[Beowulf] doubt about job queuing systems

Mark Hahn hahn at mcmaster.ca
Fri Apr 17 11:15:50 PDT 2009


> When I argue that many other computational electromagnetics programs such as
> FEKO <http://www.feko.info> can run in batch mode without any X output and
> can be sheduled directly by external job queueing system, engineer from CST
> declares that kind of  program not starting a GUI is old-fashioned...
>
> Are they telling the truth, or just lying?

normally, it's the solver that uses MPI, not the gui.  so the gui 
will set up jobs, and submit them to the cluster's queue, but itself
really only runs on a "login" node.  this is mainly motivated by a 
concern for the user's workflow, as far as I can tell - desire to make
it as IDE or "studio"-like as possible.

in the several similar packages I've integrated, there are usually
some sort of hooks available for determining the exact commandline 
for submitting a job.  after all, even if assuming LSF, a particular 
site will often have idiosyncratic parameters and switches.  normally
these hooks can be transformed into using completely different 
submission commands (though in the worst case, you could just write
a quick "bsub" emulator that did the right thing for your cluster.)

regards, mark hahn.



More information about the Beowulf mailing list