Skip to content
BytePatterns

SAA-C03 · Domain 4: Design Cost-Optimized Architectures · 20% of the exam

Task 4.2: Design cost-optimized compute solutions.

Purchase options for each kind of load, right-sizing instance family and size, serverless and containers for spiky use, and scaling down what nobody needs at night.

Study it

  • Choosing a load balancer: Application, Network or Gateway Load Balancer

    Partly covered by: EC2, Auto Scaling & Load Balancers, Load Balancing

  • Compute purchase options: On-Demand, Savings Plans, Reserved Instances and Spot

    Lesson: AWS Cost Levers

  • Right-sizing and scaling for cost: instance family and size, Compute Optimizer, hibernation

    Partly covered by: AWS Cost Levers

  • Cost visibility: Cost Explorer, Budgets, Cost and Usage Report and cost allocation tags

    Lesson coming

Sample questions

Try each one before opening the answer. Every option is explained, with the AWS documentation page that proves it.

Question 1 · choose 1

A company runs a steady baseline of compute around the clock: EC2 instances that it plans to move between instance families and Regions over the next year, plus AWS Fargate tasks and Lambda functions. It wants a commitment-based discount that still covers all of this usage. Which purchase option should it choose?

  1. AEC2 Instance Savings Plans for the instance family in use today
  2. BCompute Savings Plans for the steady baseline of hourly spend
  3. CStandard Reserved Instances for each of the current instance types
  4. DSpot Instances with an On-Demand fallback for any interruptions
Show the answer and why

  • AEC2 Instance Savings Plans for the instance family in use today

    Incorrect

    EC2 Instance Savings Plans commit to one instance family in one Region, so they stop applying when the company moves families or Regions.

  • BCompute Savings Plans for the steady baseline of hourly spend

    Correct

    Compute Savings Plans apply regardless of instance family, size, OS, tenancy or Region, and also to Fargate and Lambda usage.

  • CStandard Reserved Instances for each of the current instance types

    Incorrect

    Reserved Instances discount EC2 instances that match their attributes; they do not cover Fargate or Lambda, and they lose value as the fleet changes.

  • DSpot Instances with an On-Demand fallback for any interruptions

    Incorrect

    Spot capacity can be reclaimed, which does not fit an always-on baseline, and it is not a discount for Fargate tasks or Lambda functions.

Flexibility across families, Regions and serverless compute is what distinguishes Compute Savings Plans from the narrower options.

AWS documentation

Lesson: AWS Cost LeversTask 4.2Report a problem with this question

Question 2 · choose 1

A studio renders animation frames every night. Each frame is an independent job that can be retried, deadlines are flexible, and the jobs run on many instance types. The company wants the lowest compute cost with as few interruptions as possible. Which approach should a solutions architect recommend?

  1. ASpot Instances of a single instance type, all launched in one Availability Zone
  2. BOne-year Standard Reserved Instances sized to cover the nightly render peak
  3. COn-Demand Instances, with an EC2 Instance Savings Plan covering the nightly hours
  4. DSpot Instances in many types and zones with price-capacity-optimized allocation
Show the answer and why

  • ASpot Instances of a single instance type, all launched in one Availability Zone

    Incorrect

    Drawing on one capacity pool raises the chance of interruptions and unfilled capacity; AWS advises flexibility across types and zones.

  • BOne-year Standard Reserved Instances sized to cover the nightly render peak

    Incorrect

    Reserved Instances are billed for the whole term, including the daytime hours when nothing renders.

  • COn-Demand Instances, with an EC2 Instance Savings Plan covering the nightly hours

    Incorrect

    A Savings Plan is an hourly commitment for the whole term, and discounted On-Demand still costs more than Spot for retryable work.

  • DSpot Instances in many types and zones with price-capacity-optimized allocation

    Correct

    Spot suits fault-tolerant, flexible work, and the price-capacity-optimized strategy picks the pools least likely to be interrupted at a low price.

Retryable, flexible jobs belong on Spot. Being flexible about instance types and zones and using price-capacity-optimized allocation keeps interruptions down.

AWS documentation

Lesson: AWS Cost LeversTask 4.2Report a problem with this question

Question 3 · choose 1

Developers use a test environment of EC2 instances only from 08:00 to 18:00 on weekdays. The application builds large in-memory caches that take 40 minutes to warm up after a fresh start. The company wants to stop paying for the instances outside working hours without making developers wait every morning. What should a solutions architect do?

