Training ClusterTranslations: de
In the course we use the training cluster, which is structured similarly to this Vulcan Cluster of HLRS. Here we have collected a small summary of our working environment.
The programs are compiled with a MPI module, which we always load, to have the right environment available:
module load mpi/openmpi
In addition, we need Python for some scripts, which is also available as a module:
module load python
A workspace mechanism is used on Vulcan to store computing data in temporary directories. Such a workspace must be created by everyone for the computation:
ws_allocate cfd 7
For example, this command creates a personal workspace for 7 days. You can display the list of all your own workspaces with ws_list.
The files for the course can be found in /shared/akad-cfd-s/. In the slides we use the variable $COURSE for this path, it therefore makes sense to define this variable in your own profile:
echo 'export KURS=/shared/akad-cfd-s' >> $HOME/.bash_profile
For our own created workspace we use the variable $MYWS, it is helpful to also define this variable in the profile. Overall, the procedure for creating the workspace can be described as follows:
MYWS=`ws_allocate cfd 7` echo "export MYWS=$MYWS" >> $HOME/.bash_profile
Details on the queuing system on Vulcan can be found on the HLRS Wiki.
On the training cluster we have to share the nodes and configure the layout accordingly in the job and submit via the smp queue. This is set via PBS lines in the job script:
#PBS -l select=1:node_type=skl:ncpus=8:mpiprocs=8:mem=20gb #PBS -q smp
A script (named job.pbs for example) is then submitted for example with the following command:
into the batch system. For computations that should run overnight, use the general queue without a reservation.
Interactive jobs can be created using the following command within the reservation:
qsub -q smp -I -l select=1:node_type=skl:ncpus=8:mpiprocs=8:mem=20gb,walltime=02:00:00