About

This knowledge base documents how to design, build and validate AI and HPC infrastructure on Kubernetes, one layer at a time. It is meant to be read, taught from, and run.

How it is organized

Section	What it holds
Concepts	Mental models: the nine layers, parallelism, workload taxonomy, co-design
Layers	One module per layer (L1–L9) plus the storage data path
Control plane	Kubernetes, schedulers, operators, device plugins and DRA, which span every layer
Cross-cutting	Observability, security, reliability, multi-tenancy, cost, container runtime
Workloads	Each workload type and the infrastructure it needs
Paths	Guided journeys that walk through several layers for one goal
Labs	Runnable environments: kind, Proxmox CPU lab, GPU cloud
Reference architectures	Complete designs assembled from the modules

Content is written once, in the layer or cross-cutting module where it belongs. Workloads and paths link to it rather than repeating it.

Author

Maintained by Raj Marni: Principal Cloud, Kubernetes, AI and HPC Infrastructure Architect. See about the author.