Computing diagnostic explanations of network faults from monitoring data

Abstract

A system and method for network fault diagnosis in a network having network elements is presented. The method comprises creating a network causality model, generating Boolean expressions from the network causality model, converting the Boolean expressions into SAT sets, receiving network monitoring results, correlating these monitoring results with the SAT sets, and enumerating all possible diagnostic explanations of potential faults, properly annotated. Creating a network causality model can comprise creating, for each network element, an element-specific causality model, stitching together all network elements using the element-specific causality models and a network topology, retrieving monitoring state and propagation information, and generating the network causality model using the stitched together network elements and the monitoring state and propagation information. Stitching together network elements can comprise adding causes and implies dependency between appropriate network elements and/or adding and connecting reachable and not-reachable states. The network causality model can comprise network element states.