This study documents how to design and deploy a multi-region OpenStack environment capable of hosting Kubernetes clusters and serving AI inference workloads efficiently on CPUs.
Key Learnings
- Federation of OpenStack regions via Keystone + Magnum
- CPU-only inference pipelines with containerized models
- Network isolation using OVN and VLAN overlays
- Storage integration using Ceph and object gateways
Outcome: An end-to-end reproducible environment for AI infrastructure testing using open-source tools.