Why it matters
S3 is the backbone of nearly every modern data platform. Data lakes, ML training data, backups, static site hosting — all live in S3. Understanding its performance model and cost structure is essential for building efficient cloud systems.
The architecture
Objects are stored in buckets. Each bucket is a global namespace (bucket names are unique across all of AWS). Buckets are region-scoped for data locality; the name is global for URL uniqueness.
Storage classes: Standard (frequently accessed), Standard-IA (infrequent access), Glacier (archival), Intelligent-Tiering (auto-move). Price and access latency vary by class.
How it works end to end
PUT and GET operations are the primary API. PUT stores an object; GET retrieves it. Multipart upload for large objects (>100 MB). Presigned URLs for time-limited access without credentials.
Access control via bucket policies (JSON) and ACLs (legacy). Point-in-time versioning enables recovery from accidental deletion.
Event notifications trigger Lambda or SQS on object create/delete. Enables event-driven architectures.