Network Tomography and Reinforcement Learning for Efficient Routing

Network tomography is a powerful tool to infer the internal state of a network using end-to-end metrics observed by a few nodes at the edge of the network. However, previous research in network tomography has not focused on the objectives and challenges of specific network management applications, resulting in unsatisfactory performance. This paper proposes Subito (Shortest Path Routing with Multi-armed Bandits and Network Tomography) to address the needs and challenges of shortest path routing, a cornerstone of many network management tasks in wired and wireless networks. Subito combines network tomography with reinforcement learning to find an efficient routing strategy. Experiments on synthetic networks show that Subito provides performance improvements up to three times compared to two state-of-the-art approaches.