  1. AHibernate the instances every evening and start them each morning on a schedule
  2. BBuy one-year Reserved Instances for every instance in the test environment
  3. CStop the instances every evening and start them again each morning on a schedule
  4. DReplace the instances with Spot Instances that run around the clock
Show the answer and why

  • AHibernate the instances every evening and start them each morning on a schedule

    Correct

    Hibernation saves RAM to the EBS root volume and resumes the processes on start, and stopped instances incur no instance usage charges.

  • BBuy one-year Reserved Instances for every instance in the test environment

    Incorrect

    Reserved Instances are billed around the clock for the term, so the company keeps paying for the idle hours.

  • CStop the instances every evening and start them again each morning on a schedule

    Incorrect

    A stop clears the memory, so the caches must be rebuilt for 40 minutes every morning.

  • DReplace the instances with Spot Instances that run around the clock

    Incorrect

    Spot lowers the hourly price but the instances still run all night, and an interruption would also clear the caches.

Non-production capacity can be off at night. Hibernation keeps the warmed-up memory, so turning it off costs no morning wait.

AWS documentation

Lesson: AWS Cost LeversTask 4.2Report a problem with this question

Question 4 · choose 1

A company runs a Java web service on a fleet of x86-based general purpose EC2 instances. In tests the application runs unchanged on the Arm architecture. The company wants better price performance for this fleet with little effort. What should a solutions architect recommend?

  1. AMove the fleet to general purpose instances with AWS Graviton processors
  2. BMove the fleet to Dedicated Hosts so that the instances run on hardware reserved for the company
  3. CMove the fleet to burstable T instances with unlimited mode turned on
  4. DMove the fleet to the next larger x86 size and halve the instance count
Show the answer and why

  • AMove the fleet to general purpose instances with AWS Graviton processors

    Correct

    Graviton-based instances are built for the best price performance for workloads on EC2, and an application that already runs on Arm can move without code changes.

  • BMove the fleet to Dedicated Hosts so that the instances run on hardware reserved for the company

    Incorrect

    Dedicated Hosts are physical servers for licensing or compliance needs. They usually cost more and do not improve price performance.

  • CMove the fleet to burstable T instances with unlimited mode turned on

    Incorrect

    A web service with sustained CPU use would run above the baseline, and unlimited mode charges for the surplus credits that this uses.

  • DMove the fleet to the next larger x86 size and halve the instance count

    Incorrect

    Twice the size at half the count is about the same capacity for about the same price, so price performance does not change.

When the code already runs on Arm, moving to Graviton is the usual way to get more performance per dollar on EC2.

AWS documentation

Lesson: AWS Cost LeversTask 4.2Report a problem with this question

Question 5 · choose 1

A script runs for about 3 minutes once an hour to pull data from a partner API and write it to DynamoDB. It needs 512 MB of memory and runs on a t3.large On-Demand instance that is idle the rest of the time. Which option is the MOST cost-effective way to run it?

  1. AKeep the instance and buy a 1-year Reserved Instance for it
  2. BRun the script as a Lambda function on an Amazon EventBridge Scheduler schedule
  3. CRun the script in an Amazon ECS service on Fargate that stays running and sleeps between runs
  4. DMove the script to a Spot Instance that runs all day and keeps the same schedule
Show the answer and why

  • AKeep the instance and buy a 1-year Reserved Instance for it

    Incorrect

    The discount lowers the hourly rate, but the company still pays for an instance that is idle 57 minutes of every hour.

  • BRun the script as a Lambda function on an Amazon EventBridge Scheduler schedule

    Correct

    Lambda charges only for requests and for the time the code runs, so the company pays for about 3 minutes an hour instead of a full hour.

  • CRun the script in an Amazon ECS service on Fargate that stays running and sleeps between runs

    Incorrect

    A service that stays running is billed for its vCPU and memory the whole time, including while it sleeps.

  • DMove the script to a Spot Instance that runs all day and keeps the same schedule

    Incorrect

    Spot lowers the hourly price, but the instance is still paid for all day, and it can be interrupted.

Short, infrequent jobs are cheapest on pay-per-use compute that costs nothing between runs.

AWS documentation

Lesson: Lambda & Event-Driven DesignTask 4.2Report a problem with this question

Question 6 · choose 1

A company runs 12 small internal web services, each behind its own Application Load Balancer. Each service has its own host name, and traffic is light. The company wants to cut load balancer costs without changing the services. What should a solutions architect do?

