A Dynamic, Tunable, QoS-Fair Scheduling Scheme for Mutlimedia Streams

A great deal of packet scheduling research in high speed integrated, multimedia services networks (ISNs) aims to efficiently emulate the Generalized Processor Sharing (GPS) service discipline. The approach used by these schemes has been, first, to maximize the emulation accuracy through the use of packet-by-packet scheduling decisions and then to minimize the complexity involved in making such decisions. While reductions in algorithmic complexity are useful, we believe the straight-forward GPS emulation prescribed by these schemes is not necessary in ISNs. It is the fulfillment of quality of service QoS guarantees, not GPS accuracy, which represents the primary metric for success in ISNs. In this paper, we present a new scheduling approach, QoS-Aware Fair Sharing (QFQ), designed to reduce packet scheduling overhead while guaranteeing the QoS requirements of supported applications. Because the central goal of the QFQ discipline lies in meeting the QoS requirements of backlogged flows and not guaranteeing a weighted-fair portion of the server’s output to such flows, the QFQ approach provides ISN servers with a great deal of flexibility in determining how to service contending application flows. This paper presents this new and innovative scheme which seeks to exploit an excess capacity allocation policy in order to achieve a reduction in service overhead. This paper describes how such an objective can be achieved and shows how excess capacity can be identified for this purpose despite the presence of non-empty service queues. A set simulation experiments demonstrate the capabilities of the QFQ server and compare its performance to other ISN scheduling schemes.