Re: Supercomputers and Dawn

On Nov 22, 1:43 pm, "Jonathan Thornburg [remove -animal to reply]"
<jth...@xxxxxxxxxxxxxxxxxxxxxxx> wrote:
Robert Myers <rbmyers...@xxxxxxxxx> wrote:

[[how to tackle a difficult science/engineering problem]]

Based on my personal experience with large fluid mechanical
simulations and the social behavior of scientists in the business, I
wouldn't trust any conclusion that couldn't be supported with
empirical evidence and calculations that could be performed on an
engineering workstation costing less than $100K.  I would allow for
more expensive computers for data analysis, if necessary.

I find this strategy confusing.  Why $100K, and not (say) $10K, or $1000K,
or some other nice round number?  That is, what is it about computations
that take less than number-of-teraflops/terabytes/etc-that-$100K-will-buy
that makes them inherently more reliable and less prone to tell-the-boss-
-what-she-wants-to-hear problems than calculations (say) 10 times that
size?  Not to mention that $100K N years from now probably will buy a
lot more computation than $100K today.

Having worked for many years in research groups doing both small-scale
and supercomputer simulations, I can well appreciate the "social behavior
of scientists" problems (not to mention "social behavior of managers of
scientists"!) that can arise, but I don't see how putting a rather
arbitrary cap on the size of the computers that can be used for
simulations (though not for "data analysis") necessarily solves
these problems.


The point is to have a price at which many investigators can
participate and many calculations can be performed. If a calculation
is so large that it can be performed only on one or two computers or
only by one or two groups, it is by definition probably wrong, as far
as I'm concerned. If I speak bluntly, it's because that's the only
kind of language some people understand. I just said it, and I mean
no offense to Del: all the money being spent on one-of-a-kind
calculations and one-of-a-kind computers might just as well be put
into cargo containers and dumped into the Pacific.

If the details are so complicated that they can't really be examined
by a competent and unbiased observer with means readily available to a
competent and unbiased observer, they shouldn't be trusted. Would you
want something so important as the warheads in question to be treated
with any less skepticism?

The $100K figure is arbitrary. I figure that if some wingnut who
would be likely to belong to (say) the Union of Concerned Scientists
or IPPNW wants to badly enough, he can scrounge up access to such a
system. Maybe $1M is ok, depending on the resourcefulness of the
wingnut.

Tens of thousands of processors, megawatts of electricity, and acres
of computer center not only build empires, they can also hide
incompetence and poor judgment.

Robert.
.

Relevant Pages

  • Re: Supercomputers and Dawn
    ... simulations and the social behavior of scientists in the business, ... more expensive computers for data analysis, ... of scientists" problems (not to mention "social behavior of managers ... participate and many calculations can be performed. ...
    (comp.arch)
  • Re: Supercomputers and Dawn
    ... simulations and the social behavior of scientists in the business, ... more expensive computers for data analysis, ... and supercomputer simulations, I can well appreciate the "social behavior ...
    (comp.arch)
  • Scientists use of computer models to predict climate change is under attack
    ... Scientists' use of computer models to predict climate change is under ... which uses computers to show fans ...
    (uk.philosophy.humanism)
  • Re: Chez Watt Ballot November 8, 2008
    ... What to look forward to when creationists run ... quarries and anywhere a bit of geology poked it way ... The number of calculations is implementation ... Funny that SETI scientists are called "scientists" when SETI isn't ...
    (talk.origins)
  • Chez Watt Ballot November 8, 2008
    ... What to look forward to when creationists run ... quarries and anywhere a bit of geology poked it way through ... The number of calculations is implementation dependent; ... Funny that SETI scientists are called "scientists" when SETI isn't ...
    (talk.origins)