  1. AReplace each Application Load Balancer with a Network Load Balancer
  2. BReplace the load balancers with one Gateway Load Balancer
  3. CUse one Application Load Balancer, with a host-based listener rule for each service
  4. DRemove the load balancers and return each instance's IP address from Route 53 multivalue answer records
Show the answer and why

  • AReplace each Application Load Balancer with a Network Load Balancer

    Incorrect

    There would still be 12 load balancers to pay for, and a Network Load Balancer cannot route by host name.

  • BReplace the load balancers with one Gateway Load Balancer

    Incorrect

    A Gateway Load Balancer distributes traffic to virtual appliances such as firewalls. It does not route web requests to services by host name.

  • CUse one Application Load Balancer, with a host-based listener rule for each service

    Correct

    One load balancer can route each host name to its own target group with host-header rules, so 12 hourly charges become one.

  • DRemove the load balancers and return each instance's IP address from Route 53 multivalue answer records

    Incorrect

    Multivalue answers are not a substitute for a load balancer, and instance addresses change as instances are replaced.

An Application Load Balancer routes on host names and paths, so many light services can share one.

AWS documentation

Task 4.2Report a problem with this question

Question 7 · choose 1

A production web tier runs a fixed 20 On-Demand instances around the clock in an Auto Scaling group. Traffic needs 20 instances only for a few peak hours each day and about 5 the rest of the time, and it changes gradually. The tier cannot tolerate instance interruptions. Which change reduces cost the MOST while keeping performance?

  1. AChange the instances to the next smaller size and keep 20 of them running
  2. BBuy Reserved Instances that cover all 20 instances for one year
  3. CMove the whole group to Spot Instances with a capacity-optimized allocation strategy
  4. DAdd a target tracking scaling policy and set the group's minimum capacity to 5
Show the answer and why

  • AChange the instances to the next smaller size and keep 20 of them running

    Incorrect

    Smaller instances save money all day but leave the tier short of capacity at peak.

  • BBuy Reserved Instances that cover all 20 instances for one year

    Incorrect

    The discount applies to all 20 instances, but the company keeps paying for 15 idle instances for most of the day.

  • CMove the whole group to Spot Instances with a capacity-optimized allocation strategy

    Incorrect

    Spot Instances can be interrupted, which this production tier cannot tolerate.

  • DAdd a target tracking scaling policy and set the group's minimum capacity to 5

    Correct

    The group then runs about 5 instances most of the day and adds instances as load rises, so the company pays for capacity only when traffic needs it.

A fixed fleet sized for peak pays for idle capacity all day. Scaling with demand removes that waste first.

AWS documentation

Lesson: EC2, Auto Scaling & Load BalancersTask 4.2Report a problem with this question

Question 8 · choose 2

A company is choosing purchase options for two workloads: a production web tier with a steady baseline of 10 instances that runs all year, and a nightly batch job that saves checkpoints and can be stopped and restarted at any time. Which choices are the MOST cost-effective? (Choose TWO.)

  1. AOn-Demand Instances for the web tier's baseline, to keep it flexible
  2. BDedicated Hosts for the batch job, so that its capacity is always reserved
  3. CA Savings Plan commitment that covers the web tier's steady baseline
  4. DSpot Instances for the nightly batch job
  5. EOn-Demand Capacity Reservations for the batch job, created every night
Show the answer and why

  • AOn-Demand Instances for the web tier's baseline, to keep it flexible

    Incorrect

    On-Demand has no commitment but also no discount, so a steady baseline costs more than it needs to.

  • BDedicated Hosts for the batch job, so that its capacity is always reserved

    Incorrect

    Dedicated Hosts are whole physical servers for licensing or compliance needs, the opposite of a cost saving here.

  • CA Savings Plan commitment that covers the web tier's steady baseline

    Correct

    A steady, year-round baseline is what a 1- or 3-year commitment discounts, and Savings Plans apply it automatically to that usage.

  • DSpot Instances for the nightly batch job

    Correct

    Spot uses spare capacity at a steep discount. A job that checkpoints and restarts can absorb interruptions.

  • EOn-Demand Capacity Reservations for the batch job, created every night

    Incorrect

    Capacity Reservations hold capacity at On-Demand prices. They give no discount on their own.

Match each workload to a purchase option: commit for the steady baseline, use Spot for interruptible, flexible work.

AWS documentation

Lesson: AWS Cost LeversTask 4.2Report a problem with this question

Practise domain 4 →Practise all domains →