Use anomaly detection for exception and alert thresholds

Db2® Query Monitor can be configured to report on exceptions and alerts using either manually specified thresholds or thresholds that have been established using anomaly detection on a workload.

  • Specifying thresholds manually - Db2 Query Monitor administrators can manually determine and specify the appropriate thresholds for alerts and exceptions. This approach uses a fixed threshold value.

  • Specifying thresholds using anomaly detection - Db2 Query Monitor administrators can configure anomaly detection for an SQL workload. This enables Db2 Query Monitor to inform users with alerts or exceptions whenever the performance of SQL deviates from recent historical averages. When configuring anomaly detection, administrators can specify how sensitive the anomaly detection should be to deviations. The thresholds determined by anomaly detection change as a function of how a system operates over time.

Note: In some situations (for example, when monitoring performance against an SLA), a fixed threshold might be more appropriate than a dynamic threshold. In other situations, a variable threshold provides the advantage by telling you when the system is exhibiting behavior that is very different than how it was performing last week (even if it is still within your SLA limit).

Benefits

Anomaly detection and reporting enables Db2 Query Monitor to learn what is normal over time, providing the following benefits:

  • Db2 Query Monitor administrators are no longer required to approximate the good threshold values. Instead, adaptive thresholds can be used with alerts and exceptions, allowing machine learning to determine what is normal over time.
  • Adaptive thresholds enable Db2 Query Monitor to use machine learning to calculate what is normal for Elapsed Time, CPU, and GETPAGES.
  • An alert or exception is produced only when an SQL statement execution is a user-specified variance from normal (standard deviation).
  • Enhanced per-interval collection and reporting to add mean and variance.
  • Enriched data for conventional summary views.
  • Mean and variance from server startup and for each interval is reported.
  • Alerts and exceptions become more meaningful because they are only triggered when an execution is user-specified variance from normal and multiple successive occurrences result in increased significance.
  • Performance data history provides more value.
  • Includes calculated mean over time and per interval.
  • Provide a richer environment for problem determination.