Thanks for this fast response.

According in what I read about open-mpi and slurm the two stages procedure
would be:

salloc -N2 bash
Then
mpirun myapp

With the procedure you provided user gets:
srun: error: Unable to create job step: Requested node configuration is not
available


On Sat, Jul 28, 2012 at 4:53 PM, Smith, Jerry Don II <[email protected]>wrote:

>  If you want to do it "interactively" with salloc do it in 2 stages.
>
> salloc -N2
> Then
> srun --mpi=openmpi -n16 myapp ( if you have 2x8 core nodes. )
>
>  *From*: Diego Lendoiro [mailto:[email protected]]
> *Sent*: Saturday, July 28, 2012 08:30 AM
> *To*: slurm-dev <[email protected]>
> *Subject*: [EXTERNAL] [slurm-dev] openmpi+slurm
>
>  Hello everybody;
>
>  I've been setting up slurm into a 4 node cluster with 8 cores per node.
> One of my users is trying to launch some openmpi application with:
>
>  salloc -N2 mpirun myapp
>
>  He claims that the application is running one instance for each
> node (using 8 cores each) instead of running only one instance that uses 16
> cores I'm no expert on openmpi but maybe I forgot to add some config into
> slurm.conf
>
>  MPI related stuff on slurm.conf:
>
>  MpiDefault=none
> MpiParams=ports=12000-12200
>
>  And yes, I'm quite new with slurm, read a bunch of documentation but
> right now I'm kind lost with this issue.
>
>  Cheers.
>
>  --
> Diego Lendoiro
>
> Inestable GNU/Linux Users Group Member | http://www.inestable.org
>
> In the beginning was the command line
>
>


-- 
Diego Lendoiro

Inestable GNU/Linux Users Group Member | http://www.inestable.org

In the beginning was the command line

Reply via email to