SwRED: A Robust Active Queue Management Scheme based on Load Level Prediction

摘要

Active queue management (AQM) is a novel technique to realize a smooth congestion control in communication networks, which is expected to work effectively by combining it with end-to-end congestion control schemes such as the window control adopted in TCP. Although it has been proposed a lot of AQM schemes in the literature, an appropriate tuning of parameters used in those schemes is recognized as a challenging problem to be overcome. In this paper, we propose a new AQM scheme based on a load level prediction of bottleneck links, and on a combination of queue-based and rate-based AQM schemes in multiple time scales. Different from most of conventional AQM schemes, the proposed scheme, called Switch-RED (SwRED), employs a new packet drop strategy based on a macroscopic analysis of the queue dynamics. The performance of the resultant scheme is evaluated by simulation. The result of simulations indicates that SwRED realizes a more stable queue behavior compared with previous schemes, i.e., it realizes a low queuing delay and a low packet loss rate while keeping a high link utilization.