Cluster benchmark(s)?

[Randy_Howard at Dell.com]

Well, my intent was not to establish specific numbers but rather to 
get an idea of "bang for the buck" factors with various hardware
configurations.  For example, I wonder if there is a way of 
predicting up front for a given application whether or not 10/100
ethernet would be sufficient and not become the primary bottleneck.
I understand it is a very complex problem and this may not even 
be possible.

> -----Original Message-----
> From: Putchong Uthayopas [mailto:pu at ku.ac.th]
> Sent: Tuesday, January 07, 1997 1:21 AM
> To: Jared Hodge; Randy_Howard at exchange.dell.com
> Cc: beowulf at beowulf.org
> Subject: Re: Cluster benchmark(s)?
> 
> 
> Hi,
> 
> I agree that real performance of beowulf depends on your application.
> Anyway, if you need to get some ideas about system 
> performance . People
> usually run slbench ,
> parallel benchmark from UTK. Please obtain it by going to 
> ATLAS project.
> Net pipe and net perf seems to be a standard for testing network
> performance.
> All these benchmark and more can be found at.
> 
> http://www.netlib.org/benchweb/
> 
> Putchong.
> 
> ----- Original Message -----
> From: "Jared Hodge" <jared_hodge at iat.utexas.edu>
> To: <Randy_Howard at dell.com>
> Cc: <beowulf at beowulf.org>
> Sent: Wednesday, January 17, 2001 3:13 AM
> Subject: Re: Cluster benchmark(s)?
> 
> 
> > Randy,
> > That's kind of a funny question.  Not funny as in Ha Ha, 
> but funny as
> > in strange.  It's a good question, but the problem is that 
> people who
> > use and build Beowulf clusters (being the informed, 
> scientific type) do
> > so for specific applications which very from cluster to 
> cluster.  These
> > applications (which are the only real benchmark) very as to 
> weather they
> > are CPU limited or network latency or bandwidth limited, or 
> limited by
> > something else (like disk space or memory).  That means if 
> your going to
> > run a generic benchmark on a cluster, you need to run 
> benchmarks that
> > test these components and let the user/developer weigh the different
> > portions that they see as important.  For cpu limited 
> benchmark, there
> > is tons of stuff, but I think the parallel POVRAY ray 
> tracer is a common
> > tool for comparison.  For network latency, there are 
> various versions of
> > ping pong tests and such that can be used, and for 
> bandwidth, well, you
> > could just use ftp, although that's not really in context, 
> because not
> > to many parallel programs actually call ftp, but it should 
> give you a
> > rough idea.  I'm by no means an authority on any of this, 
> so maybe you
> > can get some input from the people on the list who are more
> > experienced.  Like I said though, I'm sure a lot of them 
> will give you
> > the generic answer that the only true benchmark is your application.
> >
> > Randy_Howard at Dell.com wrote:
> > >
> > > I am curious if there is any reference benchmark software 
> for a Beowulf
> > > cluster to test the networking and processing 
> capabilities.  More for
> > > comparitive testing between various configurations than 
> meaningful data
> > > values.  Any pointers you have would be very helpful.
> > >
> > > Randy
> > >
> > > _______________________________________________
> > > Beowulf mailing list
> > > Beowulf at beowulf.org
> > > http://www.beowulf.org/mailman/listinfo/beowulf
> >
> > --
> > Jared Hodge
> > Institute for Advanced Technology
> > The University of Texas at Austin
> > 3925 W. Braker Lane, Suite 400
> > Austin, Texas 78759
> >
> > Phone: 512-232-4460
> > FAX: 512-471-9096
> > Email: Jared_Hodge at iat.utexas.edu
> >
> > _______________________________________________
> > Beowulf mailing list
> > Beowulf at beowulf.org
> > http://www.beowulf.org/mailman/listinfo/beowulf
> 
>