A study of IP router queues with the use of Markov models

Abstract

We investigate the use of Markov chains in modeling the queues inside IP routers. The model takes into account the measured size of packets, i.e. collected histogram is represented by a linear combination of exponentially distributed phases. We discuss also the impact of the distribution of IP packets size on the loss probability resulting from the limited size of a router memory buffer. The model considers a self similar traffic generated by on-off sources. A special interest is paid to the duration of a queue transient state following the changes of traffic intensity as a function of traffic Hurst parameter and of the utilization of the link. Our goal is to see how far, taking into account the known constraints of Markov models (state explosion) we are able to refine the queueing model.