Adaptive Routing and Spectrum Assignment in Flex-grid Optical Networks

Abstract:
We propose a novel online method for adaptive routing and spectrum assignment (RSA) in flex-grid optical networks which aims at maximizing a definable reward objective. The method, denoted as MDP-RSA, is based on the theory of Markov Decision Processes (MDP). For any incoming connection request, a policy iteration algorithm is executed to decide on the optimal lightpath to allocate. Such a decision is performed online based on the current state of the network (i.e. the number of carried connections) and on the offered traffic load. The optimization target is the maximization of the overall reward of the network. A special case of the reward formulation is the request acceptance probability (i.e. the inverse of the blocking probability). For this case, the performance of the MDP-RSA method is compared to that of a Least Loaded Path (LLP) RSA scheme. It is observed that for partial mesh network topologies with high traffic load both methods are comparable, whereas for highly or fully meshed topologies MDP-RSA outperforms LLP. The performance of the MDP-RSA scheme degrades in low load regions for partial mesh networks due to the approximation method used to solve the optimization problem. However, it should be noted that the MDP-RSA scheme is much more flexible than LLP RSA as it (unlike LLP) can be tailored to different reward objectives (e.g. tariffing schemes).