Object Storage

What is object storage?

Object storage stores data as objects — the file bytes plus metadata plus a key — in flat containers called buckets, accessed via HTTP APIs rather than mounted filesystems. Amazon S3 defined the model; Google Cloud Storage, Azure Blob Storage, and S3-compatible systems (Cloudflare R2, MinIO) followed. It scales to exabytes with 11-nines durability and pennies-per-GB pricing.

Object vs block vs file

Type	Looks like	Best for
Object	HTTP API + keys	Backups, media, data lakes, static sites, ML datasets
Block	A raw disk attached to one VM (EBS)	Databases, boot volumes — anything needing a filesystem
File	A shared network drive (EFS, NFS)	Legacy apps, shared home directories

Objects are immutable — you replace, not edit — and there are no real directories, just key prefixes that look like paths.

Storage classes and lifecycle

Providers tier by access frequency: hot (standard) → infrequent access → archive (Glacier-class, hours to retrieve). Lifecycle rules move objects between tiers automatically — the difference between a data lake that costs $2k/month and the same one at$400.

Cost reality

The sticker price (~$0.02/GB-month) is the small part. The bills that surprise teams: **egress** (getting data *out* costs ~$0.09/GB on major clouds — moving 100 TB out is ~$9k, which is R2’s whole zero-egress pitch), request charges on millions of small objects, and cross-region transfer. Design rule: process data in the region where it lives.

What people get wrong

• Public buckets. The classic cloud breach; default-deny, use presigned URLs for sharing, audit with provider tools.
• Using it as a filesystem. Listing “directories” with millions of keys and rewriting objects for small edits fights the model; that workload wants block or file storage.
• Ignoring lifecycle from day one. Data accumulates forever by default — the bill only goes up until you add rules.

Object Storage vs. Block and File Storage

Object storage differs fundamentally from the other two major storage paradigms: block storage (like AWS EBS) presents raw, fixed-size blocks of storage that an operating system formats with its own filesystem, well suited to database storage and boot volumes requiring low-latency random access; file storage (like AWS EFS) presents a traditional hierarchical filesystem accessible over a network protocol, well suited to shared file access across multiple servers. Object storage instead stores data as discrete objects (each with data, metadata, and a unique identifier) accessed via a simple HTTP-based API rather than a filesystem interface, trading the ability to do fine-grained in-place modifications (you replace an entire object rather than editing part of it) for effectively unlimited scalability and much lower cost per gigabyte — a tradeoff that makes it the default choice for static assets, backups, data lakes, and any workload that reads and writes whole objects rather than needing filesystem-style partial access.