SAA-C03 domain - 20% of the exam

Design Cost-Optimized Architectures

Design Cost-Optimized Architectures is 20% of the SAA-C03 exam. These are the objectives it covers, each with practice questions, with every answer explained.

The domain in numbers

71
Practice questions
7
Objectives
8
Easy
42
Medium
21
Hard

Where people go wrong

  • Reversing the flexibility and depth of commitment types

    The wrong answers hand the wrong discount to the wrong plan. A Compute Savings Plan is offered as the deepest option when the family and Region are fixed, and an EC2 Instance Savings Plan as the flexible one. It works the other way round. A Compute Savings Plan applies across any family, size, Region and operating system, and covers Fargate and Lambda, at a shallower discount. An EC2 Instance Savings Plan locks one family in one Region in return for the deeper discount.

    See it in a sample question
  • Lowering the rate on idle hours instead of removing them

    The wrong answers cut the price of capacity that keeps running. Reserved Instances, a Savings Plan, downsizing and Spot all still bill every hour the instances sit idle, so a business-hours fleet or a sporadic API keeps paying through evenings and weekends. The correct answers stop the charge itself: scheduled Auto Scaling actions scale the group to zero outside business hours, Lambda bills per request, and Aurora Serverless v2 contracts when the database is idle.

    See it in a sample question
  • Confusing gateway and interface VPC endpoints

    Terabytes of S3 or DynamoDB traffic through a NAT gateway are billed per gigabyte, and the wrong answers reach for the wrong fix: an interface endpoint for S3 or DynamoDB, more NAT gateways, Direct Connect or public subnets. Gateway endpoints exist only for Amazon S3 and DynamoDB, carry no hourly or per-gigabyte charge and work through route table entries. The reverse also appears: a gateway endpoint offered for SQS, KMS or CloudWatch, which need interface endpoints powered by AWS PrivateLink.

    See it in a sample question
  • Swapping the jobs of Budgets, Cost Explorer, the Cost and Usage Report and tags

    Each wrong answer names a real billing feature that does a different job. Budgets alert when actual or forecast spend crosses a threshold. Cost Explorer charts spend interactively by service and account. The Cost and Usage Report delivers hourly line items to S3 for Athena and QuickSight. Activated cost allocation tags label resources so spend can be grouped by project and team. Matching the verb in the requirement (alert, chart, query, label) picks the right feature.

    See it in a sample question
  • Expecting acceleration or a load balancer swap to cut egress

    Data transfer out to the internet is billed per gigabyte each time an object is served, and the wrong answers change the network path instead of the number of origin requests. S3 Transfer Acceleration adds a surcharge and caches nothing, Global Accelerator does not cache content, and a Network Load Balancer offers no edge caching. CloudFront serves repeat requests from edge caches and prices egress lower per gigabyte than direct internet egress.

    See it in a sample question

Objectives in this domain

What to study

The official documents this domain's practice questions cite most.

Sample questions from this domain

Free sampleDesign Cost-Optimized Architectureshard

An enterprise runs a steady baseline of EC2 and Fargate compute that never drops to zero, but over the coming year it expects to shift workloads between instance families and between two AWS Regions as teams modernise. Finance wants a committed-spend discount on the constant baseline while preserving the freedom to change instance family, size, operating system, and Region without losing the discount. Which purchase option meets these goals?

  • AA Compute Savings Plan committing to a fixed hourly spend, which applies the discount automatically across instance families, sizes, Regions, operating systems, and to Fargate usage as workloads move. Correct
  • BAn EC2 Instance Savings Plan committing to spend within a single instance family in one Region, applying the discount automatically as size and operating system vary inside that chosen family.
  • CStandard Reserved Instances purchased per instance family and Region, giving a strong rate on the baseline while the reservation is exchanged through a support case whenever a family change is needed.
  • DOn-Demand Capacity Reservations held in each Region, guaranteeing capacity for the baseline so the workload can move freely between families without committing to any reduced hourly rate.
Select a Compute Savings Plan when a steady compute baseline needs a committed discount while retaining freedom across instance family, size, Region, operating system, and Fargate. A Compute Savings Plan commits the customer to a steady hourly dollar amount for one or three years and automatically applies the discounted rate to any matching EC2, Fargate, or Lambda usage no matter the instance family, size, operating system, tenancy, or Region. That breadth is exactly what a baseline expecting cross-family and cross-Region movement needs, whereas EC2 Instance Savings Plans and Standard RIs trade flexibility for a deeper but narrower rate.

