Towards fully automated adaptation of RED parameters to optimise TCP performance

Abstract

Active queue management mechanisms, such as random early detection (RED), aim to optimise TCP/IP behaviour through probabilistic congestion notification. Increasing the drop probability with the mean queue state can provide both high throughput and low delay; however, RED's effectiveness depends on tuning parameters to the specific traffic scenario. Various adaptive schemes maintain the queue length within a target range by adjusting the notification (i.e. drop) profile. But even this target range depends on the variability of the traffic. A new scheme is presented that uses the first two moments of the measured arrival rate distribution to adapt RED's queue thresholds and its maximum drop probability. Simulation results demonstrate that the scheme can simultaneously maintain low delay and high throughput.