Stochastic post-processing for topology design and capacity assignment on a multi-class network with guarantees on robustness and Quality Of Service.
Wallach, Emmanuelle Janet
Master of Science
Date of Defense:
M Jeya Chandra, Thesis Advisor/Co-Advisor
stochastic optimization; telecommunications network design; Qual
This thesis considers a domain in the Internet and proposes a strategic problem of designing its network. In particular, an optimization problem is formulated and solved for planning the capacities of the links of the multi-class network so that (a) the network is robust, and (b) quality of service (QoS) can be guaranteed for traffic flowing across the network. The complexity of the optimization problem is that the QoS constraint, in terms of the decision variables, can neither be expressed in closed-form nor as a convex function. Therefore standard optimization techniques cannot be used. A two-stage heuristic is developed, which
first solves a routing problem, ignoring the QoS constraints and then uses the QoS constraints to determine the capacity of the
various arcs. Several examples show that the heuristic performs remarkably well, both in terms of speed and solution quality.