Reinforcement Learning applied to the Routing and Spectrum Assignment in Elastic Optical Networks

Elastic Optical Networks (EON) has emerged as technology in optical networks whose architecture can respond to the growing need for elasticity in allocating optical network resources. EON imposes physical layer constraints on traffics. In this context, developing efficient Routing and Spectrum Assignment (RSA) algorithms for dynamic traffic is critical to EON's success. Strategies based on Reinforcement Learning (RL) emerge as a valid alternative to dynamic RSA, due to its ability to adapt to changes in the state of the network by learning process. In this work, we propose RL algorithms based on Q-learning to solve the dynamic RSA problem. First, we study the performance of routing algorithms that evaluate the learning Q values for a set of pre-calculated disjoint routes by selecting a route that corresponds to a minimum blocking rate. Similarly, we studied Q-learning algorithms for spectrum assignment, which selects a block of the spectrum from a pre-routed path to minimize the blocking rate. Numerical simulations on different dynamic traffic loads show that the performance of the Q-learning-based proposals is promising in obtaining a better blocking rate than the state-of-the-art heuristic approaches.