lindahl at keyresearch.com
Fri May 3 13:58:28 PDT 2002
On Fri, May 03, 2002 at 09:54:30AM +0200, Joachim Worringen wrote:
> Regarding the placement of the process (or: assignment of ranks) on
> SMP-nodes is part of the peformance strategiy that an MPI library is
> free to follow. It seems that ScaMPI does a straight round-robin mapping
> of the process-to-node-ranks, while MPICH-GM "groups" processes on the
> same node. Both approaches are, of course, valid.
If a benchmark measures process-to-process bandwidth and latency
between only 2 processes, it matters whether or not they are on the
same node (shmem) or not. It's not a matter of round robin vs. groups.
BTW, groups _are_ strongly preferred, but that's not the issue. Why?
If you have a program that does nearest neighbor communication, where
are your neighbors? With round robin you are accessing your array in
the wrong order.
More information about the Beowulf