Kubernetes Rightsizing Savings Calculator

Estimate the monthly cloud spend you can reclaim by right-sizing Kubernetes pod CPU and memory requests against real utilisation.

Pods targeted for rightsizing in the workload
Current CPU request in millicores
Measured average or P95 CPU usage for the pods
Blended infrastructure cost per provisioned vCPU-month
Optional: headroom applied on top of observed CPU usage (defaults to 10%)
Optional: RAM request per pod (defaults to 0 GB if blank)
Optional: RAM usage per pod (defaults to 0 GB if blank)
Optional: blended memory cost per GB-month (defaults to $0 if blank)

Operational planning aid only. Validate recommendations with load testing and platform SLOs before applying new limits in production.

Examples

  • 40 pods requesting 500 mCPU, using 220 mCPU, $120 per vCPU-month, 10% buffer ⇒ Estimated monthly savings: $1,238.40 (CPU: $1,238.40, Memory: $0.00). Capacity freed: 10.32 vCPU and 0.00 GB.
  • 18 pods requesting 24 GB, using 14 GB, CPU already right-sized, $6 per GB-month, 15% buffer ⇒ Estimated monthly savings: $853.20 (CPU: $0.00, Memory: $853.20). Capacity freed: 0.00 vCPU and 142.20 GB.

FAQ

Which utilisation metric should I feed into the calculator?

Use a high percentile such as P95 from recent weeks so the buffer covers typical bursts. The default 10% buffer adds guardrail headroom if left blank.

How do I include memory savings?

Populate the optional memory fields with request and observed usage. Leaving them blank assumes no memory savings and calculates CPU impact only.

Can I combine spot nodes or autoscaling impacts?

Yes. Reflect the blended compute price in the cost fields, or run separate scenarios for on-demand and spot pools to compare savings.

Why might savings be zero?

If usage plus buffer already matches or exceeds the requests, the model assumes no safe reduction. Tighten workloads or adjust the buffer to explore other scenarios.

Additional Information

  • Result unit: USD (estimated monthly cloud savings).
  • CPU buffer defaults to 10% to cover routine spikes when no custom value is supplied.
  • Memory savings require both request and observed usage values; the calculator treats blanks as zero to stay conservative.
  • Convert millicores to vCPU by dividing by 1,000; the component does this automatically in the output.

Related calculators