Advanced traffic modeling: fitting third order moments

We have presented an efficient discrete model for the generation of self-similar traffic (Potemans, J. et al., Proc. ICC´03, 2003). The model is based on a hierarchical scheme of Bernoulli sources, each working on a different timescale. The major advantage of the model is that it allows traffic to be generated with an arbitrary combination of the Hurst parameter, the variance and the mean. We further refine this model by also fitting third order moments of traffic and by compensating for non-stationary effects. A comparison of the queuing behavior of measured traffic and synthetically generated traffic shows that these refinements make the model more accurate.