Graphs don’t tell you what to do. Patterns do.
CloudWatch will happily draw a CPU graph for every instance you own. What it won’t do is tell you which of those 400 graphs deserve attention. CloudMonitor reads seven days of hourly datapoints for each EC2 instance, RDS database and Lambda function and turns the curve into a verdict you can act on.
The five patterns
- Idle — average CPU under 5%. Usually a forgotten test box, an over-provisioned standby or a workload that moved elsewhere. Stop it, downsize it or delete it.
- Underutilized — average under 20%. The classic rightsizing candidate: one size down in the same family often halves the cost with no user-visible change.
- Spiky — peaks above 50% that are more than 3.5 times the average. Often a batch window or a cron job; consider burstable instances, scheduling or moving the job to Spot.
- Sustained-high — average above 70%, or above 60% with peaks over 90%. This is a saturation risk, not a saving: scale up before it becomes an incident.
- Healthy — everything else.
Each pattern carries a score, and the fleet score is the average across resources. It is a blunt instrument on purpose: a single number your team can watch move as you work through rightsizing.
From pattern to decision
The metrics explorer shows average, minimum, maximum, p95 and latest values for every cached series, and a day-by-hour heatmap that makes schedules obvious. Pair it with the What-If Simulator to price a one-step-down rightsizing across the fleet, or with the cost optimizer to find the storage and network waste around those instances.
Coverage today
Utilization patterns are built from Amazon CloudWatch. Azure and Google Cloud resources are covered by the inventory and security checks, but their utilization metrics are not collected yet.