HPC script example¶
Installation quick notes¶
Follows the installation guide for running gospl on HPC.
In most cases, MPI, PETSc and Hdf5 libraries for Python3 will be installed.
Using pip you will be able to install the remaining libraries locally (pip install XXXX --user).
PBS script example¶
After installation on HPC, you might be able to submit a job to the queue system using for example:
qsub job.pbs
Here is a minimal PBS script provided as an example, it is worth noting that the environment will likely
depend on the HPC service you are using:
#!/bin/bash
# Project
#PBS -P PROJECTNAME
# 96 CPUs
#PBS -l select=12:ncpus=8:mpiprocs=8:mem=60GB
#PBS -l walltime=4:00:00
#PBS -M email@address
#PBS -m abe
#PBS -q ALLOCNAME
# set up environment
module load gcc/4.9.3 python/3.6.5 petsc-gcc-mpich/3.11.1
cd $PBS_O_WORKDIR
# Launching the job!
mpirun -np 96 python3 runMinimal.py
A minimal python file for gospl¶
where the runMinimal.py file will be of the form:
from gospl.model import Model as sim
input = "input.yml"
# Minimal model
model = sim(input, True, False)
model.runProcesses()
model.destroy()
Runtime¶
Here are some indicative wall times for 2 specific resolution ran for a 10 million years simulation:
Case 1¶
Resolution min:7.6 km, max:10.3 km, mean: 9.1 km
Pts number |
CPU number |
Wall time |
CPU nodes |
|---|---|---|---|
10612062 |
8 |
17:08:58 |
1 |
10612062 |
16 |
08:58:32 |
2 |
10612062 |
32 |
04:57:58 |
4 |
10612062 |
64 |
03:32:15 |
6 |
10612062 |
128 |
02:40:53 |
7 |
Case 2¶
Resolution min:4.8 km, max:7.6 km, mean: 6.0 km
Pts number |
CPU number |
Wall time |
CPU nodes |
|---|---|---|---|
17004184 |
64 |
07:28:41 |
4 |
17004184 |
96 |
06:38:11 |
4 |
17004184 |
128 |
05:29:51 |
6 |
17004184 |
144 |
04:59:49 |
6 |
17004184 |
168 |
03:31:14 |
7 |