandwidth. Active queue management algorithm (AQM) and queue scheduling algorithm (QS) implemented in middle nodes in time than in end nodes reflect the congestion, and thus become a research hotspot in network congestion control.Firstly, this paper analyses the shortcoming of congestion control mechanism implemented in end nodes, points out the nature of congestion control, and depicts the categories, performance and relationship of AQM and QS.Secondly, A AQM algorithm can't reach all of the targets such as fairness, stability and extendibility, so we propose a new self-adaptive AQM algorithm based on packet arrival time, called AI-AQM (Average Interval-AQM). AI-AQM congestion measure improves REM congestion price formula, and then gives a method which is used to provide self-adaptive parameter for AI-AQM. According to the average packet arrival time (AI) of different data flow, AI-AQM finds the non-adaptive flow.To reduce spending, AI-AQM only maintain the data flow state information whose AI is less than that of the fair limit.Thirdly, To solve the problem of divide and rule between AQM and QS, we propose AI-RR (Average Interval-RR) algorithm.We group all of data flows according to their state information in AI-AQM, and use the ratio of number of data packets and weight as the standard of bandwidth distribution. In order to aviod the unfairness which is caused by different packet length among data folws, AI-RR corrects old weight.Round-Robin is carried out descending to the weight value among different groups. We use first come first served in the inner of group, and smooth the output of Round-Robin.Finally, we set up experimental environment built on NS2. The performance of AI-AQM, RED, FRED, CHOKe and CSFQ are compared through the experiments.The simulation results show that AI-AQM has better performance than RED, FRED, CHOKe and CSFQ such as the packet loss rate, link utilization, fairness among different flows and stability of queue length.