Traffic characterization for QoS provisioning in high-speed networks

We are developing a new approach to traffic characterization motivated by the traffic-queue analysis techniques of effective bandwidth and the use of log moment generating functions. A generic representation of traffic exhibiting non-Gaussian marginal and long range dependence is modeled by three dissimilar stochastic models. Generated traces of the model traffic are put to test in a G/D/I/∞ system by computer simulation. Experimentation demonstrates that the marginal and autocorrelation pair of descriptors do not adequately characterize traffic such that different input models with similar marginal-autocorrelation produce vastly different queuing behavior. The ability of the new approach to describe and differentiate stochastic traffic is shown to be robust with regard to queuing results, whereas marginal and autocorrelation prove insufficient