AWS SAA · Domain 3 of 4

AWS SAA Domain 3: Design High-Performing Architectures

This domain covers choosing the fastest, most scalable AWS services for each need: storage, compute, databases, networking and data ingestion, matching the right service to the workload's performance profile.

Exam format
24%
Official task statements
5
Assessment estimate
~12
Suggested study time estimate
25–35 hrs

About 11 min read Reviewed October 2026

Overview

Domain 3 covers designing for performance and scalability across five task statements: high-performing storage, elastic compute, high-performing databases, scalable networking, and data ingestion and transformation. It covers choosing among storage types (object, block, file), compute options (EC2, containers, serverless), database engines (relational, key-value, in-memory, purpose-built), networking and content delivery (CloudFront, global accelerators, placement), and ingestion and streaming services (Kinesis, data pipelines).

At 24%, this domain tests service selection: given a workload's access pattern, throughput and latency needs, pick the service that performs best. The exam rewards matching the storage, compute or database type to the workload rather than defaulting to one option, and using caching and content delivery to improve performance. These choices sit within the secure, resilient designs from the earlier domains.

How it's tested

SAA-C03 questions describe a workload's performance and scaling needs and ask which service BEST fits. In this domain they give access patterns, throughput, latency or scale clues, and reward matching the right storage, compute, database, network or ingestion service rather than a default choice.

  • Match the storage type

    An access pattern is described, and you choose object, block or file storage, or the right S3 or EBS option.

  • Pick the compute model

    A workload profile is described, and you choose EC2, containers or serverless for the best performance and scaling.

  • Add caching or delivery

    A latency or read-heavy pattern is described, and you add ElastiCache, CloudFront or read replicas.

Reasoning it through

Several EC2 instances must share the same file data with concurrent read and write access. EBS is tempting, but an EBS volume attaches to one instance at a time, so it cannot serve shared concurrent access. Amazon EFS is a managed, elastic file system that many instances can mount simultaneously, which matches the shared-access requirement and scales automatically.

Task statements 5

Official task statements from the Amazon Web Services outline, in outline order.

3.1 Determine high-performing and/or scalable storage solutionsObject, block and file storage, and choosing the right S3 and EBS options.
What it means
This covers choosing among object storage (S3), block storage (EBS), and file storage (EFS, FSx), selecting S3 storage classes and EBS volume types for performance, and using hybrid storage such as Storage Gateway. It matches storage to access patterns, throughput and latency needs.
What to understand
S3 is object storage for large-scale, durable data accessed over HTTP. EBS is block storage attached to a single instance. EFS is a shared file system many instances can mount. FSx provides managed file systems for specific workloads. EBS volume types and S3 classes trade performance against cost.
How it shows up
An access pattern is described, and you must choose object, block or file storage and the right type for the performance need.
Common confusion
Confusing EBS with EFS; EBS attaches to one instance, while EFS supports shared, concurrent access from many instances.
3.2 Design high-performing and elastic compute solutionsEC2, containers and serverless, with auto scaling for elasticity.
What it means
This covers choosing compute options: EC2 instance families, containers with ECS, EKS and Fargate, and serverless with Lambda, plus auto scaling and load balancing for elasticity. It matches the compute model to the workload's performance and operational profile.
What to understand
EC2 gives full control and suits steady or specialized workloads. Containers package and scale applications, with Fargate removing server management. Serverless Lambda scales automatically and suits event-driven, variable workloads. Choosing the right instance family and scaling approach drives performance.
How it shows up
A workload profile is described, and you must choose EC2, containers or serverless and the scaling approach that performs best.
Common confusion
Defaulting to EC2 for event-driven or bursty workloads where serverless scales better with less management.
3.3 Determine high-performing database solutionsRelational, key-value, in-memory and purpose-built databases, with caching and replicas.
What it means
This covers choosing database engines: relational (RDS, Aurora), key-value (DynamoDB), in-memory (ElastiCache), and purpose-built databases, plus read replicas and caching for read-heavy workloads. It matches the database to the data model and access pattern.
What to understand
Relational databases suit structured data and complex queries. DynamoDB is a key-value store that scales to very high throughput with low latency. ElastiCache caches hot data in memory to cut latency and database load. Read replicas scale read traffic. The access pattern and scale point to the engine.
How it shows up
A data model and access pattern are described, and you must choose the database engine and whether to add caching or replicas.
Common confusion
Choosing a relational database for a high-throughput key-value workload that DynamoDB would serve better.
3.4 Determine high-performing and/or scalable network architecturesContent delivery, global acceleration, and network design for performance.
What it means
This covers Amazon CloudFront for content delivery, AWS Global Accelerator for network performance, Route 53 routing policies, placement groups, enhanced networking, and VPC design choices that affect performance and scale.
What to understand
CloudFront caches content at edge locations close to users, reducing latency for static and dynamic content. Global Accelerator improves performance for global applications by routing over the AWS network. Route 53 routing policies direct users to the best endpoint. Placement groups and enhanced networking boost intra-VPC performance.
How it shows up
A latency or global-performance need is described, and you must choose CloudFront, Global Accelerator or the right routing or placement option.
Common confusion
Confusing CloudFront with Global Accelerator; CloudFront caches content at the edge, while Global Accelerator optimizes network paths to endpoints.
3.5 Determine high-performing data ingestion and transformation solutionsStreaming and batch ingestion, and transformation services.
What it means
This covers ingesting and transforming data at scale with services such as Amazon Kinesis for streaming, AWS Glue for extract, transform and load, Amazon Data Firehose for delivery, and batch processing options. It matches the ingestion pattern, streaming or batch, to the right service.
What to understand
Streaming ingestion handles continuous, real-time data with Kinesis. Firehose delivers streaming data to stores such as S3. Glue performs serverless extract, transform and load for batch processing. Choosing streaming versus batch and the matching service is the key decision.
How it shows up
A data ingestion pattern is described, and you must choose streaming or batch and the service that fits.
Common confusion
Confusing streaming with batch ingestion; real-time continuous data suits Kinesis, while periodic bulk processing suits batch ETL.

