[Beowulf] Three notes from ISC 2006

[Greg Lindahl]

On Thu, Jun 29, 2006 at 10:10:04AM +0200, Joachim Worringen wrote:

> SPEC benchmarking is an example for this. The advantage of SPEC is that you 
> only need one machine to reproduce the results, not a whole cluster.

That's true for SPECcpu. SPEChpc, the MPI codes, require a cluster.
While the current version of these are not commonly run, there is a
new revision in the works which I hope will be more popular.

> But mostly because the vendors want to make sure they get "the right" 
> numbers for their evil whitepapers.

Note that all of the comparison numbers in my evil whitepapers were
run by other people, often system vendors, precisely because it is
important to get the best numbers to compare to. Maybe next time
Patrick will read something before he criticizes it? Better yet,
he could use my whitepaper as a suggestion of codes he could run,
because comparison results are available.

-- greg