Automatic Provisioning of Bare Metal Edge for AI Inference
summary
Claims a repeatable, user‑initiated method to provision bare‑metal edge nodes for AI inference by shipping a bootable image instead of relying on BMC/Redfish or remote management. Uses cordant wrapper around Cluster API to manage inventory and metal provisioning, and builds immutable images with Kyros ISO with k0s deployer (GitHub Actions produce ISOs containing OCI artifacts). Image supports TPM trusted boot, A/B upgrades, and boots isolated clusters that run local AI offline inference; presenter reports successful onsite deployments and demos.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 2:13Traditional bare metal workflow
- 5:46Metal³ and Cluster API
- 8:58Callback provisioning challenges
- 12:12BLURP framework tradeoffs
- 16:31Edge AI workload requirements
- 18:17Kyros provisioning
- 22:35Local AI deployment
- 24:45Q&A
related talks
36mBare metal Kubernetes on mixed x64 and ARM64
24mDemo of Kubernetes at the Edge for AI Model Deployments
34mRemote Baremetal Deployments using Bifrost
26mBuilding a Cloud Native Edge Platform for Accelerating AI Inference
21mAIaaS with Dell AI ready infrastructure
32mNo Touch Installation Concept
30mDesigning a Hybrid AI Platform on Kubernetes
32mAutoscale Your on-Premises Bare Metal Kubernetes Clusters