AWS Architecture Patterns¶

Common AWS architecture patterns for real-world applications.

Three-Tier Web Application¶

Classic pattern: presentation, application, data layers with high availability.

graph TB
    subgraph "Internet"
        Users[Users]
    end

    subgraph "AWS Cloud"
        subgraph "Availability Zone 1"
            ALB1[Application<br/>Load Balancer]
            EC2_1[EC2 Instance<br/>App Server]
            RDS_Primary[(RDS Primary<br/>PostgreSQL)]
        end

        subgraph "Availability Zone 2"
            EC2_2[EC2 Instance<br/>App Server]
            RDS_Standby[(RDS Standby<br/>Read Replica)]
        end

        S3[S3 Bucket<br/>Static Assets]
        CloudFront[CloudFront CDN]
    end

    Users -->|HTTPS| CloudFront
    CloudFront -->|Static| S3
    CloudFront -->|Dynamic| ALB1
    ALB1 --> EC2_1
    ALB1 --> EC2_2
    EC2_1 -->|Read/Write| RDS_Primary
    EC2_2 -->|Read/Write| RDS_Primary
    RDS_Primary -.->|Replicate| RDS_Standby

    style Users fill:#e1f5ff
    style CloudFront fill:#ff9900
    style S3 fill:#569a31
    style ALB1 fill:#ff9900
    style EC2_1 fill:#ff9900
    style EC2_2 fill:#ff9900
    style RDS_Primary fill:#527fff
    style RDS_Standby fill:#527fff

Components:

• CloudFront: Global CDN, caches static assets at edge locations
• S3: Object storage for images, CSS, JavaScript
• ALB: Distributes traffic across EC2 instances in multiple AZs
• EC2: Application servers running in Auto Scaling group
• RDS: Managed PostgreSQL with multi-AZ failover

Real talk: This pattern handles 10k-100k requests/day. Add auto-scaling for growth.

Serverless Microservices¶

Event-driven architecture with Lambda, API Gateway, and DynamoDB.

graph LR
    Client[Mobile/Web<br/>Client] -->|HTTPS| APIGW[API Gateway<br/>REST API]
    APIGW -->|Invoke| Auth[Lambda<br/>Auth Function]
    APIGW -->|Invoke| Users[Lambda<br/>Users Service]
    APIGW -->|Invoke| Orders[Lambda<br/>Orders Service]

    Auth -->|Read/Write| Cognito[Cognito<br/>User Pool]
    Users -->|Read/Write| UserDB[(DynamoDB<br/>Users Table)]
    Orders -->|Read/Write| OrderDB[(DynamoDB<br/>Orders Table)]

    Orders -->|Publish| SNS[SNS Topic<br/>Order Events]
    SNS -->|Subscribe| Email[Lambda<br/>Email Service]
    SNS -->|Subscribe| SQS[SQS Queue]
    SQS -->|Process| Worker[Lambda<br/>Worker Function]

    style Client fill:#e1f5ff
    style APIGW fill:#ff9900
    style Auth fill:#ff9900
    style Users fill:#ff9900
    style Orders fill:#ff9900
    style Email fill:#ff9900
    style Worker fill:#ff9900
    style Cognito fill:#dd344c
    style UserDB fill:#527fff
    style OrderDB fill:#527fff
    style SNS fill:#ff9900
    style SQS fill:#ff9900

Components:

• API Gateway: RESTful API with authentication, rate limiting, caching
• Lambda: Stateless functions, auto-scale, pay-per-invocation
• DynamoDB: NoSQL database with single-digit millisecond latency
• SNS/SQS: Async messaging for decoupled microservices
• Cognito: User authentication and authorization

Real talk: Scales to millions of requests, costs pennies at low traffic. Cold starts are 100-500ms.

