Simple Storage Service (S3)

What is S3?

Amazon S3 is the original cloud object store and still the center of gravity for cloud data: files stored as objects in buckets, addressed by key, retrieved over HTTP, with 11 nines of durability (objects replicated across at least three AZs). Nearly everything else in AWS leans on it — Athena queries it, CloudFront serves from it, Lambda triggers on it, and every data lake starts as a bucket.

Storage classes: the decision that compounds

Standard ($0.023/GB-month) is where data lands; the savings live in tiering. **Intelligent-Tiering** auto-moves objects between access tiers for a tiny monitoring fee — the correct default for unknown access patterns. **Glacier** classes drop to ~$0.004–0.001/GB for archives, with retrieval times and fees as the trade. Lifecycle rules automate the migration; a data platform without them pays Standard rates forever for data nobody reads. The other cost pillars: egress ($0.09/GB to internet — the reason CDNs front buckets), request charges (millions of small PUTs add up), and cross-region replication doubling storage plus transfer.

The security posture that matters

Public-bucket leaks remain the most famous cloud breach class, which is why AWS now ships Block Public Access on by default and account-wide. The modern access model: buckets private, access via IAM roles and bucket policies, sharing via presigned URLs, CloudFront with Origin Access Control for web serving, and S3 endpoints so VPC traffic never crosses NAT.

What people get wrong

• Small-file avalanches — millions of kilobyte objects make requests, not storage, the dominant cost; batch or compact.
• Treating S3 as a filesystem: no rename (copy+delete), prefix-listing at scale is slow, and mounting it as a disk disappoints.
• Versioning without lifecycle — deleted objects live on as billed noncurrent versions indefinitely.

Primary source: Amazon S3 documentation