Group priority scheduling

Abstract

We present an end-to-end delay guarantee theorem for a class of guaranteed deadline (GD) servers. The theorem can be instantiated to obtain end-to-end delay bounds for a variety of source control mechanisms and GD servers. We then propose the idea of group priority, and specialize the theorem to a subclass of GD servers that use group priority in packet scheduling. With the use of group priority, the work of packet schedulers can be substantially reduced. We work out a detailed example, for the class of burst scheduling networks, to illustrate how group sizes can be designed such that the worst case end-to-end delay of application data units in a real-time flow is unaffected by the use of group priority. Group priority also can be used in packet schedulers that provide integrated services (best effort as well as real-time services) to achieve statistical performance gains, which we illustrate with empirical results from simulation experiments.