並列摘要
It's becoming one of the major research issues that how to handle data congestion and avoidance of that on the network in an efficient way has become a crucial issue because of bursting data traffic. There are a variety of congestion control strategies have been proposed to avoid congestion at the bottleneck nodes. Among which, random-early-detection (RED) is widely applied in IP network nowadays. It provides high bandwidth (BW) utilisation by dropping packets randomly selected from the queue. These dropped packets will then serve as signals that notify their transmission ends to decrease the sending rate, preventing congestion from happening. Unfortunately, RED algorithm provides little protection against aggressive flows, such as flows with large packets, or congestion insensitive flows. As a result, it is easy for these flows to consume most of the bandwidth, causing unfair bandwidth sharing. In this paper, we propose a new algorithm, named as Power-RED, which aims to achieve BW fairness. Power-RED keeps a throughput statistics, and adjusts packet drop probabilities according to power law when a network tends to become congested. Its goal is to maintain a fair bandwidth share among all incoming connections. Power-RED monitors throughput of active flows, and adjusts drop probabilities individually according to their throughputs for the purpose of BW fairness. The extensive real-world experiments in a PC networks shows that Power-RED can effectively guarantee fairness not only in packet numbers but also in their sizes.