SGER: Algorithm Design for Reconfiguration Problem in Optical Networks

点击翻译按钮获取中文摘要

The optical network is a promising high-speed backbone or transportation network that requiresan extremely caution in operations. The reconfiguration is one of the indispensable operations if avirtual topology no longer serves a traffic demand which is changed over time. We propose to develop an algorithm and a complete reconfiguration model for wavelengthroutedoptical networks which includes the reconfiguration process and the policy. The reconfigurationprocess provides the choices of the reconfiguration operations (e.g., add, delete or re-routelightpaths) to maintain the high performance of the network in any traffic demand volumes andpatterns while minimally disturbs the current virtual topology. The reconfiguration policy defineswhich choice to be selected that returns the optimal expected outcome based on the cost of operationand the performance reward. We have preliminary experimental results that show that selectingthe choice with the highest immediate outcome does not return the optimal expected outcome inthe long term. Although the reconfiguration problem is an NP-hard problem and is a trade-off between performanceand number of changes in virtual topology, we can still find the solutions using a multiobjectiveevolutionary algorithm with the concept of Pareto Optimal. The algorithm provides a setof solutions in the Pareto front while the policy picks one of solutions in the Pareto front that gainsthe optimal expected outcome. The policy depends on the pattern of traffic. If the future patternof traffic can be predicted or estimated, the Markov Decision Process (MDP) can define the policy.However, the status of network or the MDP's state which effects the Pareto front needs an intensivestudy. Our goal is to develop the theory and algorithms in accordance with realistic traffic demand andprotocol. The model will be the centralized control over the network, which will not be designedonly for the specific performance objective or traffic type but be applicable for various objectivesand traffic types. The development of practical algorithms for solving these problems represent animportant step in allowing reasonable scale implementation of optical networks. Intellectual Merit of the proposed research rests on the integration between the reconfigurationprocess and the policy based on the Pareto optimal concept and MDP that has not yet happened inthe Optical network reconfiguration field. Reconfiguration problems are among the most difficultin the areas of optical networks since such problems encompass the construction of algorithmsinvolving the extremely high-speed communication over an optical fiber. The consequence problemis how or when to perform the reconfiguration process. From the practical application viewpoint,we want to advance our model to the practical optical networks. Broader Impacts of the proposed research is expected to be strong since we are proposingto integrate two disciplines, the evolutionary computing and the stochastic process for the recon-figuration in the optical networks. We expect that the work will have broad impact in industrialpractices on network provisioning, protection and restoration areas.