Search
You are here:
Home
»
Computing Resources
»
High Performance Computing Centre
»
Octave simulations with SLURM
:
Operational / Internal Site
Differences
This shows you the differences between two versions of the page.
View differences:
Side by Side
Inline
Go
Link to this comparison view
Go
Go
—
computing:hpcc:octave [2016/01/25 07:56]
(current)
Line 1:
Line 1:
+
====== Octave simulations with SLURM ======
+
+
The following provides a template for Octave simulations. Although the Octave
+
available on the cluster nodes is compiled with OpenMP support, Octave by
+
default only uses 1 CPU. The template therefore only reserves one CPU. If you
+
make use of a third-party toolbox which does use multiple CPUs you will have to
+
modify the template (similar to the multi-threaded MATLAB example).
+
+
===== Single CPU example =====
+
+
The following will run the **''single_example.m''** script in the current
+
directory, using a single Octave computational thread (i.e. one CPU).
+
+
<file bash single_example.slurm>
+
#!/bin/bash
+
#SBATCH --output=<CHANGETHIS>.log
+
#SBATCH --job-name=<CHANGETHIS>
+
#SBATCH --cpus-per-task=1
+
#SBATCH --mem-per-cpu=1000
+
#SBATCH --mail-type=END
+
#SBATCH --mail-user=<CHANGETHIS>
+
+
# Run simulation
+
srun octave --no-window-system single_example.m
+
</file>
+
+
===== Disabling workspace dump file =====
+
+
When running large numbers of Octave simulations from the same directory,
+
an issue arises when these jobs terminate at the same or nearly the same time.
+
In particular, if the jobs are all cancelled at the same time they all attempt
+
to write to the same workspace file causing filesystem problems.
+
+
To avoid this problem, and in general when using Octave, add the following
+
to the top of your Octave script file:
+
+
<code matlab>
+
crash_dumps_octave_core(0);
+
</code>
+