High-yield concepts

The terms, rules and formulas this domain turns into questions most often.

  • Object vs. block vs. file storage

    S3 (object), EBS (block), EFS and FSx (file).

    Remember
    EBS attaches to one instance; EFS supports shared access.
    Tested as
    Storage selection questions.
    Trap
    Using EBS where shared file access is needed.

  • S3 storage classes

    Tiers trading access frequency and cost, from Standard to Glacier.

    Remember
    Match the class to access frequency and retrieval needs.
    Tested as
    Storage performance and cost questions.
    Trap
    Using Standard for rarely accessed data.

  • EBS volume types

    SSD and HDD types balancing IOPS, throughput and cost.

    Remember
    Provisioned IOPS SSD for high, consistent IOPS.
    Tested as
    Block storage questions.
    Trap
    Choosing the wrong volume type for the IOPS need.

  • Serverless vs. containers vs. EC2

    Lambda scales automatically, containers package apps, EC2 gives full control.

    Remember
    Serverless suits event-driven, variable workloads.
    Tested as
    Compute selection questions.
    Trap
    Defaulting to EC2 for bursty event-driven work.

  • DynamoDB

    A key-value and document database with very high throughput and low latency.

    Remember
    It scales beyond relational for key-value access.
    Tested as
    Database selection questions.
    Trap
    Using relational for a high-throughput key-value workload.

  • ElastiCache

    In-memory caching to reduce latency and database load.

    Remember
    Cache hot data for read-heavy workloads.
    Tested as
    Caching questions.
    Trap
    Scaling the database instead of caching reads.

  • Read replicas

    Copies that scale read traffic for relational databases.

    Remember
    They offload reads, not writes.
    Tested as
    Database performance questions.
    Trap
    Expecting replicas to scale writes.

  • CloudFront

    A content delivery network caching content at edge locations.

    Remember
    It reduces latency for users near the edge.
    Tested as
    Content delivery questions.
    Trap
    Confusing it with Global Accelerator.

  • Global Accelerator

    Routes traffic over the AWS network to the best endpoint.

    Remember
    It optimizes network paths, not content caching.
    Tested as
    Global performance questions.
    Trap
    Confusing it with CloudFront.

  • Kinesis

    A service for ingesting and processing streaming data in real time.

    Remember
    Use it for continuous, real-time ingestion.
    Tested as
    Ingestion questions.
    Trap
    Using batch ETL for real-time streams.

  • AWS Glue

    Serverless extract, transform and load for batch data.

    Remember
    Use it for batch ETL and cataloging.
    Tested as
    Transformation questions.
    Trap
    Using it for real-time streaming.

Side by side

Concepts candidates mix up, compared on the point that decides the answer.

Object vs. block vs. file storage

ConceptMeaningKey differenceCommon trap
S3 (object) Durable object storage over HTTP Massive scale, not a mounted file system Treating it as a file system
EBS (block) Block volume attached to one instance Single-instance, like a disk Expecting shared access
EFS / FSx (file) Managed shared file systems Many instances mount concurrently Using EBS where EFS fits

Relational vs. DynamoDB vs. ElastiCache

ConceptMeaningKey differenceCommon trap
Relational (RDS/Aurora) Structured data and complex queries Strong consistency and joins Using it for extreme key-value throughput
DynamoDB Key-value and document store Very high throughput, low latency Expecting relational joins
ElastiCache In-memory cache Lowest latency for hot data Treating it as a primary database

CloudFront vs. Global Accelerator

ConceptMeaningKey differenceCommon trap
CloudFront Caches content at edge locations Content delivery and caching Using it purely for network path optimization
Global Accelerator Optimizes network paths to endpoints Improves network performance, no caching Expecting it to cache content

Exam traps

Each trap is an answer that looks right. Learn the clue that should change your mind.

  1. EBS for shared files

    What candidates pick
    Attach an EBS volume to share files across several EC2 instances.
    Why it looks right
    EBS is the familiar block storage.
    Why it's wrong
    An EBS volume attaches to one instance at a time. Amazon EFS provides shared, concurrent file access.

