Analysis , Approximations Andadmission Control of Amulti - Service Multiplexing Systemwith

We consider an ATM system with an architecture which is designed to accommodate users with very diierent quality of service requirements. In the base case with only two services, sources which require low loss belong to a High Priority class, and share a FCFS buuer, which has priority access to the trunk. A Low Priority class of sources with typically less stringent requirements on loss have a separate FCFS buuer, which receives the time-varying, residual bandwidth, if any, of the trunk. By administering admission control and restricting the combination of sources to an admissible set, the service guarantees for both classes may be satissed. The sources are bursty and stochastic uid models are used to handle burst-scale congestion eeects. Our contributions are: (i) we develop simple, fast and robust analytic approximations for the queue distributions in the two buuers; (ii) for admission control, we calculate the admissible set by using our analytic approximations and nd that it is reasonably accurate and a conservative approximation to the correct set. We developed a uid-ow simulator for general Markov-modulated sources, and a program which implements our analytic techniques for on-oo sources. Systems with over 100 sources are completely analyzed in a few seconds, while simulations take hours. The key element in our analysis is a characterization of the output of the high priority buuer as another Markov-modulated uid source. A reenement to the now well known eeective bandwidth approximation is used to calculate buuer content distributions.