Towards discovery of event correlation rules

Abstract

For large installations, event management is critical to ensuring service quality by responding rapidly to exceptional situations. The key to this is having experts encode their knowledge (e.g., in rules, state machines, codebooks) about the relationship between event patterns and actions to take. Unfortunately, doing so is time-consuming and knowledge-intensive. We propose reducing this burden by using offline decision support consisting of visualizing and mining event histories to discover patterns in event data. Our experience with a wide variety of production data has identified several patterns of interest such as, event bursts and partial periodicities. Herein, we use production data to illustrate how to visualize and mine event patterns, and we describe a tool we have developed to aid in pattern discovery.