[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Condor-users] Flocking MPI jobs

On Tue, Mar 08, 2005 at 09:41:55AM +0100, M?rio Costa wrote:
> What I mean is that if I have condor set up to run MPI Jobs in more
> than one pool, and if I have also the pools connected using flocking,
> will condor use the cpus of the other pool if there are not enougth
> available in the pool that the job was submited ??

No. The dedicated scheduler can only use resources that are dedicated 
to it, and a resource can only be bound to one dedicated scheduler. 

You can submit non-MPI jobs from a dedicated scheduler, and those will
flock to other pools, of course.


> On Wed, 2 Mar 2005 12:21:18 -0600, Erik Paulson <epaulson@xxxxxxxxxxx> wrote:
> > On Wed, Mar 02, 2005 at 05:54:42PM +0100, M?rio Costa wrote:
> > > Can any one tell me if it is possible to use condors flocking
> > > mechanisms to run a parallel mpi job on several condor pools ??
> > >
> > 
> > No. MPI jobs need to be submitted from a dedicated scheduler, and that
> > will not flock to another machine.
> > 
> > Now, you can use the "grid" universe in Condor to submit an MPI job to
> > another scheduler - you can use gridtype="Globus" and submit an MPI job
> > to a cluster running any batch system, or gridtype="condor" and submit
> > an MPI job to another Condor pool. That is different than flocking, though.
> > 
> > -Erik
> >