druid

Infrastructure still requires babysitting. About time it catches up to AI coding.

Exactly how I'm thinking about it. NUMA, the RAM/disk hierarchy, CXL. Operators have always abstracted over nonuniform substrates with very different latency tiers. The fabric inside a modern server is already a small network. But the argument for an OS at the cluster level isn't that the interconnect becomes seamless, but that the substrate becomes standardized, regardless of underlying hardware.

* * *

Every abstraction we tried pushed complexity down instead of eliminating it altogether. After having spent more than enough time in production on bare metal GPU clusters, we realized the problem wasn't the tooling, but the layer. ClusterdOS is what happens when you go the other direction. Happy to dig in.