©2001 by P. Blaha, K. Schwarz and J. Luitz

Back to:

Problem: Running WIEN2k in k-point and/or mpi- parallel mode under queuing systems like SLURM, PBS, SGE or Loadleveler (LSF)

Analysis: For most queuing systems you can specify only the number of processors you want to use, but not the names of the nodes. They are assigned by the queuing system later on and usually stored in some variables. Wien2k, however, need a ".machines" file for its parallelization, and this has to be generated "on the fly" by the submitted job-script.

A nice overview how to manage parallel calculations in general (and specifically on a cluster using the SGE queuing system) is given in the WIEN2k-notes of the University of Texas.

Basically, one has to write/adapt a few simple csh-lines to create the required .machines file "on the fly". You may also have a look at the "unsupported software pages" , where SRC_mpiutil helps to create a .machines file for PBS.

Fix: Below you find some examples of jobscripts which you may change according to your needs. For other queuing systems (or different versions/setups of the systems listed below), your task is to find out in which variable or file the assigned machine names are stored. With this info it should be trivial to change one of the templates according to your needs.

Peter Blaha,