Large scale symmetric network malfunction detection

Abstract

Network malfunction detection is expected to be fast while ensuring accuracy to reduce its impact and cost. Existing malfunction detection (MD) approaches are often unable to achieve both simultaneously for large scale networks. A key factor that governs the quality of a MD system is to distinguish the malfunction ones with the normal ones. Considering many networks are designed to be symmetric and malfunctions are usually only a small portion, this paper propose A4 - an automatic MD system which combines node embedding based on structural similarity in graphs (Graph-Wave which scales linearly) with density-based spatial clustering (DBSCAN) to distinguish malfunctions as the noises efficiently for original symmetric networks.