Measurement–Based MMPP Modeling of Voice Traffic in Computer Networks Using Moments of Packet Interarrival Times

Abstract

In this paper we propose a novel technique for modeling packet voice traffic in computer networks based upon the measurement and estimation of the first three moments of packet interarrival times. The moments of an interarrival process are estimated using the approximation technique proposed in [1] which is applicable to any arbitrary interarrival time distribution. We then use a technique based on the pioneering method of Heffes and Lucantoni [2] for approximating the superposition of packetized voice streams from heterogeneous sources using a correlated Markov modulated Poisson process (MMPP). This is carried out in such a way that several important characteristics of the superposed traffic are matched with those of a two-state MMPP. The performance of a voice multiplexer with heterogeneous superposed packet voice traffic is then found by modeling the multiplexer as an MMPP/D/1 queuing system. A simulation model of the system is developed. Simulation results agree more closely with results obtained by the proposed technique than by the Heffes-Lucantoni technique. Practicality of the proposed technique has been demonstrated.