OpenMPI

OpenMPI

In order to run parallel jobs across the cluster there needs to be a MPI library installed. The version of openmpi in the yum repositories do not link against the Torque library properly so a rebuild will be required. You can find the latest version of OpenMPI at the main website in the downloads section. This guide is working against version 1.6.4.

Building from source

It is easiest to download and use the provided source RPM. Do not use sudo for any of these rpm commands! These should all be done as the user. When running the rpmbuild command, the single quotes are important in order to force openmpi to use the torque build done in previous steps. This is crucial otherwise MPI jobs will not work correctly or will just fail outright.

cd ~/Downloads/
wget http://www.open-mpi.org/software/ompi/v1.6/downloads/openmpi-1.6.4-1.src.rpm
rpmbuild --rebuild --sign --define='--with-tm' ~/Downloads/openmpi-1.6.4-1.src.rpm

Copy RPMS to the webserver

On the webserver http.cluster.domain, copy the rpms and build the repo.
cd /var/www/html/Cluster_Repo
scp -r [email protected]:rpmbuild/RPMS/x86_64 .
createrepo .