Anomaly-Based Alertingin HCO is currently focused on high-cardinality performance metrics that exhibit clear cyclical patterns (daily or weekly cycles). According to theSolarWinds Platform Alerting Guide, the machine learning engine is optimized for metrics where "normal" behavior varies significantly based on time of day.
The two primary supported metrics for this feature are:
* Average CPU Load (A): CPU utilization is highly variable; a server might be idle at night but busy during business hours. Anomaly detection learns these patterns to prevent false positives during scheduled peak times.
* Percent Packet Loss (D): Network stability is a critical indicator of environmental health. By establishing a baseline for packet loss, the system can distinguish between a minor, expected "blip" in a high-traffic environment and a true anomaly that indicates a failing circuit or network congestion.
While metrics like "Disk Space Usage" (Option B) are critical, they are generally "linear" or "incremental" rather than cyclical; a disk filling up is a trend that is usually better handled by standard predictive or static threshold alerts. Similarly, while interface utilization is important, the initial release of anomaly-based features prioritizedNode-levelperformance metrics like CPU and Packet Loss to provide the most immediate value for identifying server and core network health deviations.