Detect Silent Failures with Network Sentinel

In this work, we propose ‘Network Sentinel’, a root cause analysis (RCA) model aiming at detecting silent failures in a network. Even nowadays AI comes to aid, CSPs still rely on the network management system (NMS) to remotely collect data or alarms from the network to perform fault detection or RCA. Consequently, the NMS itself can be the ‘Achilles heel’ of the network. A silent failure emerges when the NMS breaks down or disconnects so that we cannot determine the real state of the Network according to the deficient observed symptoms. Network Sentinel utilizes uncertainty levels of topologically neighboring nodes of that possible silently failed node to identify the failure. It is best embedded into a Digital Twin Network (DTN) that can predict and simulate traffic changes of the links of its physical twin.