Metal As A Service — Manage Your Bare Metal in MaaS Like a Virtual Machine
summary
Shows how Canonical MAAS is used to run a self-hosted mini-cloud that treats physical servers like VMs across three regions (Chartres, Paris, Marseille). Covers integrating PXE, BMC (iDRAC/HPE), cloud-init/curtin for persistent OS provisioning, Terraform/Pulumi for orchestration, and evaluations with OpenStack Ironic and Tinkerbell; it provisions ~300 donated servers with GPUs for local LLM workloads and uses Rust tooling (Scafandre, OpenBarMetal) to measure consumption, auto-scale, automate power on/off and reprovisioning, cutting energy use by about 3× while keeping data sovereign.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 6:03Bare metal like a VM
- 12:24Infrastructure evolution needs
- 17:21Manual server limits
- 20:47Automating bare metal choices
- 24:14Metal as a Service (MaaS)
- 31:18MaaS provisioning demo
- 36:41MaaS energy optimization
- 39:54Project impact summary
related talks
34mmetal-stack - kubernetes on bare metal
27mSupporting Security Sensitive Tenants in a Bare-Metal Cloud
06mMaturing Bare Metal Provisioning With Metal3
32mAutoscale Your on-Premises Bare Metal Kubernetes Clusters
27mAdvanced Multipurpose Infrastructure: A Self-Managed Cloud Environment with HashiCorp Terraform
27mHow to get rich with cloud!
25mAutomatic Provisioning of Bare Metal Edge for AI Inference
18mScaling Bare Metal Provisioning with Nova and Ironic at CERN