Design Cost-Optimized Compute Solutions for SAA-C03
Cover right sizing, Spot, Reserved Instances, Savings Plans, Lambda, Fargate, and scaling trade-offs for SAA-C03 compute-cost scenarios.
Compute-cost questions on SAA-C03 are about avoiding over-provisioning without breaking the workload. AWS wants you to choose the right combination of purchasing model, runtime model, instance family, and elasticity based on how predictable and interruption-tolerant the workload actually is.
What AWS is explicitly testing
The current exam guide points to cost-management tools, AWS purchasing options, distributed and hybrid compute strategies, instance families and sizes, compute-utilization optimization, and scaling strategies such as Auto Scaling and hibernation.
Cost-aware compute chooser
| Requirement | Strongest first fit | Why |
|---|---|---|
| Steady long-running usage | Savings Plans or Reserved-style commitment model | Reduces cost for predictable usage |
| Interruption-tolerant batch or flexible workloads | Spot Instances | Lowest-cost compute when interruption is acceptable |
| Bursty event-driven execution | Lambda | Pay-per-use can beat always-on servers |
| Container runtime without idle EC2 fleet management | Fargate | Can lower operational overhead and overprovisioning risk |
Purchasing model chooser
| Pattern | Usually strongest fit | What SAA-C03 is really testing |
|---|---|---|
| Predictable baseline production load | Savings Plans or committed usage strategy | Whether you notice stable usage is being paid for inefficiently |
| Elastic batch or background work | Spot | Whether interruption tolerance is part of the architecture |
| Spiky short-lived event processing | Lambda | Whether idle capacity should disappear entirely |
| Mixed production fleet with steady base and burst capacity | On-Demand base plus Spot or scaling policy | Whether one pricing model alone is too simplistic |
The cost question behind the instance choice
Ask three things:
- Is the workload steady or bursty?
- Can it survive interruption or restart?
- Is the team paying mostly for compute itself, or for idle headroom and management overhead?
That usually gets you to Spot, Savings Plans, Lambda, Fargate, or standard EC2 sizing more cleanly than memorizing product marketing.
Example: mix On-Demand baseline with Spot burst capacity
1Resources:
2 WebAsg:
3 Type: AWS::AutoScaling::AutoScalingGroup
4 Properties:
5 MinSize: '2'
6 MaxSize: '10'
7 DesiredCapacity: '2'
8 VPCZoneIdentifier:
9 - subnet-a
10 - subnet-b
11 MixedInstancesPolicy:
12 LaunchTemplate:
13 LaunchTemplateSpecification:
14 LaunchTemplateId: lt-1234567890abcdef0
15 Version: '1'
16 Overrides:
17 - InstanceType: m7g.large
18 - InstanceType: m6i.large
19 InstancesDistribution:
20 OnDemandBaseCapacity: 2
21 OnDemandPercentageAboveBaseCapacity: 20
22 SpotAllocationStrategy: price-capacity-optimized
What to notice:
- the architecture protects a small steady baseline with On-Demand capacity
- burst scale can shift toward Spot instead of paying full On-Demand cost everywhere
- SAA-C03 often prefers cost reduction that respects workload tolerance instead of blindly minimizing price
Failure patterns worth recognizing
| Symptom | Strongest first check | Why |
|---|---|---|
| Cost is high even when CPU stays low for long periods | Right sizing and scaling policy | Idle headroom may be the real issue |
| A production system cannot tolerate interruption | Spot suitability | The cheapest compute is still wrong if the workload cannot survive interruption |
| Bursty traffic keeps paying for always-on servers | Runtime model fit | Lambda or Fargate may be cheaper than oversized EC2 |
| Discounts exist but are not being captured | Usage predictability | Stable baseline usage often points to commitment options |
Common traps
- using On-Demand for predictably steady workloads forever
- choosing Spot for workloads that cannot tolerate interruption
- choosing EC2 because it feels flexible even when Lambda or Fargate fits the pattern better
- right-sizing the instance but ignoring the scaling policy that keeps it oversized most of the time
Quiz
Continue with 4.3 Database Solutions to apply the same cost lens to database engine, type, capacity, and retention choices.