Research On Adaptive RED With Self-Similar Traffic Input

With the development of network technology, the network scope extended rapidly. Internet has experienced an explosively growth,With the appearance of many kinds network applications, especially various multimedia applications and the rapid growth of Internet users which have brought forth the rapid growth of Internet traffic, This Internet traffic has caused network congestion ,the network congestion results in a significant problem that obstructs the development and application of Internet. In recent years, a great deal of measure and analyses on actual network traffic indicated that the network traffic exists distinct self-similarity. Due to the great difference between the network performance of self-similar and that of traditional model, network management and congestion control are quite different from them.In this thesis, we firstly analyze two kinds of congestion control algorithms and discuss the advantages and disadvantages of the common Active Queue Management (AQM) algorithms in detail. Then we validate that self-similar property of network traffic has great effect on the performance of AQM algorithms through simulation. Based on the characteristic of self-similar traffic, we improve the Adaptive Random Early Detection (ARED) and propose a new algorithm named improved adaptive random early detection (IARED) with self-similar traffic input. Finally, we introduce Linear Minimum Mean Square Error (LMMSE) traffic predict mode into IARED and propose a new algorithm, namely PIARED (prediction IARED) which based on the prediction of self-similar traffic. The performance of IARED algorithm is compared through simulation.The main research achievements of the thesis are as follows:First, we study the common AQM algorithms and analyze their basic principle and merits/faults, Then the performance of these algorithms with short-range dependent traffic and self-similar traffic input are compared by simulating with OPNET Modeler 10.0.We also compare the performance of the common AQM algorithms with different degree of self-similar traffic.Second, based on the characteristic of network traffic, an improved adaptive random early detection algorithm named IARED with self-similar traffic input is proposed. This algorithm uses the autocorrelation function as the weight of average queue length and dynamically adjusts the maximum packet dropping/marking probability according to two parameters, namely the change ratio of the current average queue length versus target queue length and the change ratio of current average queue length versus last average queue length. Finally the validation of IARED is test through simulation.Third, based on the predictability of self-similar traffic, LMMSE forecast model is introduced into IARED algorithm to predict the interval time of the arrival packets.It can dynamically adjust the maximum packet dropping/marking probability with the change of self-similar network traffic and made the average queue length more stable. Finally the validation of this new prediction-based algorithm is verified through simulation.