AWS Solutions Architect Associate SAA-C03 Sign in Try 10 free questions

Free SAA-C03 practice questions: High-Performing Architectures

The domain where the storage and database choice decides the answer — seven scenarios with the reasoning attached. SAA-C03, 2026 edition.

Here are 7 typical high-performing-architecture questions in the exam's own style — free, with the answer explained, no sign-up.

Straight to the 7 free questions ↓

Design High-Performing Architectures is 24% of the SAA-C03 — about 12 of the 50 scored questions — and it has the clearest identity of the four domains: every question is a service-selection problem wearing a performance requirement. You are never asked whether a service works. You are asked which of four working services fits this workload.

Every option would run — one is right

That is the SAA signature, and this domain is where it shows most. A stem describes a workload; the four options are all genuine AWS services; each of them could be deployed without anything catching fire. What separates the correct answer is a constraint hiding in the wording: “sub-millisecond latency”, “shared file access from many instances”, “millions of requests per second”, “static IP addresses”, “not cacheable”. Train yourself to read for the constraint first and the service names second. Once you can point at the deciding phrase, the other three options stop being tempting.

The storage triangle is the most reliable discriminator

No single distinction resolves more questions in this domain than block versus file versus object — the mapping in the diagram above. Amazon EBS is block storage: a fast virtual disk attached to one EC2 instance, the home of databases and boot volumes. Amazon EFS is file storage: a shared POSIX file system that many instances mount at once, across Availability Zones, growing and shrinking on its own. Amazon S3 is object storage: objects addressed by key over HTTP at internet scale — static assets, backups, data lakes.

The access pattern in the stem picks the corner of the triangle. One instance needing a low-latency disk is EBS. Many instances needing the same files through standard file operations is EFS. Anything described as objects, assets or internet-facing content is S3. The exam will happily offer all three as options; the workload’s own words disqualify two of them.

Inside the EBS corner sits a second decision the exam loves: gp3 versus io2. gp3 is the general-purpose default and provisions up to 16,000 IOPS per volume; io2 exists for the workloads beyond that — tens of thousands of IOPS, sub-millisecond latency, higher durability. When a stem states an IOPS figure, do the arithmetic: above gp3’s ceiling, the answer is io2. The HDD types, st1 and sc1, are for cheap sequential throughput and are almost always distractors in a performance stem.

The database decision line

Databases in this domain separate along one line: what shape is the data, and how fast must a single lookup be? Tables, joins and SQL put you on the relational side — Amazon RDS, or Amazon Aurora when the stem adds demands stock engines strain under: higher throughput, more read replicas, faster failover. On the other side sits Amazon DynamoDB, and the exam signposts it with a phrase you should learn to recognise on sight: single-digit millisecond latency at any scale, usually next to “key-value” and “millions of requests”. That is DynamoDB’s own definition written into the stem. Milliseconds and relational features point left; single-digit milliseconds, key-value and extreme scale point right. Amazon Redshift appears in the options to check whether you know a data warehouse is for analytics, not operational lookups.

Read-heavy has a default answer: cache

When a stem says the same data is read over and over — a product catalog, a leaderboard, session state — and asks for the lowest possible latency, the exam’s default answer is Amazon ElastiCache. Memory beats disk by orders of magnitude, and a cache removes the repeated queries from the database rather than helping it endure them. Keep the boundary with read replicas sharp: a cache serves the same hot items at microsecond speed; replicas add capacity for varied read traffic at normal database speed. Scaling the instance vertically is the option that always lurks nearby, and it is almost never correct — it treats the symptom at full price.

The edge network has two products, not one

CloudFront and Global Accelerator both put your traffic onto AWS’s edge network, which is exactly why the exam pairs them. The split is clean: CloudFront is for cacheable HTTP content — static assets, video, APIs that benefit from caching. AWS Global Accelerator is for everything that is not cacheable: TCP and UDP flows that need static anycast IP addresses and the fastest path onto the AWS backbone. “Static IPs”, “UDP” or “not cacheable” in the stem eliminates CloudFront, no matter how global the audience sounds. And for the upload direction, remember the third edge product: S3 Transfer Acceleration, which routes long-distance uploads through the nearest edge location into a bucket using the standard S3 APIs.

The domain’s favourite wrong answer

Learn this one sentence: Multi-AZ improves availability, not performance. The classic Multi-AZ standby is a synchronous copy that serves no traffic — it exists to take over when the primary fails. Yet “enable Multi-AZ” appears in performance questions constantly, because it sounds like a second server you could put to work. When the problem is slow reads, the answer is a read replica or a cache; Multi-AZ answers a different question — what happens when an Availability Zone dies. Question 7 below is built on exactly this trap, and some version of it is waiting on the real exam.

