I'm trying to execute CASTEP on a cluster using TORQUE. Due to our strict cluster configuration we do not wish to open a Gateway, thus users must submit their job manually through a submit file built by hand. I've been able to find the environment variable (DSD_MachineList) to set so that DMol3 is executed in parallel (on more than one compute node), but this doesn't work for CASTEP as proven again by the mpd.hosts file created by the job (which was fine for DMol3 after setting the DSD_MachineList environment variable).
Another (maybe) unrelated problem is that the job is executed is the \$HOME directory whatsoever the change directories made in the submit script (again this was fine with DMol3).
Anybody ran in this kind of problem before? Will this happen again for any other module (Discover, Forcite, Mesocite, GULP, etc)? Would it be possible to have a complete list of the environment variables used by each module since walking the perl files is clearly impossible? We are using Materials Studio 6.0.
Regards,
François-Michel De Rainville
Calcul Québec - Université Laval