Structural Network Metrics and Incident Generation

The monitoring of data networks to ensure their availability is an increasingly important task. The proliferation of scale and complexity renders this an ever more complex and challenging problem to solve. At the heart of the problem is the large volume of poor quality telemetry that systems produce. In this work, we investigate the role of network structure as a guide to filter out events from nodes that produce fewer service-impacting incidents. Analyzing a commercial dataset that includes both network topology and monitored incidents, we are able to identify that structural network metrics such as degree, entropy, and curvature, are a reliable indicator that a particular node will produce a higher rate of incidents. This holds out the prospect of being able to target monitoring at more critical nodes, improving the efficiency of network management systems.