The Desirability of Pareto Distribution for Modeling Modern Internet Traffic Characteristics

Abstract

The Internet is rapidly growing in number of users, traffic levels and topological complexities. A vital component to understand the requirements, its capabilities and extension of the network is its traffic analysis. In literature, conventional traffic model does not capture today’s burstiness in Internet traffic with the assumption that inter-arrival time is independent and identically distributed. The Internet forwarding scheme has given provision for correlated inter-arrival times which are rampant when streams of packets arrive to a destination at the same time. Modern researchers are contemplating of modulating Poisson process with another distribution in form of Compound Poisson Process, Markov Modulated Poisson Process or use of Pareto model to tract actual traffic characteristics. Pareto is assumed in literature to have facility to cater for burstiness. This paper uses goodness of fit test to detect the exponential nature of the inter arrival times of Pareto generated network packets. A network simulator was used to generate statistics of arrival times of traffics were traced, inter-arrival times calculated, ranked and subjected to goodness of fit test. The analysed result fell into acceptable region which implies that Pareto can stand in place of conventional exponential distribution in modeling inter-arrival times of a bursty network.