AURED – Autonomous Random Early Detection for TCP Congestion Control

Abstract

Congestion control is widely used in the Internet to prevent congestion collapse. Because data are inherently bursty, routers are provisioned with large buffers to absorb this burstiness and to achieve high link utilization. At the same time, large buffers lead to high queuing delays at congested routers. RED (random early detection) was introduced to relieve this problem so as to achieve high link utilization and low queue delay. Several adaptive techniques have also been proposed to allow for better parameter adjustment under different situation settings. However, parameter adjustment approaches in these techniques are usually based on an assumption that there exists a known combination of optimal parameter setting based on which techniques are to adjust to. Whereas, optimality of the setting very much depends on circumstantial factors which cannot be universally true, and thus the ensuing adjustment may not be even beneficial. In this paper, we propose an autonomous random early detection (AURED) technique to allow for a complete autonomous adjustment process without having to assume the aforementioned association. By tuning the packet drop probability variable according to the performance variation between two consecutive sampling periods, this technique does not require a target setting value to adapt to, thus allowing for more flexibility to accommodate for various situations.