Dolphin Wulfkit

Greg Lindahl lindahl at keyresearch.com
Mon May 6 18:06:31 PDT 2002


On Tue, May 07, 2002 at 12:30:58AM +0200, Steffen Persvold wrote:

> Greg,
> 
> Let's not turn this one into a flamewar, ok ?

As far as I'm concerned, it's over, but I'm sure the same issue will
come up again; that's the way life is.

> However, it's not the 2 process results that interest me, it's
> the larger ones (which test the scaling possibilities of the MPI
> implementation and the interconnect).

I would disagree with this sentence. Large MPI applications which I
have used have not used things like collective operations. So it's
easy to stare at a big set of microbenchmarks and not learn anything
about the scaling of an MPI application. Bisection bandwidth, now
that's interesting.

greg




More information about the Beowulf mailing list