AWS Certified Solutions Architect - Associate SAA-C03 Practice Test 68
Question 670
A DataSync task moves research data from an on-premises NAS to S3 each night. Auditors require checks that destination objects reflect source content, but engineers have disabled task verification to shorten runs. What should be configured?
- Rely only on an S3 bucket name match as proof that every file byte was copied correctly.
- Select an appropriate AWS DataSync verification strategy and schedule, and monitor task reports for transfer integrity.
- Compare S3 ETags for every transferred object and assume they are always an end-to-end checksum of the original NAS file, including multipart uploads.
- Disable all task logs and assume a successful start request proves the entire transfer completed.
Explanation
Correct Answer
B:DataSync supports verification options; validating transferred content helps expose mismatches that raw copy-count totals miss.
Incorrect Answers
A:A name match cannot establish content fidelity for thousands of transferred files.
C:An S3 ETag is not a universal source-file MD5 checksum, particularly for multipart objects or other configurations. DataSync verification modes validate transferred content using supported checks rather than assuming ETag equality proves every file.
D:Starting a task is not proof that all files passed transfer and verification.
Question 671
An on-premises DataSync agent synchronizes millions of files during a narrow WAN window. The destination S3 bucket has ample capacity, but transfers run far below the desired rate. Which performance analysis is relevant?
- Increase the S3 bucket's logical object-count quota by changing the bucket name.
- Disable checksumming and reduce all network visibility regardless of fidelity requirements.
- Measure the WAN/agent CPU and memory, file enumeration cost, task settings and concurrency before changing destination capacity.
- Provision a larger EC2 root disk for unrelated application hosts and assume the on-premises agent accelerates.
Explanation
Correct Answer
C:End-to-end DataSync speed depends on source, network, agent and task behavior, not just how large the destination storage can grow.
Incorrect Answers
A:S3 naming does not fix a saturated or poorly provisioned transfer agent.
B:Removing safety controls without diagnosing the bottleneck can compromise integrity without improving throughput.
D:Unrelated hosts do not change the transfer path or DataSync agent capacity.
Question 672
A simulation fleet runs in one Availability Zone and repeatedly reads and overwrites millions of small intermediate objects. Measurements show S3 request latency, not throughput or CPU, is the bottleneck. The objects can be regenerated and the team accepts single-AZ storage durability in return for consistent low-latency S3 API operations. Which design should it test?
- Keep all temporary data in a regional S3 Standard general-purpose bucket and increase its retention period, because longer object retention reduces GET and PUT latency.
- Move the hot temporary objects into S3 One Zone-Infrequent Access solely for its lower storage price, despite frequent operations and an explicit latency bottleneck.
- Attach one gp3 EBS volume to a single worker and treat it as a directly shared S3-compatible object endpoint for all compute nodes in the AZ.
- Create an S3 Express One Zone directory bucket in the same supported AZ as the compute, using its zonal data endpoints and compatible SDK/API operations.
Explanation
Correct Answer
D:S3 Express One Zone is designed for high-throughput, low-latency operations from directory buckets in a selected AZ. Colocating compute and the directory bucket suits ephemeral data when the single-zone tradeoff is explicitly acceptable.
Incorrect Answers
A:S3 Standard provides regional resilience and broad compatibility, but increasing retention does not address measured single-object request latency. The workload specifically accepts a zonal high-performance option.
B:S3 One Zone-IA targets infrequently accessed objects and can introduce retrieval charges. Its cost profile is not the specialized low-latency, high-operation design required here.
C:An EBS gp3 volume is block storage attached to supported compute, not a managed S3 object API shared across the fleet. This would require a separate service layer and failure design.
Question 673
A fleet compresses logs, calculates checksums, and runs deterministic simulations. During load tests CPU is consistently above 95%, but memory is mostly unused and network traffic is modest. Which EC2 family best matches the first right-sizing experiment?
- Benchmark a current compute-optimized C-family instance against the present host.
- Use R-family memory-optimized instances because high CPU by itself proves a need for a very large RAM-to-vCPU ratio.
- Choose memory-bound high-memory instance types before measuring actual memory allocation.
- Use storage-optimized I-family instances solely to fix high CPU when local I/O is already idle.
Explanation
Correct Answer
A:C-family capacity emphasizes compute relative to memory for CPU-bound workloads, although measured price/performance must confirm fit.
Incorrect Answers
B:An R-family choice targets memory pressure not evidenced by the profiling results.
C:Paying for abundant unused RAM is unlikely to resolve the observed CPU bottleneck economically.
D:Storage-optimized hosts address local storage I/O, not a compute-saturated workload by default.
Question 674
An in-memory analytics engine maintains a 280-GiB working set per node. CPU use is moderate, but cache eviction and swapping cause query latency spikes. Which EC2 capacity characteristic should drive the replacement?
- Select C-family instances solely by their CPU-to-price ratio regardless of their available memory.
- Evaluate an R-family memory-optimized instance with sufficient RAM to hold the working set plus overhead.
- Choose only the cheapest burstable T-family size without measuring sustained RAM requirements.
- Use GPU instances exclusively because all in-memory key-value queries benefit from graphics acceleration.
Explanation
Correct Answer
B:Memory-optimized ratios target large resident datasets and reduce forced eviction when CPU is not the limiting resource.
Incorrect Answers
A:Insufficient memory persists even on a fast CPU if the dataset cannot remain resident.
C:Burst behavior does not compensate for a large fixed memory footprint.
D:A GPU is not a substitute for adequate main-memory capacity in conventional in-memory analytics.
