Traffic modeling of IP networks using the batch Markovian arrival process

Abstract

In this paper, we identify the batch Markovian arrival process (BMAP) as an analytically tractable model of choice for aggregated traffic modeling of IP networks. The key idea of this aggregated traffic model lies in customizing the batch Markovian arrival process such that the different lengths of IP packets are represented by rewards (i.e., batch sizes of arrivals) of the BMAP. The utilization of the BMAP is encouraged by the observation that IP packet lengths follow to a large extent a discrete distribution. A comparative study with the MMPP and the Poisson process illustrates the effectiveness of the customized BMAP for IP traffic modeling by visual inspection of sample paths over four different time-scales, by presenting important statistical properties, and by analysis of traffic burstiness using R/S statistics. Additionally, we show that the BMAP model outperforms MMPP and Poisson traffic models by comparison of queuing performance. © Springer-Verlag Berlin Heidelberg 2002.