The impact of infinite variance processes on effective bandwidth estimation

With up-coming Quality of Service requirements raised by a wide range of communications demands, the effective bandwidth estimation is one of the approaches to satisfy these QoS requirements. This work compares the effective bandwidth estimation based on Gaussian and non-Gaussian appraximadons, for aggregated traffic, taking the finite and infinite variance cases into account and evaluating its impacts on bandwidth estimation. We briefly describe the Norros approach based on fractional Brownian motion. Also, we present the generalization method proposed by N. Laskin based on fractional Levy motion. Experimentally we compare the estimated effective bandwidth using real aggregated traffic traces providing benchmarks for both cases. We observed that Laskin model overestimates the effective bandwidth in some region, then we optimized the results in order to reduce this problem For this end, analytical, heuristical and simulation methods were used jointly to validate this work This study confirms that we have carefully differentiated the self-similar processes with finite variance from infinite variance cases, in order to estimate its effective bandwidth for aggregated traffic.