Hi,
Many MPI implementations will have some sort of core binding allocation
policy - which may impact such node sharing. Would these only be limited
to single CPU jobs? Can users request a particular core, for example for
a GPU based job some cores will have better memory transfer rates to the
GPU, so it may be good for users to request these or for SLURM to
allocate these cores?
Regards,
Benson
On 3/19/19 6:26 PM, Peter Steinbach wrote:
I've read through the parameters. I am not sure if any of those would
help in our situation. What suggestions would you make? Note, it's not
the scheduler policy that appears to hinder us. It's about how slurm
keeps track of the generic resource and (potentially) binds it to
available cores. The situation described above is that a non-GPU or
non-GRES job enters a node, runs but doesn't fill the node. Then all
of a sudden SLURM doesn't allow a GPU/GRES job to come it for some
reason.
Best,
Peter
