Reward-based online routing and spectrum assignment in flex-grid optical networks

In this paper we propose an online, reward-based routing and spectrum assignment (RSA) method for multirate flex-grid optical networks. The method, denoted as MDP-RSA, is derived from a flex-grid optical link model based on the theory of Markov decision processes. Assuming that connections arrive randomly offering a reward to the system, the objective is to take the best decision on admission, routing and spectrum assignment for each incoming connection request, such that the rate of reward from the network is maximized. This is accomplished by only admitting connections that bring a positive reward gain to the network. For selected network scenarios, the reward loss (i.e. the ratio of the reward of rejected to offered connections) caused by the MDP-RSA is compared to that of an online scheme, which admits connections on the shortest path that fulfils the contiguity and continuity constraints. It is observed that the MDP-RSA yields a lower reward loss. Furthermore, it is shown how to provide grade of service control on individual connection classes. Compared to existing online RSA schemes, the MDP-RSA is much more flexible as the connection reward parameters can be tailored to different optimization objectives (e.g. maximization of either carried traffic or economic revenue).