on
> Cc: libmesh-devel@lists.sourceforge.net
> Subject: RE: [Libmesh-devel] Recursive partitioning
>
> Ah, well. That clarifies things a bunch. If you've got a decent
> partition for all the cores (ignoring on-node and off-node
> distinctions) from metis/parmetis, perhaps the right way to to
| Phone: (512) 232-7069
Office: ROC 1.405 | Fax: (512) 475-9445
> -Original Message-
> From: Benjamin Kirk [mailto:[EMAIL PROTECTED]
> Sent: Tuesday, July 22, 2008 11:03 AM
> To: Bill Barth; Derek Gaston
> Cc: libmesh-devel@lists.sourceforge.net
> Subject: R
| Fax: (512) 475-9445
> -Original Message-
> From: Derek Gaston [mailto:[EMAIL PROTECTED]
> Sent: Tuesday, July 22, 2008 10:28 AM
> To: Benjamin Kirk
> Cc: "libmesh-devel@lists.sourceforge.net"; Bill Barth
> Subject: Re: [Libmesh-devel] Recursive par
On Tue, Jul 22, 2008 at 10:56 AM, Benjamin Kirk <[EMAIL PROTECTED]> wrote:
> Ultimately I would like to generalize the partitioner interface to work on
> an input iterator range.
>
> You could then do something similar to this...
>
> // partition into NNodes
> Partition(mesh.active_elements_begin()
> It's also unclear yet whether it's worth doing this sort of thing at alll
> given that MVAPICH is already multi-core aware and the on-node communications
> are done via shared memory.
Yeah, that is what I would like to take advantage of. My thought process is
that you may want to group "nearest
Ultimately I would like to generalize the partitioner interface to work on
an input iterator range.
You could then do something similar to this...
// partition into NNodes
Partition(mesh.active_elements_begin(),
mesh.active_elements_end(),
n_nodes);
// partition each subdomai
Definitely interesting numbers. What I find most interesting is that
MVAPICH2 has higher latency than MVAPICH1... any ideas about that?
Do you have an idea about how you would actually implement this using
Metis / ParMetis?
Derek
On Jul 22, 2008, at 9:03 AM, Benjamin Kirk wrote:
> Check ou
Check out attached...
I've been doing some MPI profiling on my 4-socket, dual-core per node
Opteron cluster. I've been curious for a while about "multilevel domain
decomposition" for this class of architectures - e.g.
(1) partition into the number of nodes
(2) partition each subdomain into the nu
