Storage, bandwidth and why the archive is plain files

People occasionally ask why the archive is not built on an object store. The answer is that it does not need one. The material is written once, read many times, and never modified — which is precisely the workload a filesystem and a web server already handle perfectly.

An object storage layer would add a second process, its own memory footprint, its own failure mode and an API nobody here uses. The only thing it would genuinely buy us is remote upload, and we upload rarely enough that a copy over SSH is simpler.

Serving is deliberately conventional: range requests so large files can be resumed and seeked, long cache lifetimes on immutable media paths, and no compression on material that is already compressed.

The result is boring, which is the point. Boring infrastructure is infrastructure you do not have to think about when something else breaks.