LAM's mpirun N/C notation should allow you to do whatever you want.
See the man page for mpirun for details, especially the section
entitled "Location nomenclature".
I believe that this stuff is described in the User's Guide as well.
On Apr 18, 2007, at 5:21 AM, Yuan Wan wrote:
> Hi all,
>
> I want to run a large scale MPI job on a 4-way SMP cluster.
> I don't know how to control CPU-process mapping in lam.
> i.e.
> ModeA ModeB
> ----------------------------------------
> process00 node0 node0
> process01 node1 node0
> process02 node2 node0
> process03 node3 node0
> process04 node4 node1
> process05 node5 node1
> process06 node6 node1
> process07 node7 node1
> process08 node0 node2
> process09 node1 node2
> process10 node2 node2
> process11 node3 node2
> ... ... ...
> process30 node6 node7
> process31 node7 node7
>
> Anyone know how to do that?
> Many Thanks
>
> --Yuan
>
>
> Yuan Wan
> --
> Unix Section
> Information Services Infrastructure Division
> University of Edinburgh
>
> tel: 0131 650 4985
> email: ywan_at_[hidden]
>
> 2032 Computing Services, JCMB
> The King's Buildings,
> Edinburgh, EH9 3JZ
>
> _______________________________________________
> This list is archived at http://www.lam-mpi.org/MailArchives/lam/
--
Jeff Squyres
Cisco Systems
|
