HPC-Mini-VM-Cluster

Repository documenting an HPC mini VM cluster I am building on a spare laptop - for fun!

🖥️ Mini HPC Cluster on a Laptop Using KVM + Slurm + Rocky Linux

This project sets up a miniature High-Performance Computing (HPC) cluster on a laptop using KVM virtualization, Rocky Linux 9.6, SLURM workload manager, and optional GPU passthrough. The goal is to simulate real-world HPC system management and learn HPC administration best practices.

---

Progress

This project has just been initiated and is in the planning phase.

TODO:

Category	Task	Status
✅ Base OS Setup	Install Rocky Linux 9.6 on host machine	✔ Done
✅ Disk Setup	Custom partitioning per best practices (XFS, LVM, EFI, /var/log, etc.)	✔ Done
🖧 Networking	Set up Ethernet and WiFi: Notes here	✔ Done
🔐 System Hardening	Disable root SSH login. See Hardening notes	✔ Done
🔐 System Hardening	Configure SSH key for admin as a test for later	✔ Done
🔐 System Hardening	Enable firewall (firewalld or nftables)	✔ Done
🔐 System Hardening	Set up fail2ban or ssh rate-limiting	☐ Optional
🧠 Virtualization	Install and configure KVM, libvirt, and virt-manager	☐ Todo
🖧 Networking	Set up bridge networking (br0) for VMs	☐ Todo
🖥️ VM Creation	Create 1 head node + 12 compute node VMs + 1 GPU VM	☐ Todo
🧪 VM Testing	Ensure all VMs boot, have network access, and SSH in/out works	☐ Todo
⚙ SLURM Setup	Install SLURM on head node, configure compute nodes	☐ Todo
🧪 Job Testing	Submit basic test jobs via SLURM	☐ Todo
🎮 GPU Integration	Set up passthrough VM w/ NVIDIA drivers + CUDA toolkit	☐ Todo
📦 File Sharing	(Optional) Set up NFS for /home, /scratch	☐ Optional
📊 Monitoring	(Optional) Install htop, nmon, or Prometheus stack	☐ Optional
🧹 Automation	Set up virt-clone, cloud-init, and Ansible roles	☐ Todo
🧼 Maintenance	Script backups, log rotation, and regular cluster health checks	☐ Todo

---

🧰 System Overview

One of many laptops I have sitting around is a semi-beefy System 76 machine which has the below specs.

Component	Detail
Host OS	Rocky Linux 9.6 (minimal install)
Laptop CPU	Intel Core i9-13900HX — 24 cores / 32 threads
Memory	32 GB DDR5
GPU	NVIDIA RTX A2000 — Ampere (Passthrough Capable, 6 GB VRAM)
Hypervisor	KVM + libvirt + virt-manager
Networking	Realtek RTL8125 — bridged networking for cluster nodes
Cluster Design	1 head node + 12 compute nodes + 1 GPU node (13 total VMs)

---

🧱 Disk Partition Plan

Target Disk: /dev/nvme0n1

Mount Point	Size	Scheme	Filesystem	Volume Label	Notes
/boot/efi	1 GB	Standard	EFI (FAT32)	EFI	UEFI system partition — required for UEFI booting
/boot	1 GB	Standard	XFS	BOOT	Separate /boot needed, must be non-LVM
swap	16 GB	LVM	swap	SWAP	Standard swap
/	120 GB	LVM	XFS	ROOT	Main OS and software
/var/lib/libvirt	350 GB	LVM Thin	XFS	VM_IMAGES	Thin pool for VM images and snapshots
/home	150 GB	LVM	XFS	HOME	User home directories, shared storage
/var/log	30 GB	LVM	XFS	LOGS	Prevents log bloat from filling /
Unallocated	~330 GB	—	—	—	Reserved for expansion, scratch space, or future volumes

---

🗃️ Cluster Layout

Node	CPUs	RAM	Notes
head-node	1	1 GB	SLURM controller, login node
node01	1	1 GB	Compute node
node02	1	1 GB	Compute node
...	...	...	...
node12	1	1 GB	Compute node
gpu-node	4	4 GB	GPU passthrough (NVIDIA RTX A2000)

---

⚙️ Features & Configuration Highlights

• 🔒 No root SSH access: Admin and user accounts (user1, ..., usern) only
• 🐧 SLURM Workload Manager: Installed on head node, configured across compute nodes
• 🚀 GPU Passthrough: Using vfio-pci, OVMF UEFI, and virtio interfaces
• 🔗 Bridged Networking: VMs are accessible from LAN with optional DNS names
• 📦 Ansible-ready: Cluster is automatable via Ansible for scaling or redeployment
• 📈 Monitoring ready: Planned support for htop, nmon, ganglia, or Prometheus

---

📡 Networking Strategy

• Host has bridged connection (br0) attached to physical NIC
• All VMs receive static IPs or DHCP reservations
• SSH enabled on all VMs
• Head node may act as NFS and DNS server for cluster

---

🧪 Learning Objectives

• Simulate a real HPC cluster using SLURM and KVM
• Practice GPU passthrough and PCI device management
• Understand disk provisioning, thin LVM pools, and file system design
• Configure automated user, job, and system management
• Implement secure networking and monitoring best practices

---

🚧 To-Do / Future Work

• Automate VM setup via virt-clone and cloud-init
• Set up /home and /scratch as NFS shares
• Deploy SLURM accounting and job history
• Add cluster monitoring (Grafana, Ganglia, etc.)
• Add backup strategy (rsnapshot, duplicity, etc.)
• Build sample SLURM job scripts (MPI, CUDA, etc.)

---