BCTCP: A feedback-based congestion control method

摘要

Delay and throughput are the two network indicators that users most care about. Traditional congestion control methods try to occupy buffer aggressively until packet loss being detected, causing high delay and variation. Using AQM and ECN can highly reduce packet drop rate and delay, however they may also lead to low utilization. Managing queue size of routers properly means a lot to congestion control method. Keeping traffic size varying around bottleneck bandwidth creates some degree of persistent queue in the router, which brings in additional delay into network unwillingly, but a corporation between sender and router can keep it under control. Proper persistent queue not only keeps routers being fully utilized all the time, but also lower the variation of throughput and delay, achieving the balance between delay and utilization. In this paper, we present BCTCP (Buffer Controllable TCP), a congestion control protocol based on explicit feedback from routers. It requires sender, receiver and routers cooperating with each other, in which senders adjust their sending rate according to the multiple bit load factor information from routers. It keeps queue length of bottleneck under control, leading to very good delay and utilization result, making it more applicable to complex network environments.