Analytic Models for Separable Statistical Multiplexing：（分析模型对分离统计多路复用）.pdfVIP

16 Analytic Models for Separable Statistical Multiplexing Keith W. Rossi and Veronique Vequt? (1) University of Pennsylvania and (2) Universite de Paris-Sud (1) Department of Systems, University of Pennsylvania, Philadelphia, PA 19104, USA. Telephone: 215-898-6069. Fax: 215-573-2065. email: ross@eniac. seas. upenn. edu (2) Laboratoire de Recherche en Informatique - CNRS, Universite Paris-Sud, 91405 ORSAY CEDEX, FRANCE. Telephone: 33-1 Fax: 33-1 email: vv@lri.fr Abstract We investigate a multiplexing scheme for ATM that statistically multiplexes VCs of the same service, but does not statistically multiplex across services. The scheme is imple- mented by allocating bandwidth to each service. In the static version, the allocations are fixed; in the dynamic version, the allocations depend on the numbers of VCs in progress. Under minimal assumptions, we show that the distribution of the VC configuration has a product form. We use the product-form result to construct an efficient convolution algo- rithm to calculate VC blocking probabilities. We give a numerical example that demon- strates the rapidity of the algorithm and the potential efficiency of separable statistical multiplexing. Keywords Admission control, loss networks, performance evaluation, statistical multiplexing. 1 INTRODUCTION It has long been known that statistical multiplexing of cell streams of the same service type can be highly cost efficient. This is true for delay-sensitive as well as delay-insensitive services. For example, statistical multiplexing of packet streams emanating from voice sources has long been used by telephone companies to increas