CM cluster documentation#
CM is the cluster of the Theory and Simulation of Condensed Matter group @UniTS!
Overview#
Computing nodes#
The cluster comprises the following computing nodes.
Node |
CPU |
Cores |
Frequency |
RAM |
PSU |
|---|---|---|---|---|---|
cm01 |
Xeon Gold 5218 |
32 (2x125W) |
2.30 GHz |
96 Gb (DDR4-2666) |
550W (x2) |
cm02 |
Xeon Gold 5118 |
24 (2x105W) |
2.30 GHz |
64 Gb (DDR4-2400) |
550W (x2) |
cm03 |
Xeon Gold 5118 |
24 (2x105W) |
2.30 GHz |
64 Gb (DDR4-2400) |
550W (x2) |
cm04 |
Xeon Silver 4210R |
10 (1x100W) |
2.40 GHz |
16 Gb (DDR4-2400) |
550W (x1) |
General rules#
The login node is
cm01.units.itDo not ssh or run processes manually in the other nodes
Submit jobs from the login node using SLURM, see next section
Avoid running long processes on the login node
Keep an eye on the data you store on disk (
du -sh ~)Regularly remove data you do not need anymore
Submitting jobs#
Interactive job
srun <command>
nohup srun -n 16 sleep 10 &
Sample batch job
#!/bin/bash
#SBATCH --ntasks-per-node=16
mpirun -n $SLURM_TASKS sleep 10
Batch job
sbatch <script>
Jobs queue
squeue
Frequently asked questions (FAQ)#
How do I install python package X?#
It is recommended to install python packages in a python virtual environment. The following command
python3 -m venv env
creates a virtual environment in folder env/. To activate the environment and install packages within it
. env/bin/activate
pip install <package_to_install>
Once you have finished using the environment and you want to go back to the default python distribution
deactivate
You can have several environments (e.g. one per project) and they can be deleted anytime by just delete the folder env/.
Can I use docker images?#
Yes, ask the admin to add you to the docker group.
How can I submit a job on a specific node?#
Typically, you should let the scheduler find the first available node for your job. But if we want to submit it on a specific node, say cm03
sbatch -w cm03 job.sh
How can I use the Intel compiler?#
To use the Intel compilers (ifort and icc) and the MKL libraries
. /opt/intel/oneapi/setvars.sh
You can put this command in your ~/.bashrc file, so that it is executed every time you log in.
Can squeue print the number of cores used by jobs?#
Add this to your ~/.bashrc file
alias squeue='squeue -o "%.7i %.9P %.8j %.8u %.2t %.10M %.6D %C %R"'
Log out, log in again and type squeue
How do I get the number of idle cores in the cluster?
Type
free_cpus
Updating this web page#
Log into the cm01 node and go to
cd /var/www/docs
Edit the file index.rst (in reStructuredText syntax, see for instance this tutorial).
Compile and deploy the webpage
make deploy
and check the result by reloading this page (ignore the warnings). When you are satisfied, commit your changes
git commit -am "...describe your changes here..."