Re: why mainframes are still used?

From: Mike Bartman (omni_at_foolie.omniphile.com)
Date: 09/08/04 

Date: Wed, 08 Sep 2004 11:54:26 -0400

On 7 Sep 2004 23:14:35 -0700, spamsink2001@yahoo.com (Alan E. Feldman)
wrote:

>Uh, you're saying that PC's are supercomputers? Maybe poopercomputers.
>:-) They're not so great when they crap out on you! (That's the
>"technical" variant of "crap out".)

That's probably why the boot-up time is one of the numbers reported
when testing new PCs. ;-)

"Supercomputer" is a term used very loosely by some people. I
remember when IBM claimed the 3090 was a "supercomputer"...despite the
fact that it maxed out at about 25 MFLOPS at a time when a Cray X/MP
48 could hit 890 MFLOPS, and when a 1 nanosecond decrease in the Cray
clock cycle time (from 9.5 ns to 8.5 ns) resulted in a 56 MFLOPS speed
increase...that's equivalent to *two* 3090s. Claiming a 3090 was a
"supercomputer", which the Cray unquestionably was, would be like
entering a kid's go-cart in the Indy 500. It's laughable.

These days what is called a "supercomputer" is usually a massively
parallel array of fairly fast CPUs. They quote speeds by summing up
the FLOPS rating of all the CPUs, and come up with numbers in the
TFLOPS range. These machines are based on the concept that if one
woman can have a baby in 9 months, 9 women can have a baby in 1
month...this isn't always true. ;-)

If your algorithm is 100% parallel, then more processors will get you
your answer sooner. If it has *any* sequential aspects, then there
will be a declining effect from adding more processors to the problem,
until eventually you are doubling the processor count for a fraction
of a percent decrease in run time...or worse. This is called "Amdahl's
Law" if I remember my schooling correctly.

The curve is pretty steep too, even for processing which is very
parallel. I think we figured out once that for a 96% parallel problem
the most processors you could effectively use was about 15. You got
very little decrease in run time from going to 30 processors for
instance...it wasn't cost-effective. With less parallel problems it
gets worse. Much worse.

Systems with thousands of processors may have impressive cumulative
TFLOPS numbers, but they are only good for running 100% parallel
algorithms. There are many such problems, but there are even more
where there is some sequentialness to the situation. For those you
really need a small number of really, really fast CPUs...or a better
algorithm that requires less calculating to start with.

Fast individual CPUs were what Cray got famous for. I once ran a
Fortran program that did antenna analysys on an 11/785 (new at the
time... ;-) and then on a Cray X/MP 12 (1 processor machine). The
*run* time for the executable image on the VAX was about 90 seconds
with the data I had. On the Cray the total time to ship the source
over from the VAX, have it compiled, linked, executed, and the results
and listings shipped back to the VAX and stored on disk was under 8
seconds...

I don't think we have any machines anymore that can truely be called
"supercomputers" in the sense that their CPUs are blindingly fast
compared with the average CPU that is generally available. Seymore's
rule was that if his current design wasn't going to be at least 10
times faster than the fastest thing available on the planet, he was
wasting his time and should start over.

It was fun working for Cray Research in the 80s...

-- Mike B.
-- Mike B.

'04 FLSTCI

----------------------------------------------------------------
  To reply via e-mail, remove the 'foolie.' from the address.
  I'm getting sick of all the SPAM...
----------------------------------------------------------------

Relevant Pages

  • Re: why mainframes are still used?
    ... We don't know how many CPUs the 3090 was using. ... which claimed this 25 MFLOPS machine was a supercomputer. ... was usually Cray. ... warrant the cost. ...
    (comp.os.vms)
  • Re: IBM tops supercomputing table
    ... > A supercomputer being built by IBM for the US Department of Energy ... > has topped a list of the world's fastest computers. ... dust by cheap clusters of off the shelf silicon CPUs. ...
    (sci.physics)
  • Re: IBM tops supercomputing table
    ... >> A supercomputer being built by IBM for the US Department of Energy ... > dust by cheap clusters of off the shelf silicon CPUs. ... > Supercomputer applications generally require hardware configured to ... Open source Linux clones written without bias ...
    (sci.physics)
  • Re: Another AMD supercomputer, 13,000 quad-core
    ... I suspect that there is a very small number of customers for Cray, ... Cray or the high-capabilty (real supercomputer?) one? ... Commerce assigned a dumping margin of 454% for NEC supercomputers. ...
    (comp.sys.ibm.pc.hardware.chips)
  • Re: Suzanne Gildert on Quantum Computing in Teleplace
    ... I'm bloody sure you won't either. ... "Jaguar is a petascale supercomputer built by Cray at Oak Ridge National Laboratory in Oak Ridge, ... We also know that specialist hardware used by people like the NSA tends to lead personal computer performance by around 50 years. ...
    (uk.politics.misc)