On-line predictive techniques for "differentiated services" networks

Abstract

We propose two on-line predictive algorithms that can provide performance estimates under a set of parameters in the context of differentiated services. These algorithms are based on sample path constructability techniques. The first one (ASA) depends on exponential lifetime distributions, while the second one (ML) is distribution independent. The information provided by these algorithms can be used to solve the admission control problem and to allocate resources to the various traffic streams.