Modeling of temporal dependence in packet loss using universal modeling concepts

Packet losses are commonplace over the Internet, and can severely affect the quality of delay sensitive multimedia applications. In the current Internet architecture it is up to the application to react to the perceived congestion level in the network. The ability of the application to react is enhanced by the availability of simple and efficient loss models. Traditionally, Markov chain models have been proposed for modeling of the dependency in packet loss. In this abstract we establish the drawbacks of using a Markov chain model and instead propose the more general Markov tree model for modeling the temporal dependency. Markov tree models are an example of universal models. In universal modeling, the most appropriate model for the observed data is chosen from a collection of models. Entropy of the data with respect to a model is used as the performance measure so that the model which gives the lowest entropy for the data is chosen as the best model. Our results show the advantage of using the Markov tree model.