Fractal Analysis and Modeling of VoIP Traffic

In this paper a fractal analysis study of VoIP traffic is presented. The characteristics of measured VoIP traffic on both call and packet level have been investigated. The results support the popular Poisson process for VoIP call arrival modeling but we argue that the call holding times follow heavy-tailed distributions rather than exponential distributions. We propose the generalized Pareto distribution for modeling the call holding times. At the packet level we have found that the exponential modeling of On and Off periods is also inappropriate and heavy-tailed characteristics have been identified in case of all the investigated VoIP codecs. The generalized Pareto distribution can be used as an accurate model for the On and Off periods too. In the analysis we revealed that the aggregated VoIP traffic has fractal characteristics and we suggest the fractional Gaussian noise model for the aggregated VoIP traffic