Bayesian Analysis of Overdispersed Count Data with Application to Teletraffic Monitoring

Abstract

Traffic modelling of modern telecommunication networks is a concern of major importance. Potential applications of efficient traffic models would be numerous including dynamic bandwidth allocation, network dimensioning or statistical multiplexing... As of today however, the characterization of network traffic is still an open and challenging statistical issue. It has first been reported by several authors that traditional traffic models (homogeneous Poisson model) are largely inappropriate for the arrival processes measured on most types of network connections, and especially for wide-area networks such as the internet. Moreover, it has been demonstrated that many teletraffic data sets exhibit nonstandard characteristics such as heavy tailed distribution or long range dependence. Finally, the huge size of the data sets involved imposes severe computational constraints on the analysis methodology.