Multiple Active Queue Management within Input/Output Buffers

Abstract

The purpose of this research is to design and evaluate a new queue management algorithm effectively controlling the heavy-tailed traffics. Among active queue managements (AQM), Random Early Detection (RED), has been widely used to improve the TCP throughput. The heavy-tailed traffic, however, generates the large fluctuation on the queue causing the degradation of fairness in terms of jitter performance. In this paper, we proposed the Stable time RED (STRED) to improve the jitter and the fairness of jitter performance. STRED enables to stabilize the calculation timing for average queue length leading to suppress the queue fluctuations. We also implemented the STRED mechanism on the ns-2 simulator. In addition, we propose the pseudo model network consisting of multiple input / output buffers. The experiments of different pattern of AQM algorithms were conducted. As the results of simulation, we extracted the following features. Firstly, STRED could sustain the jitter values and large fluctuations of each file size and Pareto distribution. Secondly, STRED at the first incoming buffer made lower and smoother jitter values than the RED method. Thirdly, STRED could improve the fairness of jitter for all file sizes, and especially for heavy-tailed traffics. Finally, STRED creates low jitters and the high fairness in term of jitter meaning that STRED can enhance the quality of mutual communication applications such as Internet telephony.