Why A is correct: Compute Savings Plans commit to an hourly dollar spend and discount any EC2, Fargate, or Lambda usage regardless of family, size, Region, or operating system, matching the baseline saving with full flexibility.

Why B is wrong: Tempting because it is a Savings Plan with a deeper rate, but it locks the discount to one instance family in one Region, so it breaks when workloads move family and Region as the stem requires.

Why C is wrong: Tempting because Standard RIs discount steady baselines, but they cannot change instance family and are tied to a Region, and they are not exchangeable, so the required cross-family freedom is lost.

Why D is wrong: Tempting because capacity reservations cover a steady baseline, but they only reserve capacity at the full On-Demand rate and provide no committed-spend discount, so the cost goal is not met.

Free sampleDesign Cost-Optimized Architecturesmedium

A development team runs an internal MySQL-compatible reporting database on a fixed Amazon RDS instance. The database sees heavy use for a few hours during business days and is almost completely idle on evenings and weekends, yet it is billed at full instance price the entire time. The team wants a relational, MySQL-compatible option whose capacity grows under load and contracts sharply when idle, so they stop paying for unused compute. Which option best fits this pattern?

  • AMigrate the database to Amazon Aurora Serverless v2 in MySQL-compatible mode so capacity scales up during reporting hours and contracts to a low minimum when idle. Correct
  • BMove the reporting workload to Amazon DynamoDB on-demand so the team pays per request and nothing while the database sits idle overnight and across the weekend.
  • CBuy a one-year RDS Reserved Instance for the existing class so the heavy weekday reporting hours are billed at the lower committed discounted rate.
  • DResize the RDS instance to a much larger class to clear the busy hours faster and rely on the headroom to keep the reporting database responsive the rest of the time.
Select Aurora Serverless v2 for relational MySQL workloads that spike briefly then sit idle to avoid paying for an always-on instance. Aurora Serverless v2 automatically scales database capacity in fine-grained increments to match current demand and contracts toward a low minimum when the workload is idle, so a MySQL-compatible reporting database that is busy only during business hours stops paying for a continuously provisioned full-size instance.

Why A is correct: Aurora Serverless v2 is MySQL-compatible and adjusts capacity in fine steps with demand, contracting to a low minimum during long idle windows, which matches a relational workload that is busy briefly and idle most of the time.

Why B is wrong: DynamoDB is a NoSQL key-value store and would require rewriting the relational reporting queries, so it fails the explicit need for a MySQL-compatible relational database.

Why C is wrong: A Reserved Instance discounts an always-on commitment and keeps paying through every idle evening and weekend, so it does not stop charges during the long periods of no use.

Why D is wrong: A larger fixed instance raises the hourly price and still bills continuously through the idle periods, so it increases cost instead of removing the charge for unused time.

Free sampleDesign Cost-Optimized Architectureshard

A data processing fleet runs on EC2 instances in private subnets and reads and writes several terabytes of objects in an Amazon S3 bucket in the same Region every day. All of this S3 traffic currently leaves the private subnets through a NAT gateway, and the monthly bill is dominated by NAT gateway data processing charges. The team wants to eliminate the per-gigabyte NAT charge for the S3 traffic without exposing the instances to the internet. Which change should the architect make?

  • AReplace the existing NAT gateway with a larger NAT gateway sized for higher throughput so the per-gigabyte data processing rate applied to the S3 traffic is reduced.
  • BCreate an interface VPC endpoint for Amazon S3 in each subnet so the bucket is reached through private elastic network interfaces instead of through the NAT gateway path.
  • CProvision an AWS Direct Connect connection from the data centre to the Region so the S3 reads and writes traverse the dedicated link rather than the NAT gateway.
  • DCreate a gateway VPC endpoint for Amazon S3 and add its route to the private subnet route tables so S3 traffic stays on the AWS network and bypasses the NAT gateway. Correct
A gateway VPC endpoint for S3 carries in-Region S3 traffic privately at no extra charge, removing NAT gateway data processing costs. NAT gateway data processing is billed per gigabyte, so routing bulk S3 access through it is expensive. A gateway VPC endpoint for S3 adds a prefix-list route to the subnet route tables, keeping the traffic on the AWS network with no hourly or per-gigabyte fee, which removes the NAT charge while the instances stay private.

Why A is wrong: A NAT gateway has no instance size to scale and its data processing charge is a flat per-gigabyte rate, so swapping it changes nothing and the S3 traffic keeps incurring the charge.

Why B is wrong: An interface endpoint for S3 works but bills an hourly charge plus a per-gigabyte processing fee, so for terabytes of daily traffic it is far costlier than the free gateway endpoint option.

