On exploiting traffic predictability in active queue management

In this paper, we explore the issue of exploiting traffic predictability to enhance the performance of active queue management (AQM). We show that the correlation structure present in long-range dependent traffic ran be detected on-line and used to accurately predict the future traffic. We then design, with the objective of stabilising the instantaneous queue length at a desirable level, a LMMSE-based controller, and figure in the prediction results in the calculation of the packet dropping probability. The resulting scheme is termed predictive AQM (PAQM). Through analytical reasoning, we show that PAQM is a generalized version of RED with a new dimension of congestion index - the amount of traffic that will arrive in the next few measurement intervals. By stabilizing the queue at a desirable level with consideration of future traffic, PAQM enables the link capacity to be fully utilized, while not incurring excessive packet loss. Through ns-2 simulation, we compare PAQM against existing AQM schemes with respect to different performance criteria. In particular., we show that under most cases PAQM outperforms SRED in stabilizing the instantaneous queue length, and adaptive virtual queue (AVQ) in reducing packet loss ratio and better utilizing the link capacity.