Proxmox Cluster

Cluster: rpc-cyber-dc-01

A 3-node Proxmox VE cluster providing high-availability virtualization for all lab workloads.

Cluster Diagram

[Insert proxmox-cluster.drawio diagram here]


Nodes

Node IP CPU Cores/Threads RAM Disk GPU VMs
pve1 192.168.x.x i9-13900H 14/20 64 GB 93 GB 100, 101, 103, 105
pve2 192.168.x.x i5-4590 4/4 16 GB 36 GB 104
bighost 192.168.x.x i5-12600K 10/16 64 GB 93 GB RTX 4070 102, 107

Why These Machines?

  • pve1 — A mini-PC with a laptop-class i9. Small, quiet, power-efficient, but packs 14 cores. Runs most of the lab.
  • pve2 — A 10-year-old desktop. Proves you don’t need modern hardware. Runs a single Windows Server VM just fine on 4 cores and 16 GB.
  • bighost — The muscle. A desktop-class i5-12600K with an RTX 4070 for GPU-accelerated AI workloads. Runs the Ollama inference server and Kali security lab.

Storage

Name Type Capacity Content Notes
local Directory 93 GB ISOs, backups, templates Per-node local storage
local-lvm LVM-thin 337 GB VM disks Thin-provisioned, per-node
usbdisk NFS 916 GB ISOs, VM disks, backups Shared NFS export from pve1 via USB disk

The NFS share (usbdisk) is key — it provides shared storage across all nodes without expensive SAN hardware. It’s literally a USB hard drive plugged into pve1 and exported via NFS.


VM Inventory

VMID Name Node vCPU RAM Disk OS IP Status
100 RonClaw pve1 2 8 GB 32 GB Linux 192.168.x.x Running
101 BethClaw pve1 2 8 GB 32 GB Linux Stopped
102 BigBrain bighost 12 40 GB 256 GB Linux 192.168.x.x Running
103 cainfra01 pve1 2 8 GB 128 GB Rocky Linux 10.1 192.168.x.x Running
104 cadc02 pve2 4 12 GB 256 GB Windows 192.168.x.x Running
105 CADC01 pve1 4 16 GB Win Server 2019 192.168.x.x Running
107 Kali bighost 4 4 GB 60 GB Kali Linux TBD Installing

GPU Passthrough — BigBrain

The RTX 4070 on bighost is passed directly to VM 102 (BigBrain) for GPU-accelerated AI inference:

  • GPU: NVIDIA GeForce RTX 4070 (AD104)
  • PCI address: 0000:01:00.0
  • CPU mode: host (full passthrough for best performance)
  • Purpose: Ollama LLM server running Qwen3 models (8B, 14B, 32B parameter)
  • Autostart: Yes — BigBrain comes up automatically on node boot

What is GPU Passthrough?

Proxmox can pass a physical GPU directly to a VM, giving it bare-metal GPU performance. This lets you run AI inference, video transcoding, or any GPU workload inside a VM as if the GPU were physically installed in that machine.


Cluster Features in Use

Feature How We Use It
Corosync quorum 3-node cluster ensures quorum even if one node goes down
Shared storage (NFS) ISOs and backups accessible from any node
Live migration Move VMs between nodes without downtime (for non-GPU VMs)
Autostart Critical VMs (BigBrain, cainfra01, CADC01) start automatically
UEFI boot Modern secure boot for newer VMs

What You Learn Building This

  • Hypervisor installation — bare-metal Proxmox deployment
  • Cluster configuration — Corosync, quorum, node management
  • VM creation — resource allocation, disk provisioning, network bridging
  • Storage architecture — local vs shared, LVM-thin, NFS exports
  • GPU passthrough — IOMMU, PCI device assignment, driver isolation
  • Capacity planning — balancing workloads across heterogeneous hardware
Scroll to Top