Data Pipeline Architecture¶

ETL pattern for processing large datasets with S3, Glue, and Athena.

graph TB
    Sources[Data Sources<br/>Logs, APIs, Databases] -->|Stream| Kinesis[Kinesis Data<br/>Streams]
    Sources -->|Batch| S3_Raw[S3 Raw Bucket<br/>Landing Zone]

    Kinesis -->|Real-time| Firehose[Kinesis Data<br/>Firehose]
    Firehose --> S3_Raw

    S3_Raw -->|Trigger| Glue[AWS Glue<br/>ETL Jobs]
    Glue -->|Transform| S3_Processed[S3 Processed<br/>Parquet Format]

    S3_Processed -->|Catalog| GlueCatalog[Glue Data<br/>Catalog]
    GlueCatalog -->|Query| Athena[Athena<br/>SQL Queries]
    GlueCatalog -->|Visualize| QuickSight[QuickSight<br/>Dashboards]

    S3_Processed -->|Train| SageMaker[SageMaker<br/>ML Models]

    style Sources fill:#e1f5ff
    style Kinesis fill:#ff9900
    style Firehose fill:#ff9900
    style S3_Raw fill:#569a31
    style Glue fill:#ff9900
    style S3_Processed fill:#569a31
    style GlueCatalog fill:#ff9900
    style Athena fill:#ff9900
    style QuickSight fill:#ff9900
    style SageMaker fill:#ff9900

Components:

• Kinesis: Real-time data streaming (alternative to Kafka)
• S3: Data lake storage (raw and processed data)
• Glue: Serverless ETL, converts JSON/CSV to optimized Parquet
• Athena: Query S3 data with SQL, pay per query ($5/TB scanned)
• QuickSight: BI dashboards, ML-powered insights

Real talk: Processes terabytes for cents. Use Parquet format (10x cheaper queries than JSON).

Well-Architected Framework¶

AWS's five pillars for building reliable, secure, efficient systems.

Security¶

Design Principles:

• Identity and Access Management - Use IAM roles, never embed credentials
• Detective Controls - Enable CloudTrail, GuardDuty, Config
• Infrastructure Protection - VPC isolation, security groups, NACLs
• Data Protection - Encrypt at rest (KMS) and in transit (TLS)
• Incident Response - Automated remediation with Lambda

Reliability¶

Design Principles:

• Multi-AZ Deployment - RDS, ALB, EC2 across 2+ availability zones
• Auto Scaling - Respond to demand changes automatically
• Backup and Recovery - Automated snapshots, cross-region replication
• Change Management - Infrastructure as code (CloudFormation/Terraform)
• Failure Isolation - Bulkheads prevent cascading failures

Performance Efficiency¶

Design Principles:

• Selection - Choose right compute (EC2 vs Lambda vs Fargate)
• Review - Continuously evaluate new services
• Monitoring - CloudWatch metrics, X-Ray tracing
• Trade-offs - Consistency vs latency, normalization vs denormalization

graph TD
    Start{Compute Need?} -->|Containers| Container{Orchestration?}
    Start -->|VMs| VM{Persistent?}
    Start -->|Functions| Lambda[Lambda<br/>Event-driven]

    Container -->|Yes| EKS[EKS<br/>Kubernetes]
    Container -->|No| ECS[ECS/Fargate<br/>Simpler]

    VM -->|Yes| EC2[EC2<br/>Full Control]
    VM -->|No| Batch[AWS Batch<br/>Job Scheduling]

    style Start fill:#e1f5ff
    style Lambda fill:#ff9900
    style EKS fill:#ff9900
    style ECS fill:#ff9900
    style EC2 fill:#ff9900
    style Batch fill:#ff9900

Cost Optimization¶

Design Principles:

• Right Sizing - Match instance size to workload (don't over-provision)
• Elasticity - Auto-scale down during off-peak hours
• Pricing Models - Reserved Instances (72% off), Spot (90% off)
• Managed Services - RDS cheaper than self-managed EC2 databases
• Cost Allocation - Tag everything for chargeback/showback

Operational Excellence¶

Design Principles:

• Operations as Code - Infrastructure as code, runbooks as code
• Frequent, Small Changes - Reduce blast radius of failures
• Refine Operations - Learn from failures, improve processes
• Anticipate Failure - Chaos engineering, game days
• Learn from Failures - Post-mortems without blame

Last Updated: 2026-01-31 | Vibe Check: Structural - These patterns are industry-standard. Not innovative, but proven at scale. Use them unless you have a compelling reason not to.

Tags: aws, architecture, well-architected, patterns