Design Cost-Optimized Architectures
Study cheat sheet Β· SAA-C03)
Generated May 13, 2026
Practice This Topic
Key Concepts
- Cost optimization is about paying only for what you use, right-sizing resources, and choosing the right pricing model
- EC2 Pricing Models:
- On-Demand β pay per hour/second, no commitment, most expensive
- Reserved Instances (RI) β 1 or 3-year commitment, up to 72% savings; Standard RI (can't change instance type) vs. Convertible RI (can change)
- Savings Plans β flexible alternative to RIs; Compute Savings Plans apply across EC2, Fargate, Lambda
- Spot Instances β up to 90% savings; can be interrupted; ideal for fault-tolerant, stateless, batch workloads
- Dedicated Hosts β physical server; used for licensing compliance (e.g., per-socket/per-core software)
- S3 Storage Classes (cheapest to most available):
- S3 Standard β S3 Intelligent-Tiering β S3 Standard-IA β S3 One Zone-IA β S3 Glacier Instant β S3 Glacier Flexible β S3 Glacier Deep Archive
- Use S3 Lifecycle Policies to auto-transition objects between tiers
- Data Transfer Costs:
- Inbound to AWS = free
- Outbound to internet = charged
- Transfer between AZs = charged (small)
- Transfer within same AZ = free
- Use VPC Endpoints to avoid NAT Gateway data processing costs
- Auto Scaling β ensures you don't over-provision; scale in during off-peak hours
- Right-sizing β use CloudWatch metrics + AWS Compute Optimizer to identify over-provisioned resources
- Managed Services β RDS, ECS, Lambda reduce operational overhead and often reduce cost vs. self-managed EC2
- Serverless (Lambda, Fargate, DynamoDB on-demand) β pay only per invocation/request; great for spiky or unpredictable workloads
- CloudFront β reduces S3 egress costs and origin load by caching at edge
- AWS Cost Tools: Cost Explorer, AWS Budgets, Trusted Advisor (cost checks), Cost & Usage Report (most detailed)
How It Works
- Spot Instance Interruption: AWS gives a 2-minute warning before reclaiming; use Spot with Auto Scaling groups or Spot Fleet for resilience
- Reserved Instance Flexibility: RIs apply to any instance in the same region (if regional) or same AZ (if zonal); Convertible RIs can be exchanged; Standard RIs can be sold on the RI Marketplace
- S3 Intelligent-Tiering: Automatically moves objects between frequent and infrequent access tiers based on access patterns β no retrieval fee, small monitoring fee per object
- Savings Plans vs. RIs: Savings Plans are commitment-based ($/hour spend), not capacity reservations β more flexible but no capacity guarantee
Key Values
| Option | Savings vs On-Demand | Interruption Risk | Best For |
|---|---|---|---|
| Reserved (1yr) | ~40% | None | Steady-state workloads |
| Reserved (3yr) | ~60β72% | None | Long-term predictable |
| Savings Plans | Up to 66% | None | Flexible compute needs |
| Spot | Up to 90% | Yes | Batch, CI/CD, stateless |
| On-Demand | 0% | None | Short-term, unpredictable |
| S3 Class | Min Storage Duration | Retrieval Fee |
|---|---|---|
| Standard | None | None |
| Intelligent-Tiering | None | None |
| Standard-IA | 30 days | Yes |
| One Zone-IA | 30 days | Yes |
| Glacier Instant | 90 days | Yes |
| Glacier Flexible | 90 days | Yes |
| Glacier Deep Archive | 180 days | Yes |
Exam Gotchas
- Spot β always the cheapest answer β if the question mentions the workload *cannot be interrupted
My Notes
Edit your saved notes here. Saving replaces what is stored.
Add More Notes
Paste new content here. It will be added below your existing notes β nothing gets overwritten.