Research On Active Queue Anagement Algorithms In IP Network

In recent years, the audio frequency, the video frequency and so on the multimedia applications unceasingly emerged on Internet, the more and more serious network jam question exposed gradually. In the network correspondence, the jam easy to create QoS performance indices and so on the detention and volume of goods handled drops, is affects network resources use factor and so on the band width, buffer key aspects, for solves these problems, only increases the Internet the capacity is by far insufficient, but also needs to have keen and the effective network congestion control method. Therefore the congestion control always is the network research area hot topic.The results of study and main idea in the paper are given as follows: In this paper, the principles of congestion control mechanisms are introduced firstly, and the congestion control in the IP layer is discussed in detail. RED algorithm in the AQM is the focus of this paper. Main research work in the paper as follows:(1) As RED just use the average queue length when packets arrives, thus the computation shows averaging of packet arrival rate rather than the current buffer occupancy. The dropping of packets will continue for a long time even after the queue has become empty. This will leads to a rapid degradation of the link utilization. To resolve the question, RED algorithm is improved in the paper, in the average formation length computation and utilization buffer storage, which combines EWMA with instantaneous queue size for dropping decisions. Finally the computer simulation is carried in network simulation software NS. Simulation studies show that the improved RED algorithm gives better loss rate and link utilization compared to the original RED algorithms.(2) Considering the nonlinear, time-varying and uncertain characteristics of networks, a new active queue management (AQM) algorithm, i.e. PSRED was proposed based on predictive control theory, which requires less model accuracy. In order to realize the fast control, a predictor was constructed using mix neural network to predict future frequently formation's discarding probability. By exploiting feedback adjustment, the proposed algorithm requires less model accuracy, together with decreased computation and fastened speed. Simulation results show that this algorithm possesses robustness against variances of network parameters.