To understand our EC2 fleet costs and identify rightsizing opportunities. The goal was to analyze actual CPU utilization across our EKS worker nodes and recommend instance types that match real usage patterns without over-provisioning.
What makes this hard?
Cost Explorer shows you spend, but not utilization. CloudWatch shows CPU metrics, but doesn't connect them to cost impact. The EC2 console shows instance types, but not whether they're appropriate for your workload. Manual investigation requires:
- Query Cost Explorer to understand current EC2 spend by instance type
- Pull CloudWatch metrics for each instance individually
- Correlate utilization patterns with instance specifications
- Research alternative instance types and their pricing
- Calculate potential savings across different scenarios
- Manually connect: spend data → utilization metrics → instance specs → rightsizing recommendations
How did Resolve AI help?
With one query, Resolve AI analyzed Cost Explorer, CloudWatch metrics, and EC2 configuration to build an evidence-based rightsizing recommendation:
- Inventoried the fleet: 4 homogeneous instances running across 3 subnets
- Retrieved actual costs: $103.93/month for compute, $416/month at current 4-instance steady state
- Measured CPU utilization: 4-9% average across all instances, peak usage only 22-38% over 24 hours
- Identified usage pattern: Consistent low baseline with occasional spikes. Ideal for burstable instances
- Validated compatibility: Graviton architecture already in use, making migration low-risk
Resolve AI connected utilization evidence (4-9% CPU with 38% peaks) to cost data ($416/month current) to recommend t4g.medium instances—matching the burst-friendly workload pattern while reducing spend by $319/month annually.
