Routing with Graph Convolutional Networks and Multi-Agent Deep Reinforcement Learning

The computer networking community has been steadily increasing investigations into machine learning to help solve tasks such as routing, traffic prediction, and resource management. In particular, due to the recent successes in other applications, Reinforcement Learning (RL) has seen steady growth in network management and, more recently, in routing. However, changes in the network topology prevent RL-based routing approaches from being employed in real environments due to the need for retraining. In this paper, we approach routing as an RL problem with two novel twists: minimizing flow set collisions and dealing with routing in dynamic network conditions without retraining. We compare this approach to other routing protocols, including multi-agent learning, to various Quality-of-Service metrics, and we report our lesson learned.