AIAI Agent Dev Labaiagentdev-lab.com
Sign in to save progressSign in
Menu

SAA-C03 · LESSON 4

Design Cost-Optimized Architectures20% · 4 topics

Learn to design cost-optimized storage, compute, databases and networks.

4.1Design cost-optimized storage solutions

4.1 Design cost-optimized storage solutions.

Concept

In S3, combining storage classes matched to access frequency with lifecycle rules is the core of cost reduction. Use S3 Intelligent-Tiering if access patterns are unknown or change, One Zone-IA for re-creatable data, and Glacier Flexible Retrieval or Deep Archive for long-term retention. For EBS, save by changing gp2 to gp3 and deleting unused volumes and snapshots.

How it works / steps

  1. Check access trends with S3 Storage Lens or storage class analysis
  2. Move old data to cheaper classes with lifecycle rules and expire what is not needed
  3. Choose the Glacier class that matches retrieval frequency and time requirements
  4. Change EBS to gp3 and delete unused volumes
  5. Automatically delete incomplete multipart uploads

Workplace example

Set a rule that keeps audit logs in Standard for 90 days, then Glacier Flexible Retrieval, and deletes them after 7 years, retrieving only for the annual audit.

Common confusions and exam tips

Mini quiz

Q1. Report data is referenced often for 30 days after creation, occasionally for the following year, and almost never after that. It can be deleted after 7 years. Which setting is most appropriate?

  1. Store it as EBS snapshots and recreate them every year
  2. Store it in Glacier Deep Archive from the start and restore it every time it is needed
  3. Store everything in S3 Standard and delete it manually after 7 years
  4. Use lifecycle rules to move to Standard-IA after 30 days and Deep Archive after 1 year, and delete after 7 years
Show answer and explanation

Answer: D Use lifecycle rules to move to Standard-IA after 30 days and Deep Archive after 1 year, and delete after 7 years

When you know how access frequency changes, use lifecycle rules to move data to cheaper classes in stages and delete it at the end of the retention period.

Q2. Storage costs of an S3 bucket with versioning enabled keep rising. Which measure is most appropriate?

  1. Use lifecycle rules to transition or expire noncurrent versions
  2. Shorten object names to reduce storage
  3. Copy the bucket to another Region and delete the original
  4. Suspending versioning automatically deletes past versions too
Show answer and explanation

Answer: A Use lifecycle rules to transition or expire noncurrent versions

With versioning, noncurrent versions are also charged for storage. Set lifecycle rules to transition or expire noncurrent versions. Suspending versioning leaves existing versions in place.

Reference links

4.2Design cost-optimized compute solutions

4.2 Design cost-optimized compute solutions.

Concept

Lower costs by combining purchase options (Savings Plans for the steady base, On-Demand for variation, Spot for interruptible work), right-sizing (AWS Compute Optimizer recommendations), moving to Graviton, stopping during unused hours and using serverless. Auto Scaling groups can mix On-Demand and Spot.

How it works / steps

  1. Identify oversized instances with Compute Optimizer
  2. Discount always-on usage with Compute Savings Plans
  3. Use Spot for interruption-tolerant batch and stateless processing
  4. Stop development environments automatically at night and on weekends
  5. Move intermittent processing to Lambda or Fargate to eliminate idle costs

Workplace example

Make CI build servers a Spot-only Auto Scaling group, and cover production web's minimum capacity with Savings Plans and peaks with On-Demand.

Common confusions and exam tips

Mini quiz

Q1. You want to find oversized EC2 instances and get recommendations for the right size. Which service is the best fit?

  1. AWS Artifact
  2. AWS Compute Optimizer
  3. Amazon Inspector
  4. AWS Pricing Calculator
Show answer and explanation

Answer: B AWS Compute Optimizer

Compute Optimizer analyzes utilization metrics and recommends right sizes and settings for EC2, EBS, Lambda and more.

Q2. In an Auto Scaling group, you want to lower costs by separating the minimum capacity that is always needed from the capacity added for load increases. Which design is most appropriate?

  1. Buy 1-year Reserved Instances for the peak count
  2. Cover all instances with Savings Plans and use no Spot
  3. Run the base on Spot and the extra on On-Demand
  4. Cover the base count with discounted On-Demand and scale the extra automatically, including Spot
Show answer and explanation

Answer: D Cover the base count with discounted On-Demand and scale the extra automatically, including Spot

Using long-term discounts for always-needed capacity and Auto Scaling with Spot or On-Demand for variable capacity balances cost and availability well. [Other options] Covering everything commits to idle hours too. A Spot base can be interrupted below the minimum. Reserving the peak wastes capacity.

