Affiliations: Room 3M-318, AT&T Bell Laboratories, 101 Crawfords Corner Road, Holmdel, New Jersey 07733, USA | Department of Computer Science, University of Massachusetts, Amherst, Massachusetts 01003, USA
Note: [] This work was supported in part by the Office of Naval Research under contract N00014-90-J-1293 and the Defense Advanced Research Projects Agency under contract NAG2-578.
Note: [] Presented in part at IEEE INFOCOM '92, Florence, Italy, 1992.
Note: [] This work is part of the Ph.D. dissertation of Ramesh Nagarajan at the University of Massachusetts, Amherst.
Abstract: Future high-speed networks (HSNs) are expected to support a wide variety of services such as voice and video and provide a guaranteed quality-of-service (QOS). Traditionally, the computation of user-oriented performance criteria such as the average delay has been carried out via steady state analysis of queueing theoretic models of communication networks. In this paper, we show that the steady state analysis is not entirely sufficient for QOS purposes in future high-speed networks as it yields long-run performance measures that are not appropriate for envisaged applications in HSNs. New QOS criteria for such applications are proposed and their computation detailed for some simple queueing models. These new QOS criteria essentially capture the network performance over the short-run and are more accurate indicators of the actual user-perceived quality of, for example, audio or video. For the numerical examples presented in this paper, our focus will be primarily on the fractional packet loss and we will be interested in the probability (frequency) of occurence of high or low loss “periods.” When our interest is in maximizing the occurence of no loss periods, we show for the M/M/1/K queue that reasonably high values of traffic load may be supported and fairly good QOS (as defined in this paper) provided to applications. In this context, steady state analysis is found to be woefully inadequate. On the other hand, when we are interested in minimizing the occurence of high loss periods, we find for a packet voice multiplexer that large values of the carried load may not be supportable while providing an acceptable QOS.