How to use the questions below

Answer first, then read past the correct letter. In this domain the explanation of why the other three fail is the study material — each wrong option is a real service with a real job, just not this one. If you can name the phrase in the stem that eliminated it, you are practising the one skill this domain scores.

Storage decision diagram mapping workloads to AWS storage services in three columns, under the heading question: how is the data accessed? Amazon EBS, block storage — pick it when a single EC2 instance needs a fast dedicated disk: databases and boot volumes, with IOPS chosen per volume through gp3 or io2. Amazon EFS, file storage — pick it when many instances must mount one shared POSIX file system that grows and shrinks automatically. Amazon S3, object storage — pick it for objects addressed over HTTP at internet scale: static assets, backups and data lakes, with no file system semantics. The exam tell beneath: attached to one instance means EBS, shared file system means EFS, objects or static assets mean S3.

7 free SAA-C03 practice questions

Answers and explanations — no email wall
SAA-C03 Question 1 of 7

A media company runs a video processing application on a fleet of EC2 instances in an Auto Scaling group across three Availability Zones. Every instance must read and write the same set of project files concurrently, using standard file system operations. Which storage service should a solutions architect recommend?

Answer: B — Amazon EFS mounted on all instances

The stem contains the two words that decide it: shared and file system. Amazon EFS is the managed NFS file system that any number of instances — across Availability Zones — can mount at once with POSIX semantics. Why not the others: Multi-Attach is limited to io1/io2 volumes, works only within a single Availability Zone, and still needs a cluster-aware file system on top — gp3 does not support it at all. S3 is object storage: durable and internet-scale, but not a file system, and per-instance buckets would defeat sharing anyway. Instance store is local, ephemeral and private to each instance — the opposite of shared.

SAA-C03 Question 2 of 7

A financial services company is migrating a transactional database to EC2. The database requires 60,000 IOPS on a single volume with sub-millisecond latency, and the volume must be highly durable. Which EBS volume type meets these requirements?

Answer: B — Provisioned IOPS SSD (io2)

The number does the work: gp3 tops out at 16,000 IOPS per volume, so a 60,000-IOPS requirement forces you into Provisioned IOPS territory. io2 delivers tens of thousands of IOPS per volume with sub-millisecond latency and higher durability than gp3 — exactly the profile of a critical transactional database. Why not the others: gp3 is the right default for most workloads, which is what makes it a tempting distractor, but the stated IOPS figure disqualifies it. st1 and sc1 are HDD types built for cheap sequential throughput — big-file streaming and cold data — and cannot be used as high-IOPS transactional disks or even as boot volumes.

SAA-C03 Question 3 of 7

A gaming backend stores player session data as key-value pairs looked up by player ID. The service must handle millions of requests per second at peak and return responses with consistent single-digit millisecond latency at any scale. No joins or complex queries are needed. Which database should a solutions architect choose?

Answer: C — Amazon DynamoDB

The stem quotes DynamoDB's own definition back at you: key-value access, millions of requests per second, consistent single-digit millisecond latency at any scale, no joins. That combination is the DynamoDB signature, and the explicit absence of relational features removes the only reason to prefer SQL. Why not the others: RDS for MySQL is relational and scales writes vertically — it is not built for millions of key-value requests per second, and Multi-AZ adds availability, not speed. Aurora is much faster than stock MySQL but is still a relational engine solving a relational problem this workload does not have. Redshift is a data warehouse for analytics over large datasets, not a low-latency operational store.

SAA-C03 Question 4 of 7

An e-commerce site keeps its product catalog in Amazon RDS. The same few thousand popular products are read millions of times a day, and the database CPU is saturated by these repeated identical queries. The company wants microsecond-level read latency for this hot data with minimal application change. What should a solutions architect recommend?

Answer: A — Add Amazon ElastiCache in front of the database

Repeated identical reads of a small hot set is the textbook caching workload: ElastiCache serves those items from memory at microsecond latency and takes the repeated queries off the database entirely. On the exam, read-heavy plus same data plus lowest latency equals cache. Why not the others: a larger instance class buys headroom but keeps answering the same query from disk and buffer over and over — it treats the symptom at full price and hits the next ceiling later. Redshift is for analytical scans, not hot single-item lookups, and would be slower here, not faster. Storage autoscaling grows capacity in gigabytes; the bottleneck is CPU on repeated reads, not disk space.

