Other Directory Sites: SeekWonder | Directory Owners Forum
The following archive was created by hippie-mail 7.98617-22 on Fri Aug 21 1998 - 17:20:06 ADT
Mike Rosing (eresrch@msn.fullfeed.com)
Fri, 3 Jul 1998 13:45:19 -0500 (CDT)
On Fri, 3 Jul 1998, Mok-Kong Shen wrote:
> The belief in sort of unconditional need of supercomputers stems
> in my opinion from (1) lobbying by manufacturers (2) prestige
> gain of owning very expensive super-modern equipments (national
> centres, etc.) and (3) personal perspectives of persons working
> at supercomputing centres. Back at the time where 8086 processors
> were common for PCs, I used a benchmark program to show that
> the computing power of one of the then modern CYBER computers
> was equivalent to only about 100 PCs. But no one in my environment
> wanted to 'know' that result.
But there are many applications for teraflop computers. I can build
them in my basement, but it still cost $100,000 in parts and $1M in
engineering time. Single large scale machines are useful for large
scale problems. The trick is to build the machine around the problem.
The manufactures need to recover their engineering costs and want
to sell as many machines as possible. So they try to build "generic"
machines. They all go out of business after a while.
Example large scale machines are code cracking, geological analysis,
ASIC simulation and board layout. lots of stacked up PC's can do the
job, but it's an expensive solution compared to specialized equipment.
That is if you take engineering time into account over the life of
the equipment.
A 100 PC's isn't the same as one large/fast machine. There are some
tasks which the distributed system can do as well as the big one,
but there are other tasks where it can't hope to match the ability.
To say that one is better than the other is to not understand the
problems that each can solve well. That's why I say build the machine
around the problem, it will be specialized to solve only one problem,
but it will to it better than anything else can. If it's a big enough
problem, you'll have lots of users and the machine can be paid for.
And of course, you'll need strong crypto to protect the users data!
Patience, persistence, truth,
Dr. mike
The following archive was created by hippie-mail 7.98617-22 on Fri Aug 21 1998 - 17:20:06 ADT