Question 675
A conventional Java API uses CPU and memory in approximately equal proportions. Profiling shows no special GPU, local NVMe, or exceptional-memory need. What is the reasonable starting EC2 class?
- Choose GPU-based P-family instances because every API request contains calculations.
- Buy a maximum-RAM high-memory instance even though memory pressure is low.
- Benchmark a current M-family general-purpose instance and right-size from measured latency and utilization.
- Select storage-optimized instances based solely on total S3 object count.
Explanation
Correct Answer
C:Balanced general-purpose compute/memory ratios suit mixed web workloads without evidence for a specialized hardware family.
Incorrect Answers
A:Ordinary Java web computations do not imply a GPU-accelerated model workload.
B:Unused capacity raises cost without solving the balanced service's observed demands.
D:S3 object inventory is not a measure of the API host's local storage I/O pressure.
Question 676
Two EC2 instance families offer the same vCPU count, but one runs the application's video codec much faster per dollar and uses a different architecture. How should the architect make a choice?
- Choose the instance with the alphabetically earlier family name because AWS vCPUs always yield equal performance.
- Compare only the listed number of vCPUs and ignore runtime cost and software compatibility.
- Assume the largest local NVMe storage capacity automatically produces the fastest CPU codec encode.
- Benchmark representative jobs for throughput, response time and total cost, including architecture compatibility.
Explanation
Correct Answer
D:Equal vCPU counts do not imply equal clock, memory, architecture, price or effective workload throughput.
Incorrect Answers
A:Instance naming order has no measurable relationship to the codec's throughput or price.
B:Processor generations, instruction sets and per-core behavior can materially change real performance.
C:The profiler indicates a compute-intensive codec; local-disk capacity is not a proxy for encode throughput.
Question 677
An x86-only commercial application is baked into an AMI with proprietary native libraries. Engineers propose migrating its instances to AWS Graviton using the unchanged AMI and binary. What must they verify?
- Confirm arm64-compatible operating-system images and native binaries or rebuild the application/dependencies for ARM.
- Change only the instance type to Graviton and assume an existing x86 AMI boots and executes identical native code.
- Add an EBS throughput setting to the x86 AMI because EBS converts CPU instructions between architectures.
- Change only the application load balancer to NLB to make its x86 shared libraries arm64 compatible.
Explanation
Correct Answer
A:Graviton uses the ARM architecture; x86-only machine code cannot execute natively on arm64 without adaptation.
Incorrect Answers
B:Cross-architecture images and binaries are not automatically interchangeable.
C:Storage configuration does not implement processor instruction-set translation.
D:Load-balancer choice does not recompile a native library or resolve processor ABI differences.
Question 678
An ECS microservice runs on both x86 EC2 tasks and Graviton arm64 compute during phased migration. A single architecture-specific container tag causes illegal-instruction failures in one fleet. Which build artifact approach fixes distribution?
- Build only an x86_64 image and assume the container runtime translates its native binaries on ARM hardware.
- Publish a tested multi-architecture container manifest with matching x86_64 and arm64 images, or select explicit compatible digests.
- Build only an arm64 binary and use the same digest on x86 without a compatibility layer.
- Pin the image to latest with no architecture metadata and assume this guarantees compatibility.
Explanation
Correct Answer
B:Container manifests can select the appropriate image architecture while preserving a common logical application version.
Incorrect Answers
A:Ordinary containers share the host kernel and CPU architecture; they do not universally emulate foreign instructions.
C:The reverse architecture mismatch is still a native execution problem.
D:An unqualified mutable tag does not provide tested platform-specific image artifacts.
Question 679
A stateless C++ service appears 25% cheaper on Graviton list prices, but a required cryptography library is compiled for x86 and throughput per connection is unknown. What evidence should precede mass migration?
- Migrate the full fleet without running the library or workload because hourly price alone proves value.
- Recompile the service entry point for Graviton while keeping the x86-only cryptography shared library unchanged in the deployment image.
- Run representative performance and compatibility tests on arm64 and compare cost per completed operation.
- Use x86 emulation for the required cryptography library on Graviton and project savings using unchanged native x86 throughput assumptions.
Explanation
Correct Answer
C:The combination of native dependencies, workload throughput and price determines actual architecture economics.
Incorrect Answers
A:An incompatible library could prevent service startup or erase the price advantage with reduced throughput.
B:The application and its native dependencies must be built for a compatible ARM64 instruction set or replaced. Recompiling only the entry point does not make an incompatible x86 binary load or execute correctly.
D:Emulation adds compatibility and potential CPU overhead and can introduce unsupported behaviors. Without benchmarking the actual cryptography path, the apparent hourly discount cannot establish lower cost per completed connection.
Question 680
A batch worker needs two extra instances when queue depth exceeds 10,000, but six extra instances when it exceeds 100,000. The required increments differ at explicit thresholds. Which Auto Scaling policy style fits?
- Use a single scheduled scale action at midnight regardless of queue depth.
- Use only a lifecycle hook with no scaling policy and assume it changes desired capacity.
- Raise target-tracking CPU utilization to 100% and assume it implements specific depth-to-instance increments.
- Configure step scaling with distinct CloudWatch threshold ranges and capacity adjustments.
Explanation
Correct Answer
D:Step scaling supports threshold-dependent adjustment sizes rather than one proportional utilization target.
Incorrect Answers
A:A fixed schedule does not react to these unpredictable backlog thresholds.
B:Lifecycle hooks manage the transition of instances, not the conditions that decide capacity changes.
C:CPU target tracking does not express the stated multistep queue-depth scale increments.
