I cleared my AWS Certified Solutions Architect - Associate (SAA-C03) exam on the first attempt, so I thought I could help others with this article, where I list some of the most common patterns in these exams and which option to choose in each case. Prepare well for the exam, and use this article when you have to revise all the concepts in one go. I hope it'll help you all. Cheers!
Each row reads keyword in the question → correct answer / service. I've grouped them by topic so related services sit next to each other, which makes the "which one of these four similar options?" questions easier to tell apart.
Security
| Keyword in the question | Answer |
|---|
| Restrict access to some countries | CloudFront (Geo Restriction) |
| Protection against SQL injection or cross-site scripting / protect your web app and API | AWS WAF |
| DDoS attacks | AWS Shield |
| Credentials in a database | AWS Secrets Manager |
| Monitor and safeguard the organization from cyberattacks (threat detection) | Amazon GuardDuty |
| Vulnerability in your application | Amazon Inspector |
| Temporary ability for one service to access another service securely | IAM Roles |
| Encrypt data during transit | SSL/TLS or client-side encryption |
S3 encryption: who manages which key?
| Keyword in the question | Answer |
|---|
| AWS manages both the data key and the master key | Server-side encryption with Amazon S3 managed keys (SSE-S3) |
| AWS manages the data key, the customer manages the master key | Server-side encryption with AWS KMS managed keys (SSE-KMS) |
| The customer manages both the data key and the master key | Server-side encryption with customer-provided keys (SSE-C) |
S3 and Storage
| Keyword in the question | Answer |
|---|
| Preventing accidental deletion of objects in S3 | Enable Versioning + MFA Delete |
| Prevention of any type of modification of objects | S3 Object Lock |
| Hosting static content | S3 static website hosting |
| Low latency + static hosting | CloudFront + S3 static hosting |
| Restrict access to S3 so it's only reachable through CloudFront | Origin Access Control (OAC), the newer replacement for Origin Access Identity (OAI) |
| Faster uploads/downloads over long distances | S3 Transfer Acceleration |
| Uploading very large files | Multipart upload (recommended above 100 MB, required above 5 GB) |
| Replicate S3 data from one region to another | S3 Cross-Region Replication |
| Querying data stored in S3 with SQL | S3 + Athena |
S3 storage classes
| Keyword in the question | Answer |
|---|
| Frequent access and high performance | S3 Standard |
| Unpredictable or changing access pattern | S3 Intelligent-Tiering |
| Infrequent access but fast retrieval | S3 Standard-IA (S3 One Zone-IA if high availability isn't mentioned and cost must be low) |
| Archive, but still needs millisecond retrieval | S3 Glacier Instant Retrieval |
| Archive, retrieval in minutes to hours is fine | S3 Glacier Flexible Retrieval |
| Preserve for a long time + slow retrieval is fine (hours) | S3 Glacier Deep Archive |
File and block storage
| Keyword in the question | Answer |
|---|
| File system shared by multiple EC2 instances / hierarchical directory structure / Linux file server | Amazon EFS |
| Windows file server / SMB file share | Amazon FSx for Windows File Server |
| HPC for Linux (high-performance computing, e.g. gaming or ML workloads) | Amazon FSx for Lustre |
| Low latency + temporary storage for EC2 instances | Instance store |
| EBS volume for consistent, very low latency (sub-millisecond) | Provisioned IOPS SSD (io1 / io2) |
| AWS Storage Gateway File Gateway | NFS or SMB |
Compute and Scaling
| Keyword in the question | Answer |
|---|
| Preserve the in-memory contents of an instance | EC2 Hibernate |
| Serverless, scalable, and elastic architecture for an API | API Gateway + Lambda |
| Work finishes in under 15 minutes | AWS Lambda (max 15 minutes per run, up to 10 GB of memory) |
| Low latency between EC2 instances for HPC | Cluster Placement Group |
| Expected spike in traffic / predictable load on certain days | EC2 Auto Scaling scheduled scaling or predictive scaling |
| Scale on a metric with a target value (like CPU at 40%) | Target tracking scaling |
| High availability (in almost all cases) | Multi-AZ deployment |
Networking and Content Delivery
| Keyword in the question | Answer |
|---|
