Decoupled generalized processor sharing: a fair queueing principle for adaptive multimedia applications

Traditionally targeted to best-effort packet-switching networks, adaptive applications implement a variety of mechanisms to make use of variable quality bearing service. In the framework of integrated services networks, reducing quality of service (QoS) requirements through the use of such applications allows a greater operation flexibility and increases statistical multiplexing gains. However, these applications must be given some QoS guarantees, relative to minimum service, bounded transmission delay, and fair sharing of the available bandwidth. This paper focuses on a fair queueing, fluid-flow model to be embedded in the network switching nodes. This model is based on the generalized processor sharing (GPS) paradigm, but avoids its inherent limitations. It is approximated by means of a dynamic priority algorithm. It is shown that an implementation having reduced complexity can be achieved and exhibits good service characteristics, relative to packet service delay and conformance to the fluid-flow model fairness. Simulation results give evidence that the resulting algorithm accurately meets the needs of adaptive applications