Why C is wrong: Direct Connect links an on-premises network to AWS and does nothing for instances already inside the VPC, so the in-Region S3 traffic would still flow through the NAT gateway.

Why D is correct: A gateway VPC endpoint routes in-Region S3 traffic privately through the VPC route table at no hourly or per-gigabyte charge, removing the NAT gateway data processing cost while keeping instances off the internet.

Free sampleDesign Cost-Optimized Architectureseasy

A company runs several projects in one AWS account and wants to break down its monthly bill by project and by team so each group can see what it spends. The resources are shared across the same services, so service-level totals alone do not reveal who is driving the cost. The team wants to label resources so spending can be grouped and filtered by those labels. Which AWS feature should the architect use to enable this breakdown?

  • ACreate a separate linked AWS account for every project and team so that each group receives its own isolated monthly bill split out by account.
  • BEnable AWS Budgets with one budget per project so that each project total appears separately on the monthly billing dashboard for every team.
  • CTurn on detailed CloudWatch billing metrics per project so each metric stream reports the resource spending attributable to that individual project and team.
  • DApply cost allocation tags such as project and team to the resources and activate them so AWS groups and filters spending by those tag values. Correct
Activated cost allocation tags label resources so AWS can group and filter spending by values such as project or team. Breaking a shared bill down by project and team needs labels on the resources themselves. Cost allocation tags attach key and value pairs to resources, and once a tag is activated in the billing console AWS includes it in cost data so Cost Explorer and the bill can group and filter spend by each tag value.

Why A is wrong: Splitting into many accounts can separate spend but is heavy operational overhead for shared resources, and the requirement is to label and group existing resources rather than re-architect the account structure.

Why B is wrong: Budgets track spend against a threshold and send alerts, but they do not label resources or break a shared bill down by project and team, so they do not provide the grouping requested.

Why C is wrong: CloudWatch billing metrics only show estimated total charges for alarms and cannot attribute shared-resource spend to a project or team, so they do not deliver the tag-based breakdown needed.

Why D is correct: Activated cost allocation tags attach project and team labels to resources, and once activated AWS surfaces them so the bill and Cost Explorer can be grouped and filtered by each tag value.

Free sampleDesign Cost-Optimized Architectureshard

A company hosts a static media catalogue in an Amazon S3 bucket and serves it directly to a global audience over the internet. The same popular objects are downloaded repeatedly, and the bill is dominated by S3 data transfer out to the internet, which is charged per gigabyte each time an object is served. The team wants to cut the recurring egress cost for these repeatedly requested objects without changing the bucket contents. Which approach is the most cost-effective?

  • AEnable S3 Transfer Acceleration on the bucket so global downloads are routed over the optimised edge network, lowering the per-gigabyte internet egress charge for popular objects.
  • BPut an Amazon CloudFront distribution in front of the bucket so repeated requests are served from cached copies at edge locations and only cache misses pull from the S3 origin. Correct
  • CFront the bucket with AWS Global Accelerator so the popular object downloads travel over the AWS backbone to the bucket, which reduces the data transfer out charge per gigabyte.
  • DApply an S3 Lifecycle rule that transitions the popular objects to S3 Glacier Flexible Retrieval so the lower storage class reduces the data transfer out charge on each download.
Amazon CloudFront caches popular S3 objects at the edge and prices egress lower than direct S3 internet transfer, cutting recurring data transfer out cost. Serving popular objects straight from S3 charges per-gigabyte internet egress on every download. CloudFront caches those objects at edge locations so most requests are served without touching S3, and CloudFront data transfer out to the internet is generally cheaper per gigabyte than S3 egress, so both the volume hitting the origin and the unit price drop.

Why A is wrong: Transfer Acceleration speeds long-distance transfers using edge routing and adds an acceleration surcharge, but it does not cache objects, so every download still incurs full S3 egress cost.

Why B is correct: CloudFront serves repeat downloads from edge caches so most requests never reach S3, and its data transfer out is priced lower than direct S3 internet egress, which cuts the recurring cost for popular objects.

Why C is wrong: Global Accelerator improves path quality for TCP and UDP traffic but does not cache content or lower S3 egress pricing, so repeated downloads keep incurring the same per-gigabyte cost.

Why D is wrong: Lifecycle transitions change storage class and retrieval behaviour for cold data, not data transfer out pricing, and archiving frequently downloaded objects would add retrieval fees and delay.

Other domains in this exam

See also the SAA-C03 cert hub, the study guide, and the cheat sheet.

Examworthy is not affiliated with or endorsed by Amazon Web Services. Original, blueprint-aligned practice material only.