An algorithmic technique for a class of queueing models with packet switching applications

The need to analyze the performance of a class of hardware packet switches leads to consideration of a queueing model consisting of a single server queue with input given by a function of a Markov chain. An algorithmic technique is developed to obtain the joint stationary distribution of the bivariate Markov chain which describes the system. Both the infinite and finite capacity cases are considered. The technique is used to study several design issues which arise when a packet switch is subjected to independent streams of bursty input traffic. Guidelines are given which aid in estimating the queueing space required to keep traffic losses acceptably small. It is seen that the commonly employed heuristic of dimensioning queueing space using the tail of the infinite capacity distribution can lead to considerable error when compared with exact results.