[slurm-users] ntasks and cpus-per-task

Christopher Benjamin Coffey Chris.Coffey at nau.edu
Thu Feb 22 13:57:54 MST 2018


Hi Loris,

"But that's only the case if the program is started with srun or some     form of mpirun.  Otherwise the program just gets started once on one     core and the other cores just idle."

Yes, maybe that’s true about what you say when not using srun. I'm not sure, as we tell everyone to use srun to launch every type of task.

Best,
Chris
—
Christopher Coffey
High-Performance Computing
Northern Arizona University
928-523-1167
 

On 2/22/18, 8:25 AM, "slurm-users on behalf of Loris Bennett" <slurm-users-bounces at lists.schedmd.com on behalf of loris.bennett at fu-berlin.de> wrote:

    Hi, Other Chris,
    
    Christopher Benjamin Coffey <Chris.Coffey at nau.edu> writes:
    
    > Loris,
    >
    > It’s simple, tell folks only to use -n for mpi jobs, and -c otherwise (default). 
    >
    > It’s a big deal if folks use -n when it’s not an mpi program. This is
    > because the non mpi program is launched n times (instead of once with
    > internal threads) and will stomp over logs and output files
    > (uncoordinated) leading to poor performance and incorrect results.
    
    But that's only the case if the program is started with srun or some
    form of mpirun.  Otherwise the program just gets started once on one
    core and the other cores just idle.  However, I could argue that this is
    worse than starting multiple instances, because the user might think
    everything is OK and go on wasting resources.
    
    So maybe it is a good ideas to tell users that if they don't know what
    MPI is, then they should forget about multiple tasks and just set
    --tasks-per-cpu for the default 1 task.  That way I wouldn't get users
    running a single-threaded application on 100 cores (the 1000-core job
    got stuck in the queue :-/ ).
    
    I think I'm convinced.
    
    Cheers,
    
    Loris
    
    [snip (53 lines)]
    
    -- 
    Dr. Loris Bennett (Mr.)
    ZEDAT, Freie Universität Berlin         Email loris.bennett at fu-berlin.de
    
    



More information about the slurm-users mailing list