Topcluster listing

Steven Timm timm at
Thu Sep 21 07:15:14 PDT 2000

> Greg Lindahl wrote:
> > 
> > > I suspect that the institutions that have the biggies are too busy with
> > > "real work" to run a benchmark on the cluster -
> > 
> > Actually, you'll find that "the biggies" in HPC are quite eager to get on
> > the Top 500 list. It's the industry biggies (google, Incyte) that definitely
> > don't care.
> I tried to get management to run linpack on our "clusters" (OK, so
> they're really farms, but...) at Fermi.  We couldn't take the time.  We
> had to get the work done.  Reality often gets in the way of desire.
> Dan
As Dan mentioned in a previous post, Fermilab does run seti at home
by way of burn-in, but nowadays we also use application-specific
benchmarks, measuring the performance of our machines on the 
actual code that they have to run.  Given the nature of experimental
high-energy physics, each node is decoupled from the others.  
The nature of the Fermilab problem is well described in Pfister's
book "In Search of Clusters" although at the time he wrote it
we were still an SGI/IBM RS 6000 shop and have since shifted
to Intel/Linux for our compute power.

A new effort at Fermilab is to use Linux-based machines to do
theoretical lattice QCD calculations.  This is a program that
is coupled across many machines.

Steven C. Timm (630) 840-8525  timm at
Fermilab Computing Division/Operating Systems Support
Scientific Computing Support Group--Computing Farms Operations

More information about the Beowulf mailing list