AWS Architecture & Services Deep Dive
AWS Compute Services: EC2, Lambda, ECS & EKS
Compute service selection is one of the most common interview topics. Understanding when to use each service demonstrates architectural maturity.
EC2: The Foundation
Instance Types & Selection
| Family | Use Case | Example |
|---|---|---|
| General Purpose (M/T) | Web servers, small DBs | m7i.xlarge, t3.medium |
| Compute Optimized (C) | Batch processing, gaming | c7i.2xlarge |
| Memory Optimized (R/X) | In-memory databases | r7i.4xlarge, x2idn.large |
| Storage Optimized (I/D) | Data warehousing, HDFS | i4i.large, d3.xlarge |
| Accelerated (P/G) | ML training, graphics | p5.48xlarge, g5.xlarge |
Interview Question: Right-sizing
Q: "Your web application uses m5.2xlarge instances but CPU averages 15%. What do you recommend?"
A: Analyze CloudWatch metrics for CPU, memory, and network. Consider:
- Downsize to m5.large (4x smaller, 4x cheaper)
- Use Compute Savings Plans for predictable workloads (up to 72% savings)
- Consider Graviton-based instances (m7g/Graviton3 or m8g/Graviton4) for additional 20-40% cost reduction
- Implement auto-scaling to handle peaks
Lambda: Serverless Compute
Key Characteristics
- Max duration: 15 minutes
- Memory: 128 MB to 10,240 MB
- Concurrency: 1,000 default (can increase)
- Cold start: 100ms-1s (varies by runtime)
When to Use Lambda
Good Fit:
- Event-driven processing (S3 triggers, API Gateway)
- Short-running tasks (< 15 min)
- Variable/unpredictable traffic
- Microservice APIs with low latency requirements
Poor Fit:
- Long-running processes (use Step Functions or ECS)
- High-frequency, consistent load (EC2 cheaper at scale)
- GPU-intensive workloads
Interview Question: Lambda vs EC2 Cost
Q: "Your API receives 10M requests/month, each taking 200ms and using 512MB. Should you use Lambda or EC2?"
A: Calculate Lambda cost:
- Requests: 10M × $0.20/1M = $2.00
- Compute: 10M × 0.2s × 0.5GB × $0.0000166667 = $16.67
- Total: ~$19/month
EC2 t3.medium could handle this but requires management overhead. Lambda wins for simplicity at this scale. Beyond ~50M requests/month, EC2 becomes more cost-effective.
ECS: Container Orchestration
Launch Types
| Type | Use Case | Pricing |
|---|---|---|
| EC2 | Control over infrastructure | Pay for EC2 instances |
| Fargate | Serverless containers | Pay per vCPU/memory-hour |
⚠ Prices change frequently. The values above are for illustration only and may be out of date. Always verify current pricing directly with the provider before making cost decisions: Anthropic · OpenAI · Google Gemini · Google Vertex AI · AWS Bedrock · Azure OpenAI · Mistral · Cohere · Together AI · DeepSeek · Groq · Fireworks AI · Perplexity · xAI · Cursor · GitHub Copilot · Windsurf.
ECS Architecture Components
- Cluster: Logical grouping of services
- Service: Maintains desired task count
- Task Definition: Container configuration (like Dockerfile + docker-compose)
- Task: Running container instance
Interview Question: ECS vs EKS
Q: "When would you choose ECS over EKS?"
A: Choose ECS when:
- AWS-native integration is priority
- Team lacks Kubernetes expertise
- Simpler deployment model is acceptable
- Vendor lock-in is not a concern
Choose EKS when:
- Kubernetes expertise exists on team
- Multi-cloud portability needed
- Rich Kubernetes ecosystem required
- Complex networking (service mesh) needed
EKS: Managed Kubernetes
EKS Pricing
- Control plane: $0.10/hour (~$73/month)
- Worker nodes: EC2/Fargate pricing
- No additional charge for managed add-ons
Interview Question: EKS Node Types
Q: "What are the trade-offs between managed node groups, self-managed nodes, and Fargate for EKS?"
| Option | Pros | Cons |
|---|---|---|
| Managed Node Groups | AWS handles updates, easy scaling | Less customization |
| Self-Managed Nodes | Full control, custom AMIs | You manage updates |
| Fargate | Serverless, no nodes to manage | Higher cost per vCPU, limited features |
Compute Decision Framework
Need GPU? → P/G instance types
Need serverless?
└── Short-lived (< 15min)? → Lambda
└── Longer? → Fargate or Step Functions
Need containers?
└── Kubernetes required? → EKS
└── Simpler model? → ECS
Default → EC2 with Auto Scaling
Interview Tip: Always discuss cost implications, scaling behavior, and operational overhead when comparing compute options.
Next, we'll explore AWS storage services and selection criteria. :::