SAP-C02 Practice Question 57: Fleet Rightsizing Evidence
AWS Certified Solutions Architect – Professional (SAP-C02) · Test 6 · question 7 of 10
A finance business partner has asked the cloud team to cut compute spend across 90 AWS accounts without a performance regression. The estate includes thousands of EC2 instances, several hundred Auto Scaling groups, a large number of EBS volumes, and about 4,000 Lambda functions. Nobody has reviewed instance sizing since the workloads were migrated three years ago.
The team wants recommendations derived from each resource's actual utilization rather than from guesswork, wants idle resources identified as well as oversized ones, and wants to see the findings for all 90 accounts in one place rather than opening each account in turn. Installing and maintaining a third-party agent on every instance has been rejected.
What should the solutions architect recommend as the first step?
Show answer and explanation
Answer:
Compute Optimizer analyses a resource's configuration and its CloudWatch utilization metrics to produce rightsizing recommendations and to identify idle resources, reporting whether each resource is optimal and what to change. That is the utilization-derived analysis the team asked for, and "identify idle resources" is called out explicitly. The resource types match the estate: EC2 instances, EC2 Auto Scaling groups, EBS volumes, and Lambda functions are all supported. Opting in from the management account of an organization is what produces the single view; AWS states you can view findings and recommendations across multiple accounts if you opt in the management account, and the service reads metrics that already exist in CloudWatch, so no agent is deployed. It also shows projected utilization for each recommendation, which is how the team avoids the performance regression Finance is worried about.
Option A ranks resources by spend, not by waste. An expensive instance running at 90 percent utilization is correctly sized, and a cheap instance running at 2 percent is the real saving; cost alone cannot distinguish them.
Option C is the agent rollout the team rejected, and it is premature. Memory data can improve EC2 recommendations later (Compute Optimizer can even ingest external memory metrics), but it is not needed to start.
Option D produces alerts, not recommendations. A CPU threshold says an instance is quiet without saying which instance type it should become, it ignores memory, network, and I/O, and it leaves 90 accounts' worth of judgement calls with individual teams.
Reference: What is AWS Compute Optimizer?