Performance evaluation of fairness-oriented active queue management schemes

Abstract

Active queue management (AQM) schemes are a class of queueing algorithms designed to surmount some of the shortcomings of classic drop-tail queues in best effort networks. Most AQM algorithms are primarily designed to improve congestion control through early notification. However, significant work has been done on more advanced AQM schemes designed to protect responsive flows against unresponsive traffic. Such schemes are designed to identify and penalise unresponsive flows, which may use an unfair share of the available resources. We evaluate six algorithms through simulation-based experiments. The schemes chosen employ lightweight mechanisms for approximating fair bandwidth sharing. They are designed to be scalable and allow for incremental deployment in the current best-effort Internet infrastructure. Evaluation of their performance is effected under various network traffic conditions. The operational complexity of the schemes is also assessed.