Automatic root cause analysis of performance problems using auto-baselining on aggregated performance metrics

Abstract

Anomalous behavior in a distributed system is automatically detected. Metrics are gathered for transactions, subsystems and/or components of the subsystems. The metrics can identify response times, error counts and/or CPU loads, for instance. Baseline metrics and associated deviation ranges are automatically determined and can be periodically updated. Metrics from specific transactions are compared to the baseline metrics to determine if an anomaly has occurred. A drill down approach can be used so that metrics for a subsystem are not examined unless the metrics for an associated transaction indicate an anomaly. Further, metrics for a component, application which includes one or more components, or process which includes one or more applications, are not examined unless the metrics for an associated subsystem indicate an anomaly. Multiple subsystems can report the metrics to a central manager, which can correlate the metrics to transactions using transaction identifiers or other transaction context data.