Adapting Fair Queueing to Improve Loss, Delay, and Admission Control Performance

Fair queueing provides minimum rate guarantees and is fair by being rate proportional, i.e, excess bandwidth due to non-backlogged connections is distributed in proportion to rate guarantees. State dependence of the scheduler is restricted to whether specific connections are backlogged or not. In this paper, we propose three alternate scheduling mechanisms that can provide better service, in terms of losses and delay variation, than fair queueing. The mechanisms preserve the minimum rate guarantees of fair queueing but distribute the excess bandwidth in a manner much more adaptive to instantaneous needs than fair queueing. We compare the performance of the proposed schemes to that of fair queueing using different traffic sources such as voice and video, as well as sources which have aggregate long-range dependent behavior. We find that the proposed schemes, in comparison to packet generalized processor sharing (PGPS), reduce packet losses and curtail the tails of delay distributions for real-time traffic and hence permit the use of significantly smaller playout buffers for the same network load. We argue that these schemes add robustness in measurement-based admission control by providing adaptation for connections whose demands have been underestimated during the measurement process.