QoS constraint reconfiguration policy based on DEB-cost model in optical wavelength routed networks

Abstract

In backbone optical networks, the changing traffic patterns may devalue the optimization of the static virtual topology design. Rapid and flexible Virtual topology Reconfiguration is a highly desired feature to build the traffic adaptive self healing next-generation optical network. The reconfiguration phase will interfere with existing traffic and disrupt network performance, causing a degradation of the quality of service perceived by the users. This reconfiguration problem is intractable due to its computation complexity and online nature. The issues that arise in reconfiguring have been studied in two ways which are regarding determining when to reconfigure the network and describing a class of strategies that determine how to retune the optical transceivers. Gencata and Mukherjee suggested the lightpath load``s threshold values for establishing and deleting a lightpath by using MILP to get those thresholds. However this optimization problem is known as NP-complete problem. Linear formulation problem by LP(Linear Programming) package is optimized and exactly correct but its computation time is not tractable and takes so much time to apply dynamic traffic pattern change in real networks. As well virtual topology reconfiguration policy can be obtained by using Markov Decision Process(MDP) to get optimal reconfiguration policy for the trace of the virtual topology. However, a numerical solution using MDP faces the ``curse of dimensionality`` for large state spaces. While the small state space by dividing the states roughly, affects the optimality. In addition, MDP solution techniques also require explicit specification of the transition probabilities of the continuous time chain at observed times. These drawbacks make this approach unpractical. At first, we define the amount of loaded traffic on the virtual topology as a network state then we derive the threshold value for deciding the reconfiguration instants. By using this threshold we suggest simple t...