Analytical frameworks to evaluate performance in traffic engineered networks
This dissertation presents two block work. One is Path computation elements (PCE's) implemen- tation and modeling. The second is the blocking probability study for different classes in modern broadband multi-service networks.
Path computation elements (PCE's) are used to compute end-to-end paths across multiple areas. Multiple PCE's may be dedicated to each area to provide sufficient path computation capacity and redundancy. An open problem is which PCE should be chosen to send the path computation request to, that may be a non trivial problem if PCE's have uneven processing capacities.
This dissertation proposes a product form queueing model to estimate the latencies in path com- putation while accounting for the arrival rate of path computation requests. The model is used to find the PCE selection policy to minimize the average expected latencies in path computation. The model is validated against two simulation benchmarks obtained using OPNET, i.e., a network of queues and the multi protocol label switching with traffic engineering (MPLS-TE) network running the PCE communication protocol (PCEP).
The study shows that the use of product form yields approximations that are up to 15% at practical offered loads. Moreover, the PCE selection policy derived under the product form assumption is showed to be effective in minimizing the overall expected latencies in path computation.
Quality of service issues are becoming more and more important in modern broadband multi- service networks. The blocking probability of each service is an interesting and important fact for the analysis of the QoS.
This dissertation gives an exact model solution of the blocking probability calculation for any classes in network. A recursive analytical calculation method is presented for this analytical model.Three approximation methods are provided to get blocking probability for large link capacity and large service classes.
The study shows that the three approximation method have advantages and disadvantages under different scenarios. These approximation methods provide useful and solid foundation for the next step network blocking probability research.