Using software provided by NESSI and EESSI

Once the environment is configured to access software provided by NESSI or EESSI, all a user has to do is loading module, for example,

module load GROMACS/2023.1-foss-2022a

Example job script using NESSI software

We demonstrate the use of NESSI in a job script using GROMACS as an example. Below is a simple job script.

#!/bin/bash
#SBATCH --account=REPLACE_WITH_YOUR_NN****K
#SBATCH --job-name=NESSI_GROMACS
#SBATCH --partition=normal
#SBATCH --mem=16G
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=2
#SBATCH --cpus-per-task=2
#SBATCH --time=5 # short runtime

module load NESSI/2023.06
module load GROMACS/2023.1-foss-2022a

# prepare test data if not yet present
if [ ! -f GROMACS_TestCaseA.tar.gz ]; then
  curl -OL https://repository.prace-ri.eu/ueabs/GROMACS/1.2/GROMACS_TestCaseA.tar.gz
fi
if [ ! -f ion_channel.tpr ]; then
  tar xfz GROMACS_TestCaseA.tar.gz
fi

rm -f ener.edr nessi_logfile.log

# run GROMACS, we use 'time' to measure how long it runs
# note: downscaled to just 1k steps (full run is 10k steps)
time gmx mdrun -ntmpi 2 -ntomp 2 -s ion_channel.tpr -maxh 0.50 -resethway -noconfout -nsteps 1000

Change the account (in line #SBATCH --account=REPLACE...) in the job script above and store it in a file named nessi_gromacs.sh Then, submit it with

sbatch nessi_gromacs.sh

When the job has ended it should have produced a file slurm-JOBID.out where JOBID is the id of the job that was run.

Example job script using locally installed software

We demonstrate the minimal change needed to the run the same software (GROMACS) using the installations provided locally on the HPC systems the following script just comments out the line loading NESSI module (# module load NESSI/2023.06)

#!/bin/bash
#SBATCH --account=REPLACE_WITH_YOUR_NN****K
#SBATCH --job-name=LOCAL_GROMACS
#SBATCH --partition=normal
#SBATCH --mem=16G
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=2
#SBATCH --cpus-per-task=2
#SBATCH --time=5 # short runtime

# module load NESSI/2023.06
module load GROMACS/2023.1-foss-2022a

# prepare test data if not yet present
if [ ! -f GROMACS_TestCaseA.tar.gz ]; then
  curl -OL https://repository.prace-ri.eu/ueabs/GROMACS/1.2/GROMACS_TestCaseA.tar.gz
fi
if [ ! -f ion_channel.tpr ]; then
  tar xfz GROMACS_TestCaseA.tar.gz
fi

rm -f ener.edr nessi_logfile.log

# run GROMACS, we use 'time' to measure how long it runs
# note: downscaled to just 1k steps (full run is 10k steps)
time gmx mdrun -ntmpi 2 -ntomp 2 -s ion_channel.tpr -maxh 0.50 -resethway -noconfout -nsteps 1000

Change the account (in line #SBATCH --account=REPLACE...) in the job script above and store it in a file named local_gromacs.sh Then, submit it with

sbatch local_gromacs.sh

When the job has ended it should have produced a file slurm-JOBID.out where JOBID is the id of the job that was run. For the examples above, the slurm-JOBID.out file contains the following information (among other things):

  • which program was actually run, e.g., GROMACS conveniently reports somthing like

    Executable:   /cvmfs/pilot.nessi.no/versions/2023.06/software/linux/x86_64/intel/skylake_avx512/software/GROMACS/2023.1-foss-2022a/bin/gmx

    Note, when using NESSI or EESSI, the exact path will be specific to the compute node the job was running on, because for each machine the version optimised for that CPU architecture is automatically selected.

  • how well GROMACS was run, e.g., it prints Performance information (higher values for ns/day are better, for hours/ns lower values are better),

  • finally, the job statistics at the end show how long the overall job was run (see Elapsed wallclock time) and how well the job used the reserved resources (see Used CPU time and Unused CPU time)

Please, do experiment with the examples above. For example,

  • In a single job script, execute the software twice (line with time gmx ...) to observe different runtimes.

  • Replace GROMACS with another software that is provided by NESSI. All available packages can be listed with

    module avail

  • Use EESSI instead of NESSI. You’ll likely observe different start-up times because NESSI and EESSI are served from different distribution networks. That means, the Elapsed wallclock time may differ for NESSI and EESSI.

Additional information