Overview
Designing resilient, cost-optimized, and highly available architectures on Amazon Web Services (AWS) requires an understanding of how foundational services interact across multiple Availability Zones (AZs).
This study guide summarizes the high-priority architectural patterns, storage classes, and database trade-offs needed for the AWS Solutions Architect exam and real-world team implementations.
High Availability VPC Architecture
A production-grade VPC is split across at least two Availability Zones with dedicated public and private subnets.
Internet
│
┌─────▼─────────────────────────┐
│ Internet Gateway (IGW) │
└─────┬───────────────────┬─────┘
│ │
┌───────▼────────┐ ┌───────▼────────┐
│ Public Subnet │ │ Public Subnet │
│ AZ-1a (NAT GW) │ │ AZ-1b (NAT GW) │
└───────┬────────┘ └───────┬────────┘
│ (Egress Traffic) │
┌───────▼────────┐ ┌───────▼────────┐
│ Private Subnet │ │ Private Subnet │
│ AZ-1a (App/DB) │ │ AZ-1b (App/DB) │
└────────────────┘ └────────────────┘
Key VPC Design Rules:
- Public Subnet: Has a direct route table entry to an Internet Gateway (
0.0.0.0/0 -> igw-xxxx).
- Private Subnet: Reaches the internet outbound only via a NAT Gateway located in a public subnet.
- Security Groups vs. NACLs:
- Security Groups: Stateful, apply at instance/ENI level, allow rules only.
- NACLs: Stateless, apply at subnet level, support allow and deny rules evaluated in numbered order.
EC2 Purchasing & Scaling Strategies
Selecting the right purchasing option can reduce compute costs by up to 90%:
| Option |
Best Use Case |
Discount |
| On-Demand |
Short-term, spiky, unpredictable workloads |
Baseline (0%) |
| Reserved Instances (1 or 3 yr) |
Steady-state database / baseline compute |
Up to 72% |
| Savings Plans |
Flexible compute usage across EC2, Fargate, Lambda |
Up to 72% |
| Spot Instances |
Fault-tolerant batch processing, big data, CI/CD |
Up to 90% |
Pro Tip: Never run mission-critical relational database masters on Spot Instances since AWS can reclaim spot capacity with a 2-minute termination notice.
Amazon S3 Storage Tier Decision Matrix
Choosing the correct S3 storage class directly impacts both retrieval latency and cost:
- S3 Standard: 99.99% availability, 11 9s durability. Millisecond retrieval. For active data.
- S3 Standard-IA (Infrequent Access): Lower storage cost, but incurs per-GB retrieval fees. 30-day minimum storage charge.
- S3 Intelligent-Tiering: Automatically moves objects between frequent, infrequent, and archive access tiers based on access patterns without operational overhead.
- S3 Glacier Flexible: Retrieval in minutes to hours. Ideal for quarterly backups.
- S3 Glacier Deep Archive: Lowest cost cloud storage ($0.00099/GB/mo). Retrieval within 12 to 48 hours.
Database Architecture: RDS vs DynamoDB
Relational Database Service (RDS / Aurora)
- Use Cases: Complex joins, ACID transactions, relational models (Postgres, MySQL).
- Multi-AZ: Synchronous standby replica in another AZ for instant automatic failover.
- Read Replicas: Asynchronous replication for horizontal read scaling (up to 15 Aurora replicas).
Amazon DynamoDB
- Use Cases: Key-value or document model with predictable single-digit millisecond latency at any scale.
- Partition Key & Sort Key: Careful partition key selection ensures uniform partition distribution and prevents hot partition throttling.
- DAX (DynamoDB Accelerator): In-memory cache delivering microsecond latency for read-heavy workloads.
# Example: Querying a DynamoDB table via AWS CLI
aws dynamodb query \
--table-name CloudStudyNotes \
--key-condition-expression "Category = :cat AND Date >= :dt" \
--expression-attribute-values '{
":cat": {"S": "AWS"},
":dt": {"S": "2026-01-01"}
}'
Key Exam & Interview Checklist