Stochastic Bounds for Loss Rates

Abstract

We propose to use stochastic comparison method in order to compute packet loss rates in IP routers for MPLS networks. We study an IP buffer with finite capacity B and we consider two kinds of packets. Those who are delay sensitive and those who are loss sensitive. We propose a new service discipline which we denoted by "LATIN" service discipline and for buffer management; we propose to use push-out. The state of the buffer can be represented by a discrete Markov chain defined on a large state space size O(B3). We propose bounded models with HOL disciplines, which are easier models to solve and using stochastic comparisons, we prove that the bounded models provide really upper and lower bounds on the packet loss rates. We present some numerical results to show the accurate bounds on loss rates.