Research On Active Queue Management Algorithms With Self-Similar Traffic Input

With the arrival of information age, the amount of information on Internet which is as the important infrastructure explosively increase, but the "Best Effort" service mode of Internet can not meet the higher performance and Quality of Service (QoS) requirement of the applications, such as distributed multimedia etc., as a result that network congestion is becoming the key factor restricting the development of Internet. On the other hand, a large number of researches indicate that network traffic exhibits ubiquitous property of long-range dependence and self-similarity, which brings new challenges to network control.The performance of Active Queue Management (AQM) with self-similar traffic input is studied and compared with the traditional traffic input through simulation. We focus on the study of the problem of the parameters setting of Random Early Detection (RED) algorithm with self-similar traffic input, and propose two new RED algorithms based on Fractional Brownian Motion (FBM) and Fractional Stabilized Motion (LFSN) respectively. Finally, we adopt FARIMA (Fractional Autoregressive Integrated Moving Average) traffic forecast model, introduce the traffic prediction result into AQM algorithm, put forward a new prediction-based AQM algorithm, and compare with the AQM algorithm based on AR model through simulation.The main research works and achievements of this thesis are as follows:First, we study the common available Active Queue Management (AQM) algorithms, and analyze their basic principle and merits/faults. The major AQM algorithms include RED, Adaptive RED (ARED), BLUE, Flow RED (FRED) and Stabilized RED (SRED). The performance of these algorithms with traditional short-range dependent traffic input is compared with that of with self-similar traffic input by simulating with OPNET Modeler 10.0.Second, a new RED algorithm based on FBM model is proposed, which Hurst coefficient of the traffic is considered into the weight function, and average queue size in the next prediction cycle is updated. We also set up the maximum and minimum queue threshold by using mean and variance of system queue size deduced from FBM model. After that, we revise the packet discard probability according to the formula of the buffer overflow probability of FBM model, and verify the validity of this new algorithm through simulation.Third, another new RED algorithm is proposed based on Fractional Stability Motion. The weight function of the algorithm is designed according to the heavy tail characteristic of network traffic, and the average queue size in the next prediction cycle is set up according to this weight function. At the same time, we also set up the maximum queue threshold, and get the packet discard probability relying on the formula of the buffer overflow probability of LFSN process. Finally, the validity of this new algorithm is verified through simulation.Fourth, adopting FARIMA traffic forecast model, we propose RED algorithm based on traffic prediction which introduces the traffic prediction result into RED algorithm, dynamically controls and adjusts the packet discard probability. This prediction-based RED algorithm is compared with that of AR-based model through simulation.