Generalized Processor Sharing scheduling enhancement through session decomposition

Abstract

In this paper a service discipline, which is based on Generalized Processor Sharing (GPS) and aims to improve resource utilization for deterministic guarantees at a single node is presented. According to the proposed seheme original sessions are decomposed into two components which are treated as separate sessions by a GPS seheduler; the work assigned to eaeh session components is mapped back to the original session. A Call Admission Control (CAC) algorithm for the proposed seheme - capable of ensuring typically better and never worse resource utilization compared to that achieved by the optimal CAC applied directly to the sessions - is derived, and some numerical results demonstrating the benefits of the proposed scheme are provided. © 2003 by Springer Science+Business Media New York.