SCARF

The MPI version of Dissolve is installed as a module on SCARF, and is built (as of v0.7.3) using the gcc compilers.

Of course, you’ll need an account on SCARF - register here.

Available Versions

To see the available versions, run:

bob@scarf:~> module avail dissolve

--------------- /apps20/contrib/modulefiles ----------------
contrib/dissolve/0.8.9   contrib/dissolve/0.9.0

To source one ready for use in a script run:

bob@scarf:~> module load contrib/dissolve/0.8.0

Example Scripts

The following example scripts for SCARF can be used as a basis for your own, or slightly modified and used as-is for the most part, but be aware that at present there are specific scripts for specific versions of Dissolve.

Before using any of these this script, you should:

1. Change JOBNAME to a short descriptive name for your job, e.g. water10k
2. Request an appropriate number of processors / tasks. How this is done depends on the version of Dissolve being run (see individual scripts for details).
3. Note the job time specified by -t is in the format HH:MM:SS, so the scripts here will run for six hours before being terminated by the queuing system on SCARF, regardless of whether the simulation has finished or not. Alternatively, you can use the format DD-HH to specify the job time in days and hours (e.g. 3-0 for three full days).
4. Set the path to the directory containing your input file for Dissolve - i.e. change ~/path/to/simulation to ~/work/water/tests/, for example.
5. Set an appropriate number of iterations for dissolve to run (100 in this example).

Note that, as written here, the associated error and log files written by the system for the run are prepended by the numerical job identifier for the run (e.g. 435003.err).

Pure MPI (v0.7.X)

#!/bin/bash
#SBATCH -J JOBNAME
#SBATCH -q scarf
#SBATCH -C scarf18
#SBATCH -n 8
#SBATCH --tasks-per-node=8
#SBATCH -t 06:00:00
#SBATCH -o %j.log
#SBATCH -e %j.err

# Load necessary modules
module load foss
module load contrib/dissolve/0.7.3

# Change to directory containing simulation
cd ~/path/to/simulation

# Run Dissolve
mpirun dissolve-mpi input.txt -n 100

This script requests eight processors (-n 8 in the #SBATCH section) all on the same physical node (--tasks-per-node=8) - highly recommended for performance reasons. This is a good place to start, but can be increased (or reduced) as necessary.

Multithreaded (v0.8.X, v0.9.X)

#!/bin/bash
#SBATCH -J JOBNAME
#SBATCH -q scarf
#SBATCH -J sles0.4
#SBATCH -q scarf
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=12
#SBATCH -t 06:00:00
#SBATCH -o %j.log
#SBATCH -e %j.err

# Load necessary modules
module load foss
module load contrib/dissolve/0.9.0

# Change to directory containing simulation
cd ~/path/to/simulation

# Run Dissolve
dissolve input.txt -n 100

For the multithreaded version we must explicitly request a single node to run one “task”, but that task can use 12 CPUs.