Report on SMP issues

Douglas Eadline deadline at plogic.com
Mon Mar 24 13:14:34 PST 2003


On Mon, 24 Mar 2003, Joachim Worringen wrote:

> Douglas Eadline:
> 		> All
> >
> > I have just posted some results and discussion concerning SMP nodes
> > and clusters.
> >
> > You can read and comment on the report at:
> > http://www.cluster-rant.com/article.pl?sid=03/03/24/1731246
> 
> It's nice to have such a comparison -  but it does not make much sense to 
> *only* compare speedups - you should add the wall clock times for the test 
> runs, too.

Hmm, the reason I used speed-up's was to normalize the
results for all the programs. I'll re-run the tests and
collect the wall clock times at some point. Although I am not sure
how it changes the observations or "general" conclusions.
I have always been curious about running two copies of
the same program on a dual node and the effects it has on
scalability. (i.e. running the NAS suite on 4 dual nodes
with n=4 and running the NAS suite on 4 dual nodes with n=8).
In general the scalability is poor for the n=8 case.


Doug
> 
>  Joachim
> 
> 

-- 
-------------------------------------------------------------------
Paralogic, Inc.           |     PEAK     |      Voice:+610.814.2800
130 Webster Street        |   PARALLEL   |        Fax:+610.814.5844
Bethlehem, PA 18015 USA   |  PERFORMANCE |    http://www.plogic.com
-------------------------------------------------------------------




More information about the Beowulf mailing list