AWS Certified Solutions Architect - Associate SAA-C03 Practice Test 90
Question 890
A service moves x86 workers to Graviton and gains a moderate price/performance advantage, but still pays for dozens of idle oversize instances around the clock. What broader optimization should accompany architecture choice?
- Treat a Graviton migration as proof every remaining idle instance is free.
- Right-size and autoscale the fleet, then model eligible longer-term commitments against its true steady baseline.
- Purchase commitments for peak capacity before measuring the new workload baseline.
- Buy long-term compute commitments for the current oversized, continuously idle instance inventory before measuring and right-sizing the post-migration baseline.
Explanation
Correct Answer
B:CPU architecture is only one cost dimension; unused capacity and pricing coverage can dominate monthly spending.
Incorrect Answers
A:Graviton instances still accrue compute charges while running.
C:Poorly sized commitments can lock in costs for idle resources.
D:A commitment may discount eligible spend, but locking in an unnecessarily large baseline can preserve structural waste. First size the workload correctly, then model coverage against stable demand.
Question 891
A bank's internal analysis app is used heavily from 09:00 to 17:00 on weekdays. Outside those hours a small minimum fleet is sufficient, but current dynamic scaling only reacts after morning queue buildup. What pricing-oriented scale strategy fits?
- Keep all peak instances on permanently because scale schedules cannot alter desired capacity.
- Reduce all application availability checks to avoid the cost of detecting startup failures.
- Use scheduled Auto Scaling adjustments before business hours combined with dynamic protection for unexpected demand.
- Change billing tags at 09:00 instead of launching compute before demand.
Explanation
Correct Answer
C:Predictable utilization can be met proactively and scaled down after hours without a permanently oversized fleet.
Incorrect Answers
A:Scheduled actions can change Auto Scaling capacity bounds or desired size.
B:Losing health assurance does not provide reliable capacity savings.
D:Tags support attribution but do not create scheduled capacity.
Question 892
A web Auto Scaling group alternates rapidly between adding and removing EC2 instances because new hosts take seven minutes to warm, but the scaling policy interprets temporary metrics instantly. The oscillation wastes instance time. What tuning is appropriate?
- Set every warmup to zero and lower evaluation periods so the group responds to every noisy datapoint.
- Disable all CloudWatch alarms while keeping the same unstable scaling target.
- Change EBS snapshot frequency to one second to stabilize CPU target tracking.
- Set realistic warmup and scaling stability controls or cooldowns based on the application's ready time.
Explanation
Correct Answer
D:Capacity should be assessed after launch readiness, or short-lived metric changes can produce costly overcorrection.
Incorrect Answers
A:This increases oscillation when capacity isn't ready yet.
B:Removing visibility does not produce a controlled scaling policy.
C:Snapshot schedules do not alter how autoscaler metric feedback works.
Question 893
An SQS-backed video-processing fleet has workers waiting on queue messages. CPU stays around 10% even as the queue grows to a dangerous oldest-message age. What metric should drive autoscaling?
- Scale from queue backlog per available worker and oldest-message age or measured request/work rate rather than CPU alone.
- Set the CPU threshold to 100% and wait for CPU to rise before serving accumulated work.
- Only increase S3 object-tag count because each tag signifies an extra worker instance.
- Set all worker timeouts to infinity and assume this reduces queue age without consuming new messages.
Explanation
Correct Answer
A:Queue demand and drain rate represent the work left to do when CPU is not correlated with throughput.
Incorrect Answers
B:A queue-driven bottleneck may remain invisible to CPU-based target tracking.
C:S3 object tags are metadata associated with stored objects and do not provision EC2, ECS or AWS Batch worker capacity. The workload must use an actual scalable compute service rather than increasing unrelated metadata counts.
D:Long timeouts cannot process backlog without adequate consumers.
Question 894
A sensor pipeline triggers a 300-ms transform only when events arrive, averaging one invocation every few minutes. It currently runs a lightly utilized EC2 server continuously. Which compute model could lower idle charges?
- Use a large always-on EC2 Auto Scaling minimum of ten solely to serve rare subsecond transforms.
- Use Lambda for short compatible event transforms and compare invocation/duration pricing to the idle EC2 fleet.
- Move each event into a dedicated EMR Hadoop cluster and keep it active between messages.
- Use provisioned Lambda concurrency at a large constant value without any latency need.
Explanation
Correct Answer
B:Event-billed compute can avoid paying 24/7 for scarce short-lived work.
Incorrect Answers
A:A permanently running ten-instance Auto Scaling minimum pays for idle servers between brief transforms. The requirement is sporadic subsecond processing, so event-triggered or more elastic compute should be compared with the always-on baseline.
C:Distributed cluster overhead is unnecessary for tiny isolated transforms.
D:Preinitialized idle environments can add recurring charges, undercutting event-billed savings.
Question 895
A service performs heavy CPU processing constantly, with instances at high utilization for months. A proposal moves every second of the sustained workload to Lambda solely because it is serverless, without modeling duration/request cost. What should the architect compare?
- Assume Lambda is invariably cheapest for any continuous CPU load because no servers are visible to the customer.
- Use Lambda for a single multi-day uninterrupted invocation regardless of maximum execution duration.