Reference links

4.3Design cost-optimized database solutions

4.3 Design cost-optimized database solutions.

Concept

For DynamoDB capacity mode, use on-demand for unpredictable load and provisioned (discounted with auto scaling and reserved capacity) for steady load. For RDS, use right-sized instances, Reserved Instances, stopping development environments, and Aurora Serverless v2 to pay according to load. Moving rarely accessed items to the DynamoDB Standard-IA table class lowers storage costs.

How it works / steps

  1. Check the load pattern (steady, variable, intermittent)
  2. For DynamoDB, choose the capacity mode and table class that fit the pattern
  3. Discount always-on production RDS with Reserved Instances
  4. Use Aurora Serverless v2 for intermittently used DBs
  5. Delete unneeded items automatically with TTL to lower storage costs

Workplace example

Run a new service on DynamoDB on-demand while load is unpredictable, and switch to provisioned with auto scaling once load stabilizes after six months.

Common confusions and exam tips

Mini quiz

Q1. An internal request system on a relational DB is used for only a few hours during weekdays. Which is the most appropriate way to cut costs?

  1. Run the request system on EC2 instance store
  2. Keep the largest RDS instance running all the time
  3. Use Aurora Serverless v2 to keep capacity automatically small according to load
  4. Add five read replicas
Show answer and explanation

Answer: C Use Aurora Serverless v2 to keep capacity automatically small according to load

For a DB used only for limited hours, Aurora Serverless v2, which adjusts capacity automatically, or stopping it during unused hours keeps costs down.

Q2. You are choosing a container runtime. Load is constant, and a large number of instances will be used for a long time. If cost matters most, which approach is most appropriate?

  1. Apply discounts to the EC2 launch type and keep instance utilization high
  2. Always use only Fargate On-Demand regardless of load
  3. Run twice the needed capacity to keep headroom
  4. Drop containers and process continuously with Lambda
Show answer and explanation

Answer: A Apply discounts to the EC2 launch type and keep instance utilization high

With constant heavy load, securing EC2 capacity at a discount and running it at high utilization tends to be cheaper. If you want less operational work, choose Fargate (Savings Plans also apply).

Reference links

4.4Design cost-optimized network architectures

4.4 Design cost-optimized network architectures.

Concept

Designing with data transfer charges in mind is important. Outbound traffic to the internet and transfers between AZs and Regions are charged. NAT gateways are billed by hours and data processed, so use gateway VPC endpoints (free) for S3 and DynamoDB. Delivering through CloudFront often has lower outbound charges than serving directly from the origin.

How it works / steps

  1. Check the breakdown of data transfer costs in Cost Explorer
  2. Switch S3 and DynamoDB traffic to gateway endpoints to reduce NAT processing charges
  3. Use CloudFront caching for delivery to users
  4. Review designs with heavy inter-AZ traffic (while balancing availability)
  5. For connecting many VPCs, compare Transit Gateway and peering costs

Workplace example

A batch in a private subnet sent 5 TB a day to S3 through a NAT gateway, so switching to a gateway endpoint eliminates the NAT processing charges.

Common confusions and exam tips

Mini quiz

Q1. You want to track AWS costs by department and be notified when a budget is likely to be exceeded. Which combination is most appropriate?

  1. Apply cost allocation tags, analyze in Cost Explorer and alert with Budgets
  2. Set CloudWatch billing alarms using per-department metrics
  3. Buy Savings Plans per department and check discount utilization
  4. Export a Cost and Usage Report per department and total it in Athena at month end
Show answer and explanation

Answer: A Apply cost allocation tags, analyze in Cost Explorer and alert with Budgets

Allocate costs to departments with tags, visualize them in Cost Explorer, and set Budgets alerts against actual and forecast spend. [Other options] CUR with Athena is a month-end tally and cannot alert before overspending. Billing alarms use the account's estimated total, not department allocation. Savings Plans are discounts, not allocation or alerts.

Q2. Cost Explorer shows data transfer costs have tripled since last month. What should you do next?

  1. Add S3 gateway endpoints to every VPC
  2. Put CloudFront in front of all services right away
  3. Break down by usage type to find which transfer path grew
  4. Buy Savings Plans to discount transfer costs
Show answer and explanation

Answer: C Break down by usage type to find which transfer path grew

First identify the type and source of the transfer. Then choose a fix for the cause: a gateway endpoint for S3 traffic through NAT, CloudFront for delivery, and so on. [Other options] Fixes before finding the cause have unknown effect. Gateway endpoints only help S3 traffic via NAT. Savings Plans do not discount data transfer.

Reference links

Lessons are written by this site based on public information and do not guarantee exam content. Product specifications change, so check the latest information at the reference links.

Notes (unofficial, original questions, no pass guarantee)