Networks of Queues with Long-Range Dependent Traffic Streams

Abstract

Traditional applications of queuing theory in the analysis and design of communication networks often use the fact that quasi-reversible queues can be interconnected with Bernoulli routing to form network models whose stationary distribution is of product form. To construct such models it is necessary to assume that the external arrival process is a Poisson process (in continuous time) or an i.i.d. sequence of Poisson random variables (in discrete time). Recently, however, several empirical studies have provided strong evidence that the traffic to be carried by the next generation of communication networks exhibits long-range dependence, implying that it cannot be satisfactorily modeled as a Poisson process.