Zinsight: A visual and analytic environment for exploring large event traces

Abstract

Information in event traces from software systems can help developers with performance analysis, debugging and troubleshooting. However, the volume of data contained in these traces can make such tasks a challenge. In this paper we propose a new tool, Zinsight, to visualize event traces from complex systems. Our contribution is a novel combination of visualizations and pattern extraction techniques, enabling user exploration, analysis and understanding of traces containing millions of events. Three complimentary views help the user answer different questions. First, the Event Flow view shows the trace in its entirety or in detail. The user sees visual patterns representing phases of processing and the relative order of events. Second, the Event Statistics view quantifies events, and presents distributions and averages enabling the user to identify outlier behavior. Third, the Sequence Context view extracts patterns of interest from the trace and represents them along with frequency and performance data in succinct execution flow diagrams. The user can navigate from patterns to their constituent instance sequences and even back to individual events in the other views. Questions can be answered and hypotheses tested using the most natural view for the task. Copyright 2010 ACM.