General solution to the losses estimation for multifractal traffic

In this work, we present a new approach for loss probability estimation in a single server link. We show how to get the estimates analytically once we assume multifractal input traffic. In order to make the estimation procedure numerically tractable without losing the accuracy, we propose the use of a Gaussian mixture model to represent the heavy tail distribution of modern network traffic trace. The adopted evaluation procedure is based on two performance measures: empirical traffic arrival load distribution and loss probability at connection. Extensive experimental tests validate the efficiency and accuracy of the proposed loss probability estimation approach against the results obtained by simulations with real traffic and by comparing with other multifractal approaches suggested in the literature.