Improvement On Parameter Settings Of RED In Router

1. IntroductionWith the rapid development of computer networks,the problem of network congestion was becoming increasely serious. When too much of a comepetion in the same router output link, the router will produce the queue overflow, some packets have been discarded by the router, which took place a discarded by the network is congestion. When congestion was heppaning, network performance delince, such as end-end delay intensify, the response time increase, significantly decreased throughput and increased dropping. Congestion control is to prevent and respond to congestion or router on the end of some control strategy. Active Queue Management is an important means to achieve congestion control. Congestion control is a key element in ensuring robust, as well as various management and control mechanisms on the basis of application, So as a hot issue in the current network.According to the position of the algorithm, Congestion control algorithms can be divided into two categories: link algorithms and source algorithms.2. EA-RED Algorithm2.1 Proposed of AlgorithmRED algorithm consists of two main parts. The first part is to compute the average queue length in order to determine the degree of unexpected flow allowed by the queue.The second part is to calculate the probability of packet marking to determine the probability Routing used to mark the data in the current extent of congestion. We found that there are two major problems of the Algorithm in the research of RED algorithm. One is that the parameters which are suitable in any case of vary conditions of load are unable to be found. The other one is the problem of fairness between different RTT, congestion window size, packet size, target speed and TCP/UDP's sharing bandwidth.Based on the averge queue and instantaneous of RED analysis of the fluctuation, we propose to vary the parameter w q and max padaptively based onthe variations of average queue length for improving the stability of RED, decreasing packets loss rates.2.2 Principle of EA-REDAs the average queue length raise between min thand max th, and close to max th, for restraining mandatory dropping, we change the parameter max p on different position. The average queue length increase fast then the grater the value of max p, but not arbitrarily increase, we will set up the maximum much to prevent packet loss probability.When the link located at the TCP Congestion Avoidance state, the instantaneous queue length is descended, if w q can susceptivity reflect the change, and make the average queue length to follow the instantaneous queue length for avoiding a great probability of packet loss. The tune can make w qto dynamicly response packets input queue rate and reduce the computational overhead. When the average queue length is up stage, the great w qwil urge the queue into mandatory loss, RED algorithm will become DropTail, so that we do not tune w qas queue length ascend.3. ConclusionIn the two components of RED algorithm, max pdecide the greatest loss probability of average queue length between the upper and lower limits, wq reflect in the degree of sensitivity of the average queue relative to instantaneous queue, which immediately reflected in the Network load. When the average queue length ascend we increase max p, when the workload is ascend we tune wq . On the basis of principle EA-RED algorithm has been proposed to tune max pand wq in the different stage of the queue.In this paper we have reported on EA-RED, and looked at the performance of the algorithm in relation to four algorithms. We have identified the impact of varying max p and wq on the link utilization and packet loss rate and stability of queue. We demonstrated our proposal have validaty. We conclude that improvedalgorithm maintains a stable average queue size and reduce RED's parameter sensitivity, improve the link utilization.