Deelnemer: Alwin Stegeman

Titel scriptie: Modeling traffic in high-speed networks by ON/OFF models

Anyone who has experience with computer networks knows that system crashes occur quite often. In order to increase the performance of the network, efforts have been put into the understanding of the traffic or workload (i.e. bytes per time unit) in computer networks. Empirical studies of the workload in computer networks have shown that the workload process exhibits long memory (long-range dependence) and that the cumulative workload process is self-similar. To understand why this is the case, theoretical models have been developed, describing the traffic in computer networks.

One class of these models are so-called ON/OFF models. In an ON/OFF model a computer network is represented by a number of so-called ON/OFF sources. An ON/OFF source can be compared with a fileserver: it sends data through the network if it is ON and remains silent if it is OFF. The lengths of the ON- and OFF-periods are stochastic.

The ON/OFF model is used to give an explanation for the empirically observed long memory and self-similarity. This is done by obtaining convergence results for the cumulative workload process. This approach is not totally convincing, however, since by choosing different limit regimes, different limiting processes can be obtained. Some limiting processes do not exhibit long memory. Therefore, the ON/OFF model does not give an unambiguous explanation for the empirically observed phenomena.