Using mpiRAxML on HPC

The parallel version of RAxML has been compiled with GNU OpenMPI, so your MOAB script needs to source that environment. The command to launch mpiRAxML is "mpiraxml" or, as listed in the documentation, "raxmlHPC-MPI." You should specify one more process than the number of CPUs that you designate, since one process is just a 'bookkeeper.' The "-b" option designates the bootstrap seed number, though parallel bootstrap runs are not reproducable. A minimal sample MOAB script to perform 100 bootstraps using ten HPC nodes on a sample DNA dataset is given here:

#!/bin/bash
#MOAB -l nodes=10
#MOAB -j oe
#MOAB -m abe
#MOAB -N RAXML-OPENMPI
source /usr/local/profile.d/openmpi-gnu.sh
mpirun -np 11 /opt/Bio/raxml/mpiraxml -s $HOME/test/subset.phy \
-m GTRMIX -b 12345 -# 100 -n bootstrappedraxml

Notes:
1) For an ultra-fast running program like RAxML often using far fewer nodes (here 10) than the actual number of computations to be performed (here 100 bootstrap replicates) will make for faster execution, given that you specify the backfill queue.
2) The backslash, "\," does not work as a line continuation character in MOAB scripts and is only shown here to denote that all of the command needs to be on a single line.
3) The "-#" bootstrap option is absolutely required for the operation of mpiRAxML!