| The proposed EPIT approach in this thesis is a hybrid of simulation and analytical methods. By treating a network as a black box, it scales well against complex network topologies. EPIT can predict the end-to-end QoS performance in a network under any traffic model (including LRD traffic models) that has a closed-form joint distribution function. In addition, EPIT has no constraints on the behaviour of the traffic itself.;The resulting approach of Simulated Annealing (SA) integrated with EPIT (SA-EPIT) provides a traffic engineering solution, so that multiple end-to-end QoS requirements are satisfied while the network resource utilization is maximized. The value of this research is that, in fulfilling end-to-end QoS requirements while maximizing the network utilization, EPIT is an attractive solution to the deployment of applications such as QoS-aware VPN and Peer-to-Peer communications.;The EPIT approach is effective, as it can project a wide range of probabilities of end-to-end overflow events accurately, with a fixed computation cost, in a network with multiple interconnecting nodes and multiple LRD input sources. Insights into the efficiency of EPIT are gained through the analysis on the efficiency of EPIT under a single-queue scenario. It is important to note that, even though this is a single-queue scenario, to the best of the author's knowledge, there has not been this type of analysis on efficiency in such scenario with LRD input. This is another contribution of this thesis. The analysis shows that, with a constant computation burden, the normalized variance of EPIT does not increase exponentially as the probability of overflow approaches zero asymptotically. |
