Modelagem de tráfego VoIP utilizando misturas gaussianas ajustadas por algoritmo de maximização de esperança e comprimento de descrição mínimo

Abstract

The increase in the number of applications that generate VoIP (Voiceover IP) traffic requires that suitable models for this kind of traffic be used to bet-ter dimension and predict network resources. In this paper,we propose the applica-tion of Gaussian mixture adjusted by an EM algorithm (Expectation-Maximization)with selection of classes using the principle of Minimum Description Length (MDL-Minimum Description Length) in order to describe VoIP traffic. Real VoIP traces werecollected and synthetic time series were generated according to these data. The propo-sed algorithm is compared to the EM algorithm and the MDL (Minimum DescriptionLength) based algorithm in terms of performance in modelingthe considered VoIPtraffic traces.