Balfrin¶

Balfrin is a high-performance computing (HPC) cluster used by MeteoSwiss, located in Lugano and hosted by CSCS (Swiss National Supercomputing Centre). It is an MPI cluster managed by Slurm , primarily used for research and development (R&D) and as a production failover system.

Note

C2SM does not officially support Balfrin. However, instructions for setting up ICON on Balfrin are available in the Compile section of ICON.

Hardware Specifications¶

Balfrin's hardware is designed for scalability. The maximum configuration is outlined below, but resources can be scaled down based on demand in the future.

Node type	Nodes	GPU type	GPUs	GPU RAM per GPU	CPU model type	CPU Sockets	Cores per socket	Threads per core	Threads	RAM
Login node	3	-	0	-	AMD EPYC 7713	2	64	2	256	256 or 512 GB
GPU node	46	NVIDIA A100	4	96 GB	AMD EPYC 7713	1	64	2	128	512 GB
Postprocessing node	15	-	0	-	AMD EPYC 7713	2	64	2	256	512 GB
User Access Node*	1	-	0	-	AMD EPYC 7713	2	64	2	256	512 GB

Storage and Quota¶

Storage¶

For detailed storage documentation, refer to the CSCS Storage Documentation .

Quota¶

Home Directory Limits:

Soft limit: 80 GB
Hard limit: 100 GB

To check your quota usage, use one of the following commands:

# Check home directory usage
df -h $HOME

# Check Lustre attributes
lsattr -p -d $HOME

# Check project quota (example for /users)
lfs quota -p 74 /users -h

Network Access¶

The login nodes can be accessed via balfrin.cscs.ch. This will forward you to a random login node, so you can profit form their redundancy.

Example Job Script¶

Below is a template for submitting jobs using Slurm on Balfrin:

#!/bin/bash
#SBATCH --job-name=my_job
#SBATCH --output=output.txt
#SBATCH --error=error.txt
#SBATCH --time=01:00:00
#SBATCH --nodes=2
#SBATCH --ntasks=4
#SBATCH --cpus-per-task=2
#SBATCH --mem=4G
#SBATCH --partition=partition_name
module load my_module
srun my_program

Software Stack¶

Overview¶

CSCS provides a curated software stack for general-purpose and domain-specific applications. The stack is installed in /mch-environment/ and is organized by version, with each version being immutable but subject to eventual decommissioning.

General purpose software¶

The $USER_ENV_ROOT environment variable points to the latest production-grade software stack.

Loading Modules¶

To use the modules in the stack:

# Add the stack's module directory to your MODULEPATH
module use $USER_ENV_ROOT/modules

# List available modules
module avail

Tip: Consider adding module use $USER_ENV_ROOT/modules to your ~/.bashrc for convenience.

Domain-Specific Software (UENV)¶

Some software (e.g., ncview) is available in user environments (uenvs). To use these:

Install the CSCS UENV CLI: Follow the Getting Started Guide .
Using UENV: There are three ways to use uenvs:

Option 1: Load modules from a uenv

uenv start --view=modules climana/24.7:v1-rc4
module load ncview/2.1.9
ncview

Option 2: Directly add applications

uenv start --view=default netcdf-tools/2024:v1
ncview

Option 3: Run scripts with a specific uenv

uenv run {uenv/version:tag} -- ./job-using-uenv.sh

Spack and Spack-C2SM¶

Spack¶

Spack is a flexible package manager for HPC. It allows users to install software without sudo rights and manage complex dependencies.

Spack-C2SM¶

Spack-C2SM is an extension of Spack, tailored for MeteoSwiss. It includes proprietary packages and machine-specific configurations.

Setup¶

# Clone the Spack-C2SM repository
git clone --depth 1 --recurse-submodules --shallow-submodules \
  https://github.com/C2SM/spack-c2sm.git -b latest

# Source the environment
. spack-c2sm/setup-env.sh $USER_ENV_ROOT

Usage¶

# List available packages
spack find

# Install a package (e.g., libelf)
spack install libelf

Note

Spack-C2SM is designed to complement the CSCS software stack, providing additional tools and libraries specific to MeteoSwiss workflows.