- Evaluate sustained EC2 or container compute versus Lambda duration/request economics and operational requirements.
- Choose EC2 only by the number of S3 prefixes in the account.
Explanation
Correct Answer
C:Always-on compute can be more economical for a continuous high-volume workload even if event services reduce management overhead.
Incorrect Answers
A:Lambda still charges for invocations and resource-duration use.
B:Lambda has a per-invocation runtime limit and is not an arbitrary long-running process replacement.
D:Storage object namespace doesn't measure sustained compute requirements.
Question 896
A Lambda image transform uses 128 MiB and completes in 12 seconds. At 512 MiB it completes in two seconds. The team assumes the lower memory size is automatically cheaper. Which comparison is needed?
- Use the minimum memory configuration without measuring how long the function runs.
- Choose maximum memory for every function irrespective of diminishing speed gains.
- Increase S3 object lifecycle days as the primary measure of Lambda compute price.
- Benchmark Lambda billed duration versus configured memory and request volume while meeting latency goals.
Explanation
Correct Answer
D:Lambda allocates CPU along with memory, so a higher-memory run can consume fewer GB-seconds if much faster.
Incorrect Answers
A:Per-duration billing means a slower small allocation is not automatically cheapest.
B:Extra allocated memory can raise cost if duration does not improve enough.
C:Storage retention is separate from per-invocation compute billing.
Question 897
A public checkout Lambda requires very low cold-start latency, so operations considers provisioned concurrency. Finance expects to pay only for actual invocation time. What charge must be included?
- Include provisioned-concurrency allocated idle duration and associated invocation costs in the latency-versus-cost model.
- Treat provisioned concurrency as free until the first customer invocation begins.
- Assume reserved concurrency and provisioned concurrency are identical pricing and latency features.
- Model only S3 data transfer cost as if it replaces Lambda capacity charges.
Explanation
Correct Answer
A:Preinitialized Lambda capacity has a separate billed reservation component even when it is waiting for requests.
Incorrect Answers
B:Keeping execution environments ready incurs billed capacity over time.
C:One partitions concurrency limits; the other maintains warm initialized environments.
D:Storage/network spending does not account for warm execution allocation.
Question 898
A startup runs several APIs in ECS. Demand is variable and the team has no infrastructure staff to patch EC2 worker nodes. It is willing to pay a task premium for operational simplicity. Which compute model should it consider?
- Choose ECS on self-managed EC2 solely because it has no worker patching or capacity management responsibility.
- Use ECS on Fargate with task sizing and service autoscaling matched to demand.
- Choose dedicated GPU hosts for all APIs although no accelerator processing is required.
- Keep maximum Fargate desired task count permanently instead of scaling down with demand.
Explanation
Correct Answer
B:Fargate removes customer management of worker instance fleets and bills provisioned task resources.
Incorrect Answers
A:EC2-backed ECS requires a worker fleet lifecycle that the team wants to avoid.
C:Specialized always-on hardware would not match the described variable web services.
D:Serverless task operations do not remove task resource charges while running.
Question 899
A mature enterprise runs hundreds of container tasks at nearly full utilization around the clock. It already operates a reliable EC2/ECS worker fleet and can pack tasks densely. What option may reduce per-task spend?
- Assume Fargate always costs less than a dense EC2 fleet at high steady utilization.
- Create one new Fargate task per container for the whole year without considering reserved resources.
- Benchmark ECS on EC2 with efficient instance packing and eligible capacity commitments against the Fargate task bill.
- Use Lambda solely because it can execute any persistent container daemon for months within a single invocation.
Explanation
Correct Answer
C:When worker utilization is high and operations are already handled, EC2-backed task placement can be economical.
Incorrect Answers
A:Task pricing and EC2 instance amortization can favor different models.
B:Continuous provisioned task CPU/memory can become costly when densely packed worker instances would suffice.
D:AWS Lambda imposes a bounded maximum invocation duration and does not provide a perpetual container-daemon runtime for months inside one invocation. Long-running persistent processes need an appropriate container or instance service.
Question 900
A Fargate task requests 4 vCPUs and a large memory allocation, but metrics never exceed 0.5 vCPU or 2 GiB. The app has no special reserve-memory needs. What change targets resource waste?
- Lower the number of running ECS tasks with service scaling while retaining the oversized CPU and memory allocation of each remaining task.
- Set each task to 0.5 vCPU while keeping a 16-GiB memory allocation, without checking the Fargate sizing matrix.
- Purchase Compute Savings Plans only and keep the original oversized task definition unchanged.
- Benchmark supported smaller Fargate CPU-memory combinations and revise task sizing with safe peak headroom.
Explanation
Correct Answer
D:Fargate charges for the task’s configured vCPU and memory allocations, not just measured use. Load-test the smallest supported combination that accommodates real memory peaks, CPU burst demand and safe headroom before adjusting the task definition.
Incorrect Answers
A:Service scaling can reduce overall task count at low demand but does not address waste in the specified per-task 4-vCPU, high-memory resource shape.
B:Fargate enforces supported CPU-memory combinations; 0.5-vCPU Linux Fargate tasks do not support a 16-GiB memory setting.
C:Commitment discounts can lower eligible compute rates but do not correct overprovisioned per-task resources, nor remove resource waste at peak.
