A Reinforcement Learning-Based Lightpath Establishment for Service Differentiation in All-Optical WDM Networks

Abstract

In this paper, we propose a lightpath establishment method based on reinforcement learning for providing the service differentiation in all-optical WDM networks. In our proposed method, the optimal policy for the lightpath establishment is derived with Q-learning. With the derived policy, each node decides whether a lightpath establishment request of each class should be accepted or not. This method can be available even if the number of wavelengths is large and there is no assumption about the lightpath establishment. We also discuss how the proposed method is utilized with Generalized Multi-Protocol Label Switching (GMPLS). In numerical examples, we investigate the impacts of learning parameters on the performance of the proposed method. Then, we show that our proposed method can provide the service differentiation for the lightpath blocking probability, while utilizing wavelengths effectively.

Self-learning Routing for Optical Networks

“…This gives us a possible way to overcome the difficulty of feature extraction for adaptive routing. In [2,8,10,13], the authors apply reinforcement learning to the routing of optical networks. These methods allow the RL-based agent to learn a routing policy during the interaction with the network environment.…”