A short primer on object storage

Object storage is one of the most-used and least-understood primitives in cloud infrastructure. This post is a short primer on what object storage is, what it is good at, and where it falls down relative to the alternatives.

The model

An object store maps an opaque key (a string) to an opaque value (bytes plus a small fixed metadata bag). The key namespace is flat; the apparent directory hierarchy is a UI convention enforced by delimiters in the listing API.

Where it shines

Read-heavy workloads where the reads are large enough that the per- request overhead does not dominate. Backup, archival, static asset hosting, and data lake patterns all fit naturally.

Where it falls down

Transactional workloads where multiple objects need to update atomically. Object stores typically offer single-object atomicity only; coordinating across objects is a layer the application has to build.

← All posts