AWS Certified Solutions Architect - Associate (SAA-C03) #13 Domain 4-1 Cost Optimization — Pricing Models
#12 wrapped up the high-performing domain (24%). The final domain is Cost Optimization (20%), which focuses on designs that deliver the same result more cheaply. The first topic is pricing models. #9 Choosing Compute covered purchasing options already, but this time we examine them more precisely from a cost-saving perspective.
EC2 Purchasing Options — Cost Perspective #
The core divide in cost questions is “how steady is the workload” and “how much flexibility is required.”
| Option | Savings | Flexibility | Fits |
|---|---|---|---|
| On-Demand | 0% | Maximum | Short-term , unpredictable |
| Spot | Up to ~90% | Interruptible | stateless , batch |
| Savings Plans | Up to ~72% | Varies by option | Steady load |
| Reserved Instances | Up to ~72% | Low | Steady + fixed attributes |
Savings Plans — Compute vs. EC2 Instance #
Savings Plans give you a discount in exchange for committing to “N dollars per hour” for 1 or 3 years. The trade-off between the two types appears on the exam.
| Type | Coverage | Discount |
|---|---|---|
| Compute Savings Plans | EC2 + Fargate + Lambda, any region/family | Lower |
| EC2 Instance Savings Plans | A specific instance family in a specific region | Higher |
If “the instance family may change and you also use Fargate/Lambda,” go with the flexible Compute Savings Plans; if “family and region are fixed,” go with the deeper-discount EC2 Instance Savings Plans.
Reserved Instances — Standard vs. Convertible #
- Standard RI — a large discount but limited attribute changes.
- Convertible RI — lets you change attributes like instance family and OS, but the discount is smaller.
These days Savings Plans are widely recommended for their flexibility and coverage, but the exam asks about both concepts.
S3 Cost Structure #
S3 cost is not just about storage capacity.
- Storage cost — varies by class (#11). Choosing the class that matches access frequency is the key to savings
- Request cost — the number of requests like PUT/GET
- Data retrieval cost — when pulling data out of IA , Glacier
- Data transfer cost — outbound transfer
The standard saving methods are moving infrequently accessed data to a lower-cost class via a lifecycle policy, and when you don’t know the pattern, automating it with Intelligent-Tiering.
Data Transfer Costs #
This is an area that’s easy to overlook in cost questions. The key principles are as follows.
| Path | Cost |
|---|---|
| Internet → AWS (inbound) | Free |
| AWS → Internet (outbound) | Charged |
| Same AZ, private IP | Free |
| Across AZs (cross-AZ) | Charged |
| Across regions (cross-region) | Charged (more) |
The ways to reduce costs are these.
- CloudFront — caches responses at the edge instead of origin outbound, reducing transfer cost and load
- VPC Endpoint — routes S3/DynamoDB traffic through a private path without NAT (#4)
- Co-locate in the same region , same AZ — avoid unnecessary cross-AZ/region transfer
- Be aware of the NAT Gateway data processing cost — it’s significant for high traffic
Architecture Choices That Lower Cost #
- right-sizing — bring over-provisioned instances to the right size (Compute Optimizer recommends them, #14)
- Move to serverless/managed — when there’s a lot of idle time, use Lambda , Fargate , Aurora Serverless
- Auto stop/scale — stop unused resources at night and on weekends, absorb variability with ASG
- Storage tiering — lifecycle policies + the right S3 class
Exam Question Patterns #
- “Steady load + instance-family flexibility/includes serverless.” → Compute Savings Plans
- “Fixed family , region, maximum discount.” → EC2 Instance Savings Plans / Standard RI
- “Batch that tolerates interruption, maximum savings.” → Spot
- “Reduce data transfer cost.” → CloudFront / VPC Endpoint / co-locate in the same region
- “Right-size over-provisioned instances.” → right-sizing with Compute Optimizer
- “Reduce cost of infrequently accessed S3 data.” → lifecycle + IA/Glacier
Common Traps #
1) Thinking inbound transfer is charged #
Inbound data transfer from the internet into AWS is free. Charges mainly apply to outbound and cross-AZ/region.
2) Recommending Reserved/Savings Plans for variable workloads #
Commitment-based discounts are for steady loads. For variable or intermittent loads, On-Demand , Spot , serverless are the fit.
3) Confusing Compute and EC2 Instance Savings Plans #
Compute is flexible (includes Fargate/Lambda, any family) with a smaller discount. EC2 Instance is fixed with a bigger discount.
4) Thinking transfer within the same region is always free #
Same-AZ private IP is free, but cross-AZ is charged.
Wrap-up #
What this post locked in:
- EC2 savings — Spot (tolerates interruption), Savings Plans/Reserved (steady). Compute SP (flexible) vs. EC2 Instance SP (fixed , bigger discount)
- S3 cost — storage + requests + retrieval + transfer. Save with class , lifecycle
- Data transfer — inbound free, outbound , cross-AZ/region charged. Reduce with CloudFront , VPC Endpoint
- Architecture — right-sizing, move to serverless, auto scale, storage tiering
Next — Domain 4-2 Cost Monitoring #
Now that we’ve locked in pricing models, we move to the tools that view and manage cost as the finale of the cost domain.
In #14 Domain 4-2 Cost Monitoring we’ll organize Cost Explorer for analyzing and forecasting cost, AWS Budgets for alerting on budget overruns, Trusted Advisor for checking cost , security , performance, Compute Optimizer for right-sizing recommendations, and consolidated billing plus cost allocation tags.