2.7. Reflection Checkpoint: Core AWS Services & Global Infrastructure

Key Takeaways

• AWS's global infrastructure (Regions, Availability Zones, Edge Locations) is the physical foundation everything else sits on: Regions give you data-residency and disaster-recovery choices, AZs give you fault tolerance within a Region, and Edge Locations (via CloudFront) push content physically closer to users to cut latency.
• Compute choice is really a question of how much operational burden you want to keep: EC2 gives full OS control, Lambda removes servers entirely for event-driven code, and ECS/Fargate sit in between for containers.
• Storage type follows access pattern, not just cost: S3 for object storage accessed over HTTP, EBS for block storage tied to one EC2 instance, EFS for a file system shared across many instances, and S3 Glacier for cold archival data you rarely touch.
• RDS/Aurora (relational) and DynamoDB (NoSQL) aren't interchangeable — pick based on whether your data needs ACID transactions and a fixed schema, or massive horizontal scale with a flexible schema.
• IAM, CloudWatch, CloudTrail, Config, and Trusted Advisor together answer "who can do what," "is it healthy," "what happened," "is it still configured correctly," and "how could this be better" — losing any one of them leaves a real blind spot.

Connecting Forward

You now know AWS's core building blocks. Phase 3 shifts from "what services exist" to "who is responsible for securing them" — the Shared Responsibility Model, and the concrete security concepts (network security, encryption, access control, compliance) that put it into practice.

Self-Check Questions

• A mobile gaming company needs single-digit-millisecond reads at massive, unpredictable scale with a schema that keeps changing. Why would DynamoDB be preferred over RDS here, and what would go wrong if they chose RDS instead?
• Your team just discovered a security incident and needs to know exactly which IAM user made which API call last Tuesday. Which service answers that question, and why wouldn't CloudWatch alone be enough?