SAA-C03 Question 5 of 7

A multiplayer game studio serves players worldwide over a custom UDP protocol. The traffic is not cacheable, the studio needs two static IP addresses that clients can allowlist, and player traffic should enter the AWS network at the nearest edge location for the lowest possible latency. Which service meets these requirements?

Answer: B — AWS Global Accelerator

Three requirements in the stem — UDP, static anycast IP addresses, non-cacheable traffic onto the AWS backbone at the edge — are Global Accelerator's exact feature list. Why not the others: CloudFront also uses the edge network, which is what makes it the classic distractor, but it is a content delivery network for cacheable HTTP content — it does not proxy arbitrary UDP and does not give you static IPs. Route 53 latency-based routing picks a Region at DNS resolution time, but the traffic then crosses the public internet and there are no fixed IPs to allowlist. Application Load Balancers are HTTP/HTTPS only and regional — they cannot terminate a global UDP flow.

SAA-C03 Question 6 of 7

Field teams on several continents upload multi-gigabyte survey files to a single Amazon S3 bucket in us-east-1. Uploads from distant locations are slow and unreliable. The company wants to speed up these long-distance uploads with a minimal change — the application should keep using standard S3 upload APIs. What should a solutions architect recommend?

Answer: A — Enable S3 Transfer Acceleration on the bucket and use the accelerated endpoint

S3 Transfer Acceleration is built for precisely this: clients upload to the nearest edge location and the data rides AWS's optimized backbone to the bucket — same PUT and multipart APIs, one endpoint change. Combine it with multipart upload and the long-haul problem largely disappears. Why not the others: Cross-Region Replication copies objects out of a bucket after they arrive — it does nothing for the slow inbound leg. CloudFront accelerates delivery of content to users; it is the download-direction sibling of the pattern the question needs in the upload direction. Snowball is a physical device for one-off bulk migrations measured in weeks, not a fix for ongoing daily uploads.

SAA-C03 Question 7 of 7

A production Amazon RDS for PostgreSQL database runs in a Multi-AZ deployment. The analytics team now runs heavy reporting queries against it during business hours, and application write latency has degraded. How should a solutions architect improve performance for both workloads?

Answer: B — Create a read replica and direct the reporting queries to it

Read replicas exist to scale reads: an asynchronous copy takes the reporting load, the primary keeps its capacity for application writes, and the two workloads stop competing. Why not the others: the Multi-AZ standby in a classic Multi-AZ deployment serves no traffic at all — it is a synchronous copy waiting for failover, so there is nothing to point the reporting tool at. Resizing that standby buys availability insurance, not read capacity; Multi-AZ is the domain's favourite wrong answer precisely because it sounds like a second server you could use. More IOPS speeds up the disk under a contended instance but leaves both workloads fighting for the same CPU and memory — it treats the wrong bottleneck.

That is exactly how every question in the course works — answer, explanation, why-not. The real set continues in the practice player: 10 free questions, no sign-up.

Continue with the 10 free questions →

Those 7 questions were the start.

The exam does not test whether you recognise a term — it tests whether you can rule out three plausible answers under time pressure. That is what the explanations above are for, and there are 780 more questions built exactly like them.

Collecting questions yourself

  • Scattered across forums, of unknown age
  • Answer keys without reasoning
  • No idea which domain you are weak in

Practising with a system

  • 780 questions in 12 full tests, SAA-C03 (2026)
  • Every option explained — including the wrong ones
  • Readiness per exam domain, and drills for your weakest
Start free — 10 questions

Straight into the player. No account, no email.

Frequently asked questions

How much of the SAA-C03 exam is this domain?
Design High-Performing Architectures carries 24% of the scored content according to the official SAA-C03 exam guide — roughly 12 of the 50 scored questions, making it the third-largest domain after Design Secure Architectures (30%) and Design Resilient Architectures (26%). Our simulator draws to these official weights, so a practice exam confronts you with about the same share.
Why do performance questions feel like service catalog questions?
Because that is what they are. The exam rarely asks how fast a service is in the abstract; it describes a workload — its access pattern, its latency target, its read/write mix — and asks which service fits. Every option offered is a real service that could plausibly be wired in. The skill being tested is matching workload characteristics to service characteristics: block versus file versus object, relational versus key-value, cache versus replica. Study the discriminators, not the spec sheets.
Are these real exam questions?
No. They are original questions we wrote for this page, in the style and difficulty of the exam. Real exam items are protected by the AWS certification agreement, and sharing or using them can cost a candidate their certification. Ours come with something a leaked item never has: the reasoning, including why the wrong answers are wrong.

Updated for SAA-C03 (July 2026). The sample questions above are our own work in the style of the exam — not real exam items. The exam itself is set and marked by the certification body.