The big difference is that you can get a larger allocation with the
salloc, spawn a single shell on a login node, and from your shell use
all of the job's resources. You can make that the default as shown
below and explained at
http://slurm.schedmd.com/faq.html#salloc_default_command
SallocDefaultCommand="srun -n1 -N1 --mem-per-cpu=0 --pty
--preserve-env --mpi=none $SHELL"
Quoting Jason Bacon <[email protected]>:
Well, UW-Milwaukee is now officially a SLURM site. SLURM is now the
primary RM on two clusters running CentOS 6.5 (formerly LSF) and
FreeBSD 9.2 (formerly Torque).
One usage question has come up that the FAQ and Google are fuzzy on:
How to open a scheduled interactive shell on a compute node.
The FAQ at https://computing.llnl.gov/linux/slurm/faq.html#prompt
seems to be incomplete, stating
srun -u bash -i
Without --pty, though, editors and the like won't work.
I came across another site suggesting
salloc -n1 srun -n1 -N1 --pty --preserve-env $SHELL
The salloc seems redundant to me, but maybe there's some advantage
I'm unaware of. Wouldn't this
be exactly the same?
srun -n1 -N1 --pty --preserve-env $SHELL
Thanks,
Jason
