Numbers don't lie...

[Bill Stewart]

At 07:23 AM 2/27/96 -0800, jamesd at echeque.com wrote:
>At 11:18 AM 2/18/96 -0500, A. Padgett Peterson P.E. Information Security wrote:
>>Finally, no cost is allocated to the sustem required to program/evaluate
>>the ponderings of these 100's of ASICs. As anyone who has ever programmed
>>a massively parallel computer (which is what they are talking about in their
>>brute force machine, it is the boundary communications that kill you.
>
>Again:  Interprocess communication is trivial.  A brute force key cracking
>machine is *not* a general purpose massively parallel computer.

Yeah - the interprocess communications includes handing out keyspace
(pretty simple, especially if each chip has a way to find out its name),
broadcasting the cyphertext and known plaintext, letