Abstract
Discrete service disciplines have been shown capable of emulating the ideal Generalized Processor Sharing (GPS) discipline within one maximum-sized packet's transmission time. As a result, the accuracy of such schemes increases with improvements in link speed due to the corresponding reduction in packet transmission delay. However, the merits of such accuracy diminish with improvements in link speed as the impact on call admission decisions decreases. Meanwhile, overhead, in terms of the number of scheduling decisions to be made per unit time, increases with the number of packets transmitted. In response, this paper will present an overview of the QFQ service discipline which enables GPS emulation servers to dynamically "tune" their service quanta at run-time based upon the QoS requirements of their currently backlogged connections rather than the statically defined maximum network packet size. Through simulation, the paper will also demonstrate how the overhead of a GPS emulation server can be reduced in high speed networks without jeopardizing QoS guarantees or adversely impacting fairness.