Empirical modeling of Internet traffic at middle-level burstiness

In this paper we use empirical data from Internet traffic measurements. Collected measurements are analyzed for different protocols, such as TCP and UDP. We perform statistical analysis through the correlation coefficients, covariance, and self-similarity degree, i.e. Hurst parameter. Our experimental studies captured traffic with Hurst parameter around 0.7-0.75, which is near half way between values of 0.5 (it is not a self-similar) and 1 (strong self-similar properties). We use maximum likelihood approach to fit the obtained time series to existing distributions, such as Pareto and exponential distribution, where the first one is a self-similar process and the second is not. The analysis pointed out that Internet traffic with such values for the Hurst parameter could be modeled with similar accuracy using both Pareto and exponential distribution.