Performance modeling of a finite capacity polling system with ATM bursty and correlated input traffic

Abstract

This paper is concerned with the mean delay and the probability of cell loss that bursty and correlated arrivals incur in an ATM switching system which can be modeled as a finite capacity polling system with nonexhaustive cyclic service. The arrival process to each input port of the system is modeled by a Markov Modulated Bernoulli Process (MMBP) which is able to describe the bursty and correlated nature of the ATM traffic. A practical polling system with finite capacity, as the one we deal with here, does not lend itself to an exact solution. In this paper, we introduce a tractable approach to provide an analytical approximation. This approach is validated extensively by comparing it against simulation results under different configurations. It is shown that both the mean delays and the cell loss probabilities obtained from this analysis provide highly accurate estimates.