A HPC Cluster (Slurm) on Proxmox VMs

A local HPC cluster that uses the Slurm Scheduler.

• Step 1: Install Proxmox.
• Step 2: Setup the Slurm Scheduler.
• Step 3: Install Application (Module, MPI).

Requirements

• A Router.
• A baremetal machine such as Intel NUC PC. We used a Mac-mini with 4CPUs and 8GB memory.

Proxmox VE

Proxmox VE is an open-source server management platform for enterprise virtualization. We created three Ubuntu virtual machines (Cloud Init) to set up a HPC cluster.

HPC Cluster

The local HPC cluster uses the Slurm Scheduler and Module for management.

ubuntu@manager:~$ sinfo
PARTITION AVAIL  TIMELIMIT  NODES  STATE NODELIST
mac*         up   infinite      2   idle compute[01-02]

ubuntu@manager:~$ sbatch hello-submit.sh
Submitted batch job 36

ubuntu@manager:~$ squeue
             JOBID PARTITION     NAME     USER ST       TIME  NODES NODELIST(REASON)
                36       mac hello-wo   ubuntu  R       0:02      1 compute01

ubuntu@manager:~$ sacct -j 36
       JobID    JobName  Partition    Account  AllocCPUS      State ExitCode
------------ ---------- ---------- ---------- ---------- ---------- --------
36           hello-wor+        mac research_+          1    RUNNING      0:0
36.batch          batch            research_+          1    RUNNING      0:0

References:

• https://glmdev.medium.com/building-a-raspberry-pi-cluster-784f0df9afbd
• https://wiki.fysik.dtu.dk/Niflheim_system/Slurm_database/
• https://wiki.fysik.dtu.dk/Niflheim_system/Slurm_accounting/
• https://southgreenplatform.github.io/trainings/hpc/slurminstallation/
• https://linuxcommandlibrary.com/man/sacctmgr
• https://hpc.gsi.de/virgo/slurm/accounts.html