Distributed Network Tomography for Failure Localization

In today's complex and interconnected networks, efficiently and accurately identifying performance degradation and failures is essential for service management, application performance, and network reconnaissance. Network Tomography can infer the state of the internal elements of the network via end-to-end measurements from the periphery of the network. Existing solutions rely on a centralized approach where the measures taken by the monitors are processed and analyzed by a central unit, or they assume that monitors can share all the information gained by communicating through a different, independent and non-faulty network. In this paper, we present D2NeT (Distributed and Dynamic Network Tomography) an innovative approach to network monitoring that extends the principles of network tomography by enabling a real-time and decentralized monitoring solution. We propose a framework based on distributed interactions among monitoring nodes, incorporating an advanced Bayesian decision-making support system. We validate our solution by testing it on emulated networks showing that it outperforms other baseline and state-of-the-art solutions in static failure scenarios. We then considered D2NeT in a dynamic scenario with ongoing failures and restorations, demonstrating its exceptional ability to promptly detect sudden changes in the nodes' state.