    Clue to watch for Multiple instances need concurrent read and write access to the same files.

  2. Relational for key-value

    What candidates pick
    Use RDS for a workload needing massive key-value throughput.
    Why it looks right
    Relational databases are familiar.
    Why it's wrong
    DynamoDB scales to very high throughput with low latency for key-value access, where relational would struggle.

    Clue to watch for The access pattern is simple key lookups at very high scale.

  3. Scale the database for reads

    What candidates pick
    Increase the database size to handle a read-heavy load.
    Why it looks right
    A bigger database seems to handle more.
    Why it's wrong
    ElastiCache or read replicas offload reads far more efficiently than scaling the primary.

    Clue to watch for The workload is read-heavy with repeated queries.

  4. CloudFront is Global Accelerator

    What candidates pick
    Use CloudFront purely to optimize network paths to a non-cacheable endpoint.
    Why it looks right
    Both improve global performance.
    Why it's wrong
    CloudFront caches content at the edge; Global Accelerator optimizes network paths without caching.

    Clue to watch for The need is network path optimization for non-cacheable traffic.

  5. EC2 for bursty events

    What candidates pick
    Run a bursty, event-driven workload on always-on EC2 instances.
    Why it looks right
    EC2 is a default compute choice.
    Why it's wrong
    Lambda scales automatically for event-driven, variable workloads with no idle servers.

    Clue to watch for The workload is event-driven, intermittent or unpredictable.

  6. Batch for real-time

    What candidates pick
    Use batch ETL for continuous real-time data.
    Why it looks right
    Batch tools are familiar.
    Why it's wrong
    Real-time streaming suits Kinesis; batch ETL introduces delay.

    Clue to watch for The data is continuous and must be processed in real time.

Answer framework

For any Domain 3 performance scenario, use these steps to find the answer SAA-C03 expects.

  1. Read the workload's access pattern, throughput, latency and scale clues before choosing any service.
  2. For storage, match object, block or file to the pattern, using EFS for shared access and the right S3 class or EBS type for performance.
  3. For compute, choose serverless for event-driven variability, containers for packaged scaling, or EC2 for steady or specialized needs.
  4. For databases, match the engine to the data model, using DynamoDB for key-value scale and adding ElastiCache or read replicas for read-heavy loads.
  5. For networking and ingestion, use CloudFront for caching, Global Accelerator for path optimization, and streaming versus batch services to match the data pattern.

Practice question

A fleet of EC2 instances in an Auto Scaling group must all read from and write to the same set of files concurrently, with the storage scaling automatically. Which AWS storage service BEST meets this requirement?

Want more like this? The AWS SAA practice test has questions for every domain.

Checklist

0 of 10

Tick each one you can do without notes. Your ticks stay in this browser.

Study strategy

This domain is about service selection, so study by access pattern rather than by memorizing service lists. For storage, build a mental table of object, block and file with the clue that signals each, especially shared concurrent access pointing to EFS. For compute, learn when serverless, containers or EC2 win, and stop defaulting to EC2.

For databases, practice reading the data model and access pattern: structured and relational versus key-value scale with DynamoDB, and read-heavy loads calling for ElastiCache or read replicas. For networking, keep CloudFront (caching) and Global Accelerator (path optimization) distinct. For ingestion, separate real-time streaming with Kinesis from batch ETL with Glue.

Hands-on time with S3, EFS, DynamoDB and CloudFront in a free-tier account makes the trade-offs concrete. Afterwards, move to Domain 4, cost-optimized architectures, which revisits these same service choices through a cost lens. Hands-on practice in a free-tier AWS account cements these choices far faster than reading, and reviewing missed practice questions by domain shows where to focus next.

FAQ

How much of SAA-C03 is Domain 3?

Design High-Performing Architectures is 24% of the scored content.

What is the difference between EBS and EFS?

EBS is block storage that attaches to a single instance at a time. EFS is a shared file system that many instances can mount concurrently, scaling automatically.

When should I use DynamoDB instead of a relational database?

For key-value or document access at very high throughput and low latency, where a relational database would struggle to scale.

What is the difference between CloudFront and Global Accelerator?

CloudFront caches content at edge locations to reduce latency. Global Accelerator optimizes the network path to your endpoints over the AWS backbone without caching content.

How do I improve performance for a read-heavy database?

Add ElastiCache to serve hot data from memory, or read replicas to offload read traffic, rather than scaling the primary database.

When should I choose serverless compute?

For event-driven, intermittent or unpredictable workloads, where Lambda scales automatically with no idle servers to manage.

How does Domain 3 connect to the rest of the exam?

Performance choices sit within the secure, resilient designs of Domains 1 and 2, and they trade off against cost, which Domain 4 addresses.

Other Cloud Computing certifications

Where the same subject shows up in other exams. Cards marked "Same subject" open the matching domain.

Test your Domain 3 knowledge

You have covered high-performing storage, compute, databases, networking and ingestion. Test whether you can match the right service to a workload's performance profile under exam conditions.

Take the AWS SAA practice test Check your AWS SAA readiness

Keep going