AWS SAA-C03: Design Cost-Optimized Architectures Practice Questions

Configuring a Lifecycle expiration policy to delete objects after 7 days eliminates ongoing storage costs entirely for data that will never be accessed again — the cheapest possible outcome. S3 Intelligent-Tiering adds a per-object monitoring fee (applies to objects ≥ 128 KB; 50 KB objects are always stored at Standard tier within Intelligent-Tiering with no tiering benefit) and does not delete unneeded data, so costs accumulate indefinitely. Moving to Glacier Instant Retrieval still incurs storage costs indefinitely and adds transition request costs, which is wasteful for data never accessed after 7 days. Compressing into ZIP archives reduces object count but increases operational complexity, and Standard-IA still incurs storage and retrieval costs — it also has a minimum storage duration charge of 30 days, penalizing objects deleted before 30 days.

Compute Savings Plans apply to Fargate usage and provide up to 50% savings over On-Demand for steady, committed workloads — ideal for 24/7 services. Fargate Spot provides up to 70% discount for tasks that can tolerate interruption, making it perfect for bursty batch jobs. Running steady-state services on Fargate Spot is risky because Spot capacity can be reclaimed, causing service disruptions for 24/7 workloads. Using On-Demand EC2 for bursty jobs would be more expensive and harder to scale than Fargate Spot. Purchasing Reserved Instances for bursty, unpredictable batch jobs is wasteful because utilization will be low and the commitment does not match the usage pattern. Running everything on On-Demand Fargate forgoes available discounts.

Lambda is billed on GB-seconds (memory × duration). At 512 MB / 800 ms: 0.5 GB × 0.8 s = 0.4 GB-s per invocation. At 1024 MB / 350 ms: 1.0 GB × 0.35 s = 0.35 GB-s per invocation. The higher-memory configuration produces fewer GB-seconds, reducing cost. This is not merely due to duration; the calculation must account for both memory and duration together. Cost does not increase simply because memory doubled — the duration improvement more than compensates. Lambda is billed on both invocations AND duration (GB-seconds), so invocation count alone does not determine cost. Lambda does not automatically optimize resource usage; memory must be explicitly configured.

An AWS Budget on the management account with a forecasted threshold of $50,000 correctly triggers an alert when projected spend is on track to exceed the limit across all consolidated accounts. Cost Allocation Tags enable chargeback by tagging resources to teams/cost centers; activating and enforcing them is the standard mechanism for team-level cost attribution. An SCP denying large EC2 instance launches is a preventive control that stops over-sized instances from being created at all, which is more effective than reactive detection. Cost Anomaly Detection is designed to detect unusual spending patterns relative to baseline—not to alert on a fixed projected total—making it incorrect for requirement (1). AWS Service Quotas controls the number of resources, not the size/type of EC2 instances launched, making it incorrect for requirement (3). AWS Config rules are detective (after-the-fact) controls; instances are launched first and then terminated, which could still result in unwanted spend and doesn't prevent the action.

AWS CodeBuild is a fully managed CI/CD build service that charges only for the compute minutes used during active builds (~$0.005/build minute for general1.small). With 500 × 15-minute average builds/day = 7,500 build-minutes, CodeBuild eliminates all idle compute costs. No fleet management is required. An Auto Scaling group with On-Demand instances reduces idle cost but still requires instances to start and run; scale-in lag means instances remain running after jobs complete, and there is a per-hour minimum billing granularity. Spot Instances in an Auto Scaling group improve cost further but require interruption handling logic (requeue interrupted builds), adding operational complexity and potential build delays during Spot reclamation. A blend of Reserved + On-Demand Instances is less flexible — Reserved Instances commit to capacity that may sit idle, and On-Demand burst still costs more than CodeBuild's per-minute billing for this bursty pattern.