Dynamic reconfiguration of the logical topology of a network is generally used to adjust the shape of the network to adhere to changes in traffic. In this thesis the logical topology reconfiguration problem for a WDM network is combined with the multi-path routing of traffic throughout the network. Changes to the logical topology are made using branch exchanges to minimize the disruption to the network traffic. A Multi-timescale Markov Decision Process (MMDP) model is used to capture the different frequencies of changes to the network topology (slow timescale) and the routing of traffic (fast timescale). Heuristic policies are developed for the slow and fast time scales and rollout is used to improve the performance of the heuristic. To measure the performance of the heuristic and rollout policies, the Model Reference Adaptive Search (MRAS) method is used to find reference topologies that are close to optimal for a given static traffic demand matrix. To implement the MRAS method a random graph generation algorithm is needed. To address this need, a very efficient random graph generation algorithm is developed that competes with existing algorithms in the literature.