Anomaly Detection and Prediction

Static Thresholds quite often do not work well in today's application environments. You either miss important signals or have to deal with spurious alerts. Managing such thresholds in an environment with thousands of Microservices each with their own computing stack is almost impossible. While thresholds apply to metrics and are challenging, so are filters applied to log messages. In fact they may just be a lot more tricky.

Both thresholds on metrics and filters on logs only deal with known failure modes and are applied to a small set of "important" metrics and logs. Which is why in spite of their wide spread presence, we often get surprised by failures and degraded behavior.

Intelligeni does away with the need to set such static thresholds and filters. Instead Intelligeni uses machine learning models to detect anomalous behavior in systems. Using both uni-variate and multi-variate models combined with our algorithms that process log messages and extract Quality of Service Impact semantics Intelligeni learns what are normal and abnormal behaviors of systems. Further you don’t need to restrict this to a few metrics and logs but instead analyze all data. Anomalies detected by Intelligeni become alerts that are then acted upon by Operations Engineers or automated Bots.

Intelligeni also applies predictive models to identify potential anomalies in metric data or event stream even before they happen.