🚀 go-pugleaf

In article <1992Jul31.051715.25994@convex.com>, tighe@convex.COM (Mike Tighe) writes:
|> fouts@hpl.hp.com writes:
|> 
|> > Sigh.
|> 
|> Sigh.
|> 
|> > 1) Supercomputers are not widely used in command and control situations
|> >   because they don't have the high mean time to failure.
|> >
|> > 2) Most US supercomputers are used for research which would probably
|> >   stop in the kind of war being suggested above, and certainly not
|> >   critical to the prosecution of such a war.
|> 
|> I don't mean to sound harsh, but unless you have actually worked in those
|> [black] areas, you are not really qualified to draw those conclusions, are
|> you? And if you have worked there (as I have), then you know you have to
|> keep your mouth shut and smile. :-)
|> 

Of course I am:

1) Estimate the number of supercomputers produced.

   This can be done by reading the annual financial statements
   from supercomputer companies and making educated guesses about
   their production costs.

2) Subtract the number of supercomputers know to be visible to the public

This gives you an estimate of the number of supercomputers not visible,
and therefor used in "black" projects.

Then,  use your favorite issue of Aviation Leak to estimate the black
budget for command and control, not all of which is black anyway.

Notice that the number of supercomputers times the best guess price
for supercomputers is small compared to the command and control budget
for machine resources.

Make acceptable general statement above.

[By the way, I am told that people who have or have had security
 clearances are required to limit knowledge of the existance and
 extend of their access.  If I had ever worked in the black,
 I certainly wouldn't be saying so in a group read by the NSA. (;-)]

|> Anyway, there is at least one system that is public knowledge. In fact, it
|> was the topic of the keynote speech at the Banquet during Supercomputing
|> 91. It was an interesting talk, since it gave a rare public glimpse of how
|> the US uses supercomputers [for military purposes] to solve real problems.
|> 

1 does not counter the claim "not widely used."  Compared to the
entire command and control structure, the entire output of the US
supercomputer industry doesn't counter the claim. (;-)

|> > 3) Not buying supercomputers for command and control does not imply
|> >    not buying supercomputers for research.
|> 
|> I agree here.
|> 
|> > Just because the NRO needs to function in a war doesn't mean NASA
|> > can't buy a supercomputer for computational chemistry.
|> 
|> ditto.
|> 
|> Remember though that any "National Security" argument used to justify
|> buying domestic computers can also be used by other countries for not
|> buying US computers, and using their own home-grown product.
|> 

Exactly.  The whole problem with tradewars is that the other side can
always use the same argument.  Most US computer companies sell 40-60%
of their product off shores, so any foreign sale prevented in the US
can easily be countered in the other country.

|> And one other point. Content. Many of those chips (except Convex GaAs!),
|> disk drives, etc. all come from the Land of the Rising Sun. So most of
|> those domestic computers are not all that home-grown after all.
|> -- 

Unless Convex has changed their machine content a lot in the last
three years, Convex machines also have a high part dependency on off
shore components, especially for periphreals.  Unless you build a PC
with Connor disk drives, no crt or floppy, and memory from Micron,
you can't market a machine without some off shore content. (;-)

|> Mike Tighe, (214) 497-4206
|> tighe@convex.com

-- 
Martin Fouts                  |  M/S 1U-14
Member of Technical Staff     |  HP Laboratories
EMAIL:  fouts@hpl.hp.com      !  MAIL:  P.O. Box 10490
PHONE:  (415) 857-2971        |  SHIP:  1501 Page Mill Road
  FAX:  (415) 857-8526        |  Palo Alto, CA 94304-1126

novae erratae!