SAA-C03 Design Cost-Optimized Architectures Practice Question

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✓

  Purchase a one-year Standard Reserved Instance for each of the six running instances in the same instance family and Region.

  Why this is correct

  Standard Reserved Instances for a specific instance family and Region provide a significant discount compared to On-Demand for steady-state usage and are a valid way to reduce cost when the instance family is stable. Because the fleet is constant and the term matches the company's one-year willingness to commit, this achieves the cost reduction without altering the architecture or availability.

• ✗

  Enable detailed monitoring on all instances and configure an Auto Scaling target tracking policy at 40% CPU.

  Why it's wrong here

  Detailed monitoring improves metric granularity and target tracking can right-size capacity, but the scenario states the workload is already steady and predictable, so there is little waste to remove. Detailed monitoring also adds a small per-instance charge. Neither action lowers the hourly compute rate, so the compute cost for this constant fleet would not meaningfully decrease.

• ✗

  Replace the On-Demand instances with Spot Instances and let the Auto Scaling group replace any interrupted capacity.

  Why it's wrong here

  Spot Instances are unsuitable here because the requirement explicitly rules out reducing availability, and Spot capacity can be reclaimed with only a two-minute interruption notice. Although Spot is cheaper, the risk of losing all six instances during a capacity event directly violates the stated availability constraint, so this is not an acceptable cost optimization for this workload.

• ✓

  Purchase a one-year Compute Savings Plan covering the expected steady-state compute usage.

  Why this is correct

  Compute Savings Plans apply to EC2 usage regardless of instance family, size, tenancy, operating system, or Region, and they also cover Fargate and Lambda. Because the workload is steady and predictable, a one-year commitment will discount the qualifying compute spend without requiring any change to instance types, Availability Zone spread, or the application architecture.

• ✗

  Move the instances into a placement group and enable cluster networking between them.

  Why it's wrong here

  Placement groups influence network latency and throughput between instances, not the price paid for compute. Cluster placement groups can actually reduce resilience because they pack instances close together, which conflicts with the requirement to keep availability high across three Availability Zones. This action does not change the On-Demand billing rate at all.

Go deeper

Related to this question