Workconserving vs. non-workconserving packet scheduling: an issue revisited

Abstract

Many packet schedulers for QoS networks are equipped with a rate control mechanism. The function of a rate control mechanism (rate controller) is to buffer packets from flows which exceed their negotiated traffic profile. It has been established that rate controllers lead to reduced buffer requirements at packet switches, and do not increase the worst-case delays in a deterministic service. On the other hand, rate controllers make a scheduler non-workconserving, and, thus, may yield higher average end-to-end delays. In this study, we show that by properly modifying a rate controller, one can design a scheduler which balances buffer requirements against average delays. We present a scheduler, called earliness-based earliest deadline first (EEDF), which achieves such a balancing using a tunable rate control mechanism. In simulation experiments, we compare EEDP with a rate-controlled EDF scheduler and a workconserving version of EDF.