On Mon, 5 May 2003, Rohan Inamdar wrote:
> Consider following hostfile (used with lamboot)
Appologies for taking so long to answer; it's getting hectic here trying
to release LAM 7.0. :-)
> #hostfile for lamboot
> sparc7.cs.iit.edu
> sparc8.cs.iit.edu
> sparc9.cs.iit.edu
>
> Consider following MPI_Info object which contains only one key
>
> MPI_Info_create(&info);
> MPI_Info_set(info, "lam_spawn_sched_round_robin", "n1");
>
> (Ref: http://www.lam-mpi.org/MailArchives/lam/msg02641.php)
>
> And I spawn only one process using this info object like this
>
> MPI_Comm_spawn("./test_tree", param, 1,
> info, 0, MPI_COMM_SELF, &everyone,
> MPI_ERRCODES_IGNORE);
>
> 1. On which machine/host/node is the process spawned if i run the
> program from sparc7 (i.e. first machine in the hostfile)
> 2. From sparc8 (second machine in the hostfile)
> 3. And from sparc9 (third machine)
>
> I believe in all three cases new process will always be spawned on n1,
> which corresponds to sparc8.cs.iit.edu.
>
> Is my understanding correct?
Yes.
LAM's MPI_COMM_SPAWN currently only has the capability to spawn on
absolute nodes, not relative nodes (i.e., "start on the first node after
'my' node").
--
{+} Jeff Squyres
{+} jsquyres_at_[hidden]
{+} http://www.lam-mpi.org/
|
