This disclosure describes techniques for monitoring computer system metrics that exhibit cyclical (periodic) behavior. Per techniques of this disclosure, function-based thresholds are utilized for metric monitoring and generation of system alerts. The function-based thresholds are time-varying and are aligned with the troughs and peaks of the underlying metric being monitored. Metric measurements are obtained for a particular period of time. A function such as a Discrete Cosine Transform (DCT) is fitted to the obtained metric data. A threshold is set for the metric based on the fitted function at a fixed offset from the function derived value. Function-based thresholding can enable monitoring metrics that exhibit cyclical properties with greater sensitivity when compared to fixed thresholding. Deviations of metric values can be accurately detected even at troughs in the metric data while also mitigating false positives at peaks in the metric data.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Turner, Joseph, "Alert Generation for Cyclically Varying Metrics", Technical Disclosure Commons, (November 15, 2021)