Pierre Lindenbaum
2014-07-07 15:37:31 UTC
Hi all,
I'm learning slurm.
I'm coming from SGE where I used to run my jobs using 'qmake' a
parallelized version of GNU-make for SGE.
That was very simple because we just had to specify the number of
parallel jobs and that's all.
$ qmake -l arch=lx24-amd64 -cwd -v PATH -- -j 5
now we're running slurm. We've patched GNU-make with (
http://slurm.schedmd.com/faq.html#parallel_make ) and I've been told
that I should first run salloc. Something like:
| $ salloc--nodes=8 --ntasks-per-node=10|
I'm still puzzled about using it. If my makefile calls itself some
multithreaded program, how should I config salloc ? If too many
resources are reserved, shall I prevent the other users from running
their jobs ? Furthermore, if I run another 'salloc+make', I can see
salloc stalled and waiting for resources to be released while some
nodes clearly don't run many jobs. Can't we tell slurm to run jobs as
they appear in the queue ?
Thank you for your help.
Pierre
I'm learning slurm.
I'm coming from SGE where I used to run my jobs using 'qmake' a
parallelized version of GNU-make for SGE.
That was very simple because we just had to specify the number of
parallel jobs and that's all.
$ qmake -l arch=lx24-amd64 -cwd -v PATH -- -j 5
now we're running slurm. We've patched GNU-make with (
http://slurm.schedmd.com/faq.html#parallel_make ) and I've been told
that I should first run salloc. Something like:
| $ salloc--nodes=8 --ntasks-per-node=10|
I'm still puzzled about using it. If my makefile calls itself some
multithreaded program, how should I config salloc ? If too many
resources are reserved, shall I prevent the other users from running
their jobs ? Furthermore, if I run another 'salloc+make', I can see
salloc stalled and waiting for resources to be released while some
nodes clearly don't run many jobs. Can't we tell slurm to run jobs as
they appear in the queue ?
Thank you for your help.
Pierre