What is rightsizing?

Why it matters

Rightsizing compares a resource's configuration with how much of it a workload uses, such as CPU, memory, network and disk activity. It then recommends a smaller, larger or different type, and may flag idle resources to stop or delete. AWS Compute Optimizer, for example, analyses 14 days of utilisation metrics by default to produce its recommendations.

A smaller instance has fewer vCPUs and less memory, so it draws less electricity and takes a smaller share of its server's manufacturing emissions. Rightsizing can therefore cut carbon as well as cost. Showing both savings side by side gives engineers a second reason to act on recommendations that cost alone has not pushed through.

An example

A rightsizing tool suggests moving an over-provisioned instance to a smaller type, and terminating another that sits idle. For the first, the saving is the gap between the two instance types' per-hour footprints over the hours the workload runs. For the termination, the idle instance's whole footprint is saved.

How Greenpixie measures it

Greenpixie's docs include a guide to adding energy, carbon and water to Cloudability rightsizing recommendations by joining the Greenpixie rate card on instance type. Each rate card row gives the footprint of running one instance for one hour in a region, with that region's factors already applied:

• electricity per instance-hour, in kWh
• carbon per instance-hour, in tonnes CO2e
• water per instance-hour, in litres

Multiplying each figure by hours running gives the current and recommended footprints, and the difference is the saving. Recommendations that span regions need a rate card per region, because power usage effectiveness, grid and water factors change from region to region.