Exams / SAP-C03 / Study guide / Cost-optimized compute and rightsizing (Graviton, Compute Optimizer)
Cost-optimized compute and rightsizing (Graviton, Compute Optimizer): SAP-C03 study notes and practice questions
What to know about Cost-optimized compute and rightsizing (Graviton, Compute Optimizer) for the SAP-C03 exam, from the Cost-Optimized Architecture Design domain (15% of the exam guide), then 1 free question to test it. Part of the SAP-C03 study guide.
SAP-C03 asks you to redesign a workload so it costs less while still meeting its performance goals. The scenarios mention overprovisioned instances, low CPU use, idle environments at night, or a workflow whose bill grew with traffic.
Choosing the compute model
AWS Graviton (Arm) instances give better price performance than comparable x86 instances for most Linux workloads that can be recompiled or already run on Arm runtimes (Java, Python, Node.js, .NET, containers with multi-arch images). Lambda functions and Fargate tasks can also run on arm64 at a lower price per unit of duration.
Lambda is billed per request and per GB-second. More memory also means more CPU, so a function can finish faster and cost the same or less; tune memory with Compute Optimizer recommendations or by testing.
Containers on Fargate remove idle node capacity; ECS or EKS on EC2 is cheaper when you can keep nodes densely packed.
Step Functions Standard bills per state transition and suits long-running (up to one year), auditable, exactly-once workflows. Express bills per request and duration, runs up to five minutes and is at-least-once, so it is far cheaper for high-volume, short, idempotent workflows such as event processing. See serverless workflows.
Rightsizing tools
AWS Compute Optimizer analyzes utilization and recommends sizes for EC2 instances, Auto Scaling groups, EBS volumes, Lambda functions, ECS services on Fargate, RDS and Aurora databases and commercial software licenses. It flags under- and over-provisioned resources and shows Graviton options. Enhanced infrastructure metrics (paid) extend the lookback period, and memory metrics need the CloudWatch agent.
Cost Optimization Hub gathers recommendations (rightsizing, idle resources, Graviton migration, Savings Plans and Reserved Instances) across accounts and Regions in one place and estimates savings net of existing discounts.
Enable both at the organization level with a delegated administrator so one team can see every account.
Scaling on metrics
Need
Use
Keep a metric near a target (CPU, requests per target, queue depth per task)
Target tracking
Known daily or weekly pattern
Scheduled scaling, or predictive scaling for EC2 Auto Scaling
Dev and test idle at night and weekends
Scheduled scaling to zero, or Instance Scheduler-style automation
Containers that tolerate interruption
Fargate Spot or Spot capacity providers
Queue-based workers
Scale on backlog per task or instance, not on CPU
Scaling in removes waste only if the application is stateless and instances or tasks can be replaced safely.
How to choose
Scenario wording
Pick
“CPU averages 10% on large instances”
Compute Optimizer recommendation, downsize, then buy commitments
“Linux microservices, cut compute cost with little code change”
Move to Graviton with multi-arch images
“Millions of short workflow executions per day”
Step Functions Express
“Workflow waits days for a human approval”
Step Functions Standard (or Lambda durable functions), not Express
“Recommendations across 200 accounts in one view”
Cost Optimization Hub with organization access
“Test environments idle outside office hours”
Scheduled scaling or stop schedules
Common traps
Moving to larger instances for headroom instead of scaling out on a target tracking policy.
Using Express workflows for a run that lasts longer than five minutes or must run exactly once.
Trusting CPU-only rightsizing for a memory-bound workload without the CloudWatch agent’s memory metrics.
Graviton for a workload tied to x86-only binaries or licensed software.
Buying Savings Plans before rightsizing locks in the waste.
A streaming site runs its video-encoding workers on a fleet of general-purpose x86 instances. The fleet runs at a low average CPU load, and AWS Compute Optimizer has recommended a smaller and cheaper instance option. The fleet has about 300 instances.
The build pipeline already produces container images for both x86 and arm64. Each encoding job runs for several hours and cannot checkpoint, so an interrupted job must start again, and the nightly window has no spare time.
The team wants the cost reduction without changing the encoding output.
Which action meets these requirements?
Choose one answer.
AA. Keep the x86 instances and buy more Savings Plans for them, so the same hourly spend covers a larger fleet at a lower rate.
BB. Move the encoding workers to Spot Instances across every Availability Zone, using a capacity-optimized allocation strategy and several instance types to reduce interruptions.
CC. Move the workers to the largest instance type in the family, so each job finishes sooner and the fleet can be reduced to a few machines.
DD. Review the Compute Optimizer recommendations, test the encoding jobs on Graviton instance types with the arm64 images, and move the fleet once the output matches.
Show the answer and explanation
Answer: D
A. Wrong: Savings Plans reduce the rate on the instances already in use but do not reduce how many instances are needed. The fleet is oversized, and that waste continues.
B. Wrong: Spot lowers the hourly rate, but an interruption restarts a multi-hour job that cannot checkpoint, and the nightly window has no room for reruns. It also leaves in place the oversizing that Compute Optimizer found.
C. Wrong: Larger instances cost more per hour, and a few oversized machines do not fix the low utilization. The change increases spend rather than reducing it.
D. Right: Correct. Graviton instances give better price performance for many workloads, and Compute Optimizer identifies the right size from observed load. Testing against the existing output confirms the encoding results before the move.
Explanation
Compute Optimizer's sizing advice combined with Graviton, tested against the arm64 images, lowers the cost without changing the output. Buying more Savings Plans keeps the waste, and bigger or Spot instances do not fix the oversizing.
Want more on Cost-optimized compute and rightsizing (Graviton, Compute Optimizer)? The 5 full-length practice sets have 5 more Cost-optimized compute and rightsizing (Graviton, Compute Optimizer) questions, 375 in all, none of them in the free exam. One payment of $14.99, no subscription. See the practice sets →
Study Cost-optimized compute and rightsizing (Graviton, Compute Optimizer)