[mvapich-discuss] Re: Running MVAPICH2 with SLURM Process Manager
wei huang
huanwei at cse.ohio-state.edu
Thu May 25 22:06:39 EDT 2006
> * hard nproc 16384
> * soft nofile 4096
> * hard nofile 65536
>
> * hard memlock unlimited
> * soft memlock unlimited
>
This only says you are allowed to set to unlimited. But it doesn't mean
the memlock limit has been set to that value.
>
> # End of file
> [molson] (root) security>
>
> THE SLURM DEAMONS WERE RESTARTED BETWEEN THESE TWO SNIPS
>
> [molson] (slurm) slurm> srun sh -c 'ulimit -l'
> 512
It shows that memlock limit is set to 512kB, which is not enough to run
mvapich2. You need to have the value set correctly so that the command
shows `unlimited'. Here is a brief instruction:
http://nowlab.cse.ohio-state.edu/projects/mpi-iba/mvapich_user_guide.html#x1-350007.2.3
Hope that works.
Thanks.
--Wei
> [molson] (slurm) slurm>
>
>
>
>
>
> wei huang <huanwei at cse.ohio-state.edu>
> 05/25/2006 09:38 AM
>
> To
> Don.Dhondt at Bull.com
> cc
> mvapich-discuss at cse.ohio-state.edu
> Subject
> Re: Running MVAPICH2 with SLURM Process Manager
>
>
>
>
>
>
> Hi,
>
> > The latest suggestion from LLNL is to make sure we have unlimited max
> > locked memory for our MPI tasks with:
>
> > srun sh -c 'ulimit -l'
>
> We wonder if you have tried this command? If things are setting up
> properly, then for every code it should return "unlimited" or the maximum
> size of physical memory you have on each node.
>
> > Below are the latest "traces" of the error.
>
> > 1: cannot create cq
>
> In most cases the "ulimit" is the cause for failing to create cq. So
> please let us know once you've tried it.
>
> Thanks
>
>
> Regards,
> Wei Huang
>
> 774 Dreese Lab, 2015 Neil Ave,
> Dept. of Computer Science and Engineering
> Ohio State University
> OH 43210
> Tel: (614)292-8501
>
>
> On Thu, 25 May 2006 mvapich-discuss-owner at cse.ohio-state.edu wrote:
>
> > As list administrator, your authorization is requested for the
> > following mailing list posting:
> >
> > List: mvapich-discuss at cse.ohio-state.edu
> > From: don.dhondt at bull.com
> > Subject: Re: [openib-general] Running MVAPICH2 with SLURM Process
> Manager
> > Reason: Post by non-member to a members-only list
> >
> > At your convenience, visit:
> >
> > http://mail.cse.ohio-state.edu/mailman/admindb/mvapich-discuss
> >
> > to approve or deny the request.
> >
>
>
>
>
More information about the mvapich-discuss
mailing list