[Beowulf] multi-threading vs. MPI

[Brian D. Ropers-Huilman]

On Dec 7, 2007 2:41 PM, Toon Knapen <toon.knapen at gmail.com> wrote:
> Greg Lindahl wrote:
> >
> > In real life (i.e. not HPC), everyone uses message passing between
> > nodes.  So I don't see what you're getting at.
>
> Many on this list suggest that using multiple MPI-processes on one and
> the same node is superior to MT approaches IIUC. However I have the
> impression that almost the whole industry is looking into MT to benefit
> from multi-core without even considering message-passing. Why is that so?

Because most of the "real" world does everything in a single node.
Given multi-cores, they have to move down the MT path.

The complications and effort of mixing MT on a node with the MP
between nodes -- absolutely required in our world -- is seen as too
great compared with the performance benefit. Simply staying with
multiple MP tasks, whether they are between nodes or within a node, is
much simpler -- it's what we already do.

No, the question of whether MP as we currently do it will scale to
larger and larger systems remains to be seen, but that's another
thread.

-- 
Brian D. Ropers-Huilman, Director
Systems Administration and Technical Operations
Minnesota Supercomputing Institute                 <bropers at msi.umn.edu>
599 Walter Library                                   +1 612-626-5948 (V)
117 Pleasant Street S.E.                             +1 612-624-8861 (F)
University of Minnesota                               Twin Cities Campus
Minneapolis, MN 55455-0255                       http://www.msi.umn.edu/