Compute Optimizer
AWS Compute Optimizer analyses the configuration and utilisation of resources in an account and recommends changes: a smaller instance type where one is over-provisioned, a larger one where performance is constrained, and idle resources that should be stopped or deleted.
It is a cost and governance tool that happens to have architectural consequences — the canonical page for it in this handbook is AWS Compute Optimizer. This page covers what an architect needs from it when sizing a design.
What it does
Section titled “What it does”Three things, in order:
- Identifies over-provisioning, which is where most cloud waste sits.
- Identifies under-provisioning, which is where latency and error-rate problems that look like application bugs often originate.
- Identifies idle resources that can be stopped or removed outright.
It runs per account or, from an organisation’s management account, across every member account at once — which is the only practical way to use it at any scale.
How it works
Section titled “How it works”After you opt in, Compute Optimizer reads each resource’s configuration and its CloudWatch metrics for the preceding 14 days, models the utilisation pattern, and produces recommendations with projected utilisation for each option so you can compare price against performance.
Two settings materially change the output:
- Enhanced infrastructure metrics (a paid feature) extends the lookback window from 14 days to 93, which matters for anything with a monthly or quarterly cycle. A 14-day window over a quiet fortnight will confidently recommend downsizing an instance that is correctly sized for month-end.
- External metrics ingestion lets Compute Optimizer read EC2 memory utilisation from Datadog, Dynatrace and similar, which removes the largest blind spot in its EC2 recommendations — the hypervisor cannot see memory use, so without either the CloudWatch agent or an external source, memory is simply absent from the model.
Supported resources
Section titled “Supported resources”Compute Optimizer covers considerably more than EC2:
- Amazon EC2 instances
- Amazon EC2 Auto Scaling groups
- Amazon EBS volumes
- AWS Lambda functions
- Amazon ECS services on AWS Fargate
- Commercial software licenses
- Amazon Aurora and Amazon RDS databases
- NAT gateways
- Amazon DynamoDB
- Amazon ElastiCache
- Amazon MemoryDB
- Amazon DocumentDB
- Amazon WorkSpaces
- Amazon SageMaker
Older material commonly lists only the first five. If a rightsizing exercise has been scoped from one of those lists, the database, cache and gateway recommendations — often the largest single line items — will have been left out.
Using it in a design review
Section titled “Using it in a design review”Recommendations are inputs, not instructions. Two habits make them useful:
- Check the lookback window against the workload’s actual cycle before acting on anything.
- Treat an under-provisioning finding as a performance investigation rather than a purchase decision; the fix is sometimes the instance type and sometimes the query.