SAA-C03
Deep Dive
20% of the exam (D4) tests how to optimize costs. For compute, the key is choosing the right pricing model for the usage pattern: On-Demand (flexibility), Reserved/Savings Plans (steady load), Spot (interruption-tolerant).
Contents
EC2 has 5 pricing models. The exam presents a usage scenario and asks you to identify the most cost-efficient one.
On-Demand
Best for
New apps, unpredictable workloads, short dev/test cycles, occasional spikes
Don't use when
Steady 24/7 workloads — this is the most expensive option here
Reserved Instances (1 year)
Best for
Web servers, production databases with predictable steady load
Don't use when
Variable workloads or instances that might not run all year
Reserved Instances (3 years)
Best for
Critical infrastructure that will definitely run for 3 years. DBs, NAT Gateways.
Don't use when
If there is uncertainty about the 3-year need
Savings Plans
Best for
Organizations that want instance-type flexibility while keeping Reserved-level savings
Don't use when
If you need per-instance-specific savings (EC2 RIs can give a deeper discount)
Spot Instances
Best for
Batch jobs, ML training, data processing, rendering, CI/CD — INTERRUPTION-TOLERANT
Don't use when
Applications with persistent state, databases, apps that cannot restart
| Type | Applies to | Flexibility | Max discount |
|---|---|---|---|
| Compute Savings Plans | EC2 (any family/region), Lambda, Fargate | Maximum — any instance type, region, OS | Up to 66% |
| EC2 Instance Savings Plans | EC2 (specific family in one region) | Medium — only the committed family/region | Up to 72% |
| Standard Reserved Instances | EC2 (specific type, AZ, OS) | Low — fixed type, can be sold on the Marketplace | Up to 72% |
| Convertible Reserved Instances | EC2 (can change family/OS) | Medium — convertible to other families | Up to 54% |
AWS recommendation
AWS recommends Compute Savings Plans for most workloads: more flexible than Reserved Instances (automatically applies to EC2, Lambda, and Fargate), competitive discount, and it doesn't "lock" you into a specific family/region. Reserved Instances only make sense when you need a specific AZ (Capacity Reservation).
Spot Fleet — pool diversification
Spot Fleet lets you specify multiple instance types and AZs. AWS picks the pool with the best price/availability. If one type runs out, it launches from another pool automatically.
lowestPrice strategy
Maximizes savings — picks the cheapest pool at every moment.
diversified strategy
Spreads across multiple pools — better availability, fewer simultaneous interruptions.
Spot Instance interruptions
When AWS needs the capacity back, it gives a 2-minute warning. Your app must handle this with checkpoints or a stateless design.
Best practices:
• Save progress every few minutes (checkpoints to S3)
• Design idempotent tasks (able to restart)
• Use an ASG with an EC2 Spot + On-Demand mix
ASG with Spot mix (recommended)
ASG with MixedInstancesPolicy: e.g. 20% On-Demand (guaranteed baseline) + 80% Spot (maximum savings). If Spot is interrupted, the ASG temporarily launches On-Demand.
Rightsizing = matching instance type/size to actual usage. The most common mistake: provisioning m5.4xlarge instances for workloads that use 10% CPU — an m5.large is enough and 8x cheaper.
AWS Compute Optimizer
Analyzes 14 days of CloudWatch metrics and recommends the optimal instance type. Supports EC2, Auto Scaling Groups, EBS, and Lambda.
When NOT to rightsize
Occasional traffic spikes
A smaller instance can get saturated at the peak — use an ASG instead.
DBs with expected growth
Better to provision with headroom to avoid migrating frequently.
Legacy applications with licenses
Some licenses are billed per vCPU — verify before downsizing.
Lambda — cost optimization
Tune memory
Lambda allocates CPU proportional to memory. More memory = more CPU = can be cheaper if it reduces duration.
ARM/Graviton2
ARM Lambda functions are ~20% cheaper and 10% faster than x86. Migrate if the code is compatible.
Compute Savings Plans
Savings Plans automatically apply to Lambda — no extra configuration needed.
Reduce cold starts
Provisioned Concurrency has a fixed cost — calculate whether it is worth it vs. the UX cost of cold starts.
Fargate — cost optimization
Fargate Spot
Fargate Spot = Spot Instances for ECS/EKS containers. Up to 70% discount for interruption-tolerant tasks.
Right-size containers
Define vCPU and memory precisely. Don't request 4 vCPUs for a container that uses 0.5.
Compute Savings Plans
Also apply to Fargate. A $/hour commitment cuts costs by up to 50%.
ARM/Graviton2
ECS Fargate supports ARM — 20% savings for the same performance.
Did you get this topic?
Test what you just learned
A company has a fleet of 50 m5.4xlarge instances running log-analysis applications that process data in batches overnight (6 hours). The jobs can restart if interrupted. Which is the most cost-effective strategy?
Put what you learned to the test with original SAA-C03 exam questions.
Start SAA-C03 mock examUnlimited mock exams with the Pro plan
Log in to track your progress.