Parameters Setting Scheme of RED with Long-Range Dependent Traffic Input

Abstract

Random early detection (RED) is the most famous active queue management (AQM) algorithm and is widely used in current Internet. But classical RED is based on Poisson model which is short-range dependence. And what is more, RED is sensitive to parameters setting. Current Internet traffic has ubiquitous characteristic of self-similarity or long-range dependence (LRD). It is necessary to recheck the RED with LRD traffic input. In this paper, the issue of RED parameters setting with LRD traffic input is studied. LRD characteristic of the network traffic is taken into account in the setting of the RED parameters. By using auto-correlation function as filter weight of RED, the current average queue length is predicted. The packet dropping/marking probability is computed based on the analytical result of Fractional Brownian Motion (FBM) model. The maximum and minimum queue threshold is derived based on the critical time which makes the queue length to achieve its maximum value in bursty Period. Intensive simulations are performed to validate the accuracy and effectiveness of RED parameters setting scheme, and results indicate that the proposed scheme can well control the average queue length and raise the performance of RED algorithm.