Blend CPU, memory and storage utilisation into one figure, then size the fleet the observed peak actually needs.
Utilisation is blended with CPU carrying the most weight because compute pricing follows vCPU, while storage is included at a lower weight since it is cheap but frequently over-provisioned. Right-sizing converts the observed peak into an absolute demand and divides by the utilisation you are prepared to run at, so the answer keeps deliberate headroom rather than fitting the peak exactly. Average utilisation below 20% is normal and is the single largest pool of recoverable cloud spend, but resizing on averages instead of percentiles causes outages. Cost does not always scale linearly with vCPU across instance families, so verify the projected saving against your provider's pricing before acting.
Cloud Resource Utilization
blended utilisation = CPU × 0.45 + memory × 0.35 + storage × 0.20; required vCPU = provisioned vCPU × peak CPU ÷ target utilisation; reclaimable spend = spend × reclaimed vCPU ÷ provisioned vCPU.
blended utilisation = CPU × 0.45 + memory × 0.35 + storage × 0.20; required vCPU = provisioned vCPU × peak CPU ÷ target utilisation; reclaimable spend = spend × reclaimed vCPU ÷ provisioned vCPU. Utilisation is blended with CPU carrying the most weight because compute pricing follows vCPU, while storage is included at a lower weight since it is cheap but frequently over-provisioned. Right-sizing converts the observed peak into an absolute demand and divides by the utilisation you are prepared to run at, so the answer keeps deliberate headroom rather than fitting the peak exactly.
Average utilisation below 20% is normal and is the single largest pool of recoverable cloud spend, but resizing on averages instead of percentiles causes outages. Cost does not always scale linearly with vCPU across instance families, so verify the projected saving against your provider's pricing before acting.
This calculator takes 8 inputs: vCPUs provisioned, Average CPU utilisation, 95th-percentile CPU utilisation, Average memory utilisation, Storage provisioned, Storage actually used, Monthly spend on this fleet, Target peak utilisation. The pre-filled defaults are a realistic starting point — replace them with figures from your own environment for a result you can act on.
The peak you measured is not the peak you will see. Sizing to a target such as 65% leaves room for growth, for traffic spikes and for the load that shifts onto the survivors when one instance fails — which is exactly when you least want to be at capacity.