← All modules

SAA3 — Design High-Performing Architectures

Why this module exists

Domain 3 is 24% of scored content, and it has more task statements than any other domain: five. It is also the domain where the exam leans hardest on numbers. A resilience question can often be answered from the shape of the requirement. A performance question often can't: "more than 16,000 IOPS", "a single file system shared by Windows servers", "sub-millisecond reads", "two static IP addresses" — each of those phrases eliminates options on a documented limit, not on judgement.

The five task statements, verbatim from the exam guide:

• 3.1 — "Determine high-performing and/or scalable storage solutions"
• 3.2 — "Design high-performing and elastic compute solutions"
• 3.3 — "Determine high-performing database solutions"
• 3.4 — "Determine high-performing and/or scalable network architectures"
• 3.5 — "Determine high-performing data ingestion and transformation solutions"

Notice the verb. Four of the five say determine. The exam is not asking you to design a storage system from scratch; it is asking you to pick the right one off the shelf from a stated requirement.

The one idea to hold onto

Every performance question is a bottleneck question. Find the resource that runs out first — then pick the service whose limit is above the requirement and whose cost isn't absurdly above it.

   THE STEM SAYS                          THE RESOURCE THAT RUNS OUT      LESSON
   ─────────────                          ──────────────────────────      ──────
   "IOPS", "latency", "throughput"    →   storage                     →   1
   "CPU-bound", "bursty", "batch"     →   compute                     →   2
   "read-heavy", "connections"        →   database                    →   3
   "global users", "on-premises link" →   network                     →   4
   "streaming", "terabytes nightly"   →   ingestion                   →   5
   "query the data lake", "Parquet"   →   transformation / analytics  →   6

AWS's own RDS documentation says it better than any summary: Provisioned IOPS "provides a way to reserve I/O capacity … However, as with any other system capacity attribute, its maximum throughput under load is constrained by the resource that is consumed first. That resource might be network bandwidth, CPU, memory, or database internal resources." That sentence is the whole domain.

What you'll be able to do

1. Choose between object, file and block storage, and between the EBS volume types, from an IOPS, throughput or durability number.
2. Say when instance store is right, and name the four events that erase it.
3. Pick an EC2 instance family from its name, size a Lambda function by memory, and choose between EC2 Auto Scaling, AWS Auto Scaling, Batch and Fargate.
4. Choose a database type and engine from the access pattern, and place read replicas, RDS Proxy, ElastiCache and DAX correctly — including where each one does not help.
5. Choose between CloudFront and Global Accelerator, between VPN and Direct Connect, and between a cluster, partition and spread placement group.
6. Size a Kinesis stream in shards, choose between Data Streams and Firehose, and pick the right transfer service for a hybrid or bulk migration.
7. Build a data lake answer: S3 + Glue Data Catalog + Lake Formation + Athena, with Parquet and partitions for performance.

Lessons

Where this sits in the exam

From the SAA-C03 exam guide — Content Domain 3 (verified 2026-09-25), Design High-Performing Architectures — 24% of scored content:

Every "Knowledge of" and "Skills in" item, mapped

Overlap with other modules — stated, not repeated

• SAA2 already teaches decoupling (SQS/SNS/EventBridge/Step Functions), Auto Scaling groups, the load balancer types and ECS/EKS/Fargate/Lambda. Task 3.2 lists several of the same items. Lesson 2 here adds only the performance angle — which metric to scale on, how big to make the thing — and points back to SAA2 for the rest.
• The quicksight/ track is the deep treatment of Amazon Quick. Lesson 6 links to it rather than compressing it.

A note on sources

This module cites the HTML exam guide at docs.aws.amazon.com/aws-certification/latest/solutions-architect-associate-03/, not the d1.awsstatic.com PDF — see ../PLAN.md for why.

Every number in this module comes from a page fetched 2026-09-25 and listed in the lesson's sources: frontmatter. Three things I found while re-fetching are worth knowing, because they contradict numbers commonly repeated in prep material:

1. Aurora throughput. The Aurora overview now says "up to 6x the throughput of stock PostgreSQL and up to 6x the throughput of stock MySQL". Older material quotes different multiples. Use the page.
2. Snowball Edge. The Snowball Edge guide now opens with "AWS Snowball Edge is no longer available to new customers." The in-scope services list still names "AWS Snow Family", but Domain 3's task statements do not. Lesson 5 covers it in one paragraph and says so.
3. EFS Max I/O. The EFS performance page calls Max I/O "a previous generation performance type" and recommends General Purpose "for all file systems". An answer choosing Max I/O for performance is now choosing the slower option.

Where a page I fetched did not state something examinable, the lesson says so inline with ⚠️ and names the page to read. The largest gaps: the per-engine maximum number of RDS read replicas, the Aurora Serverless capacity range, Fargate task sizes, FSx for NetApp ONTAP and OpenZFS, Amazon MSK and Amazon Redshift.

Facts verified 2026-09-25 against the pages cited in each lesson. Because a summarising fetch can paraphrase, every number and every italic-quoted phrase in the notes was then re-checked against the raw HTML text of its source page (downloaded with a script, not summarised); quotes that didn't match word-for-word were corrected.

Keeps playing into the following modules — 141 min from here to the end of certification prep.