| Accessing DynamoDB or S3 from a private subnet within the AWS network | VPC endpoint (a gateway endpoint for S3 and DynamoDB) |
| Handle requests with low latency for TCP / UDP | Network Load Balancer (NLB) |
| Path- or host-based routing with HTTP/HTTPS traffic | Application Load Balancer (ALB) |
| Global audience with TCP-based connections / reduce latency for global users (TCP/UDP, static IPs) | AWS Global Accelerator |
| Instance-level firewall rules | Security Groups (stateful, allow rules only) |
| Subnet-level firewall rules, or blocking a specific IP address | Network ACLs (stateless, support deny rules) |
| Low-cost, immediate on-premises to AWS connection without large bandwidth | Site-to-Site VPN |
| Dedicated private link from an on-premises data center to AWS | AWS Direct Connect |
Route 53 routing policies
| Keyword in the question | Answer |
|---|
| Route traffic based on the user's location (country or continent) | Geolocation routing |
| Route based on the location of users and resources, and shift more or less traffic to a region (bias) | Geoproximity routing |
| Send users to the region with the lowest latency for them | Latency-based routing |
| Send traffic to a backup region when the primary is unhealthy | Failover routing |
Databases and Analytics
| Keyword in the question | Answer |
|---|
| Increase read performance in RDS | Read Replicas |
| NoSQL database with high performance / unclear database schema / key-value based | Amazon DynamoDB |
| Structured data with relationships, transactions, ACID compliance | Amazon RDS (Amazon Aurora if high performance is mentioned) |
| SQL database deployed across multiple AWS regions | Aurora Global Database |
| Flexible JSON document storage, schema-less design (MongoDB-compatible) | Amazon DocumentDB |
| Complicated analytical queries using joins (data warehouse) | Amazon Redshift |
| ETL (extract, transform, load) jobs / convert .csv files to Parquet | AWS Glue |
Messaging and streaming
| Keyword in the question | Answer |
|---|
| Decoupling apps / asynchronous processing | Amazon SQS |
| Duplicate processing of messages in the queue | Increase the visibility timeout (use a FIFO queue if the question needs exactly-once processing and strict order) |
| Real-time ingestion of streaming data (logs, IoT, clickstream, event data) | Kinesis Data Streams |
| Delivering streaming data to destinations like S3 or Redshift | Kinesis Data Firehose (now called Amazon Data Firehose) |
| Real-time analytics on streaming data using SQL or Apache Flink | Kinesis Data Analytics (now Amazon Managed Service for Apache Flink) |
Migration and Data Transfer
| Keyword in the question | Answer |
|---|
| Migrating an NFS or SMB server from on-premises to AWS | AWS DataSync (or AWS Storage Gateway for hybrid cloud use cases) |
| Large-scale offline data migration (10 TB to petabytes) | AWS Snowball |
| Transferring exabytes (100+ PB) of data | AWS Snowmobile (retired by AWS in 2024, so newer questions use multiple Snowball devices instead) |
Monitoring and Governance
| Keyword in the question | Answer |
|---|
| Track resource configurations and how they change over time / check compliance | AWS Config |
| Who made an API call or changed a resource (audit trail) | AWS CloudTrail |
| Centralized management and governance of many accounts | AWS Organizations (with Service Control Policies) |
AI Services
| Keyword in the question | Answer |
|---|
| Image and video analysis (face detection, object recognition) | Amazon Rekognition |
| Natural language processing: sentiment analysis, key phrase extraction (what's written in the text) | Amazon Comprehend |
| Text-to-speech conversion | Amazon Polly |
| Extract text and data from documents (OCR + table extraction) | Amazon Textract |
Quick tie-breakers
When two options both look right, these usually decide it:
- "Least operational overhead" points to the managed or serverless option (Lambda over EC2, Fargate over self-managed containers, Aurora over self-managed MySQL).
- "Most cost-effective" points to the cheaper storage class or pricing model that still meets every stated requirement, and nothing more.
- "Highly available" means Multi-AZ. "Disaster recovery across regions" means cross-region replication or a global database.
- "Decouple" almost always means SQS (or SNS for fan-out to many subscribers).
Good luck with the exam!