On the Design of Dynamic Recon guration Policies for Broadcast WDM Networks

We study the issues arising when considering the problem of recon guring broadcast optical networks in response to changes in the tra c patterns. Although the ability to dynamically optimize the network under changing tra c conditions has been recognized as one of the key features of multiwavelength optical networks, this is the rst indepth study of the tradeo s involved in carrying out the recon guration process. We rst identify the degree of load balancing and the number of retunings as two important, albeit con icting, objectives in the design of recon guration policies. We then formulate the problem as a Markovian Decision Process and we develop a systematic and exible framework in which to view and contrast recon guration policies. We also apply results from Markov Decision Process theory to obtain optimal recon guration policies even for networks of large size. The advantages of optimal policies over a class of threshold-based policies are also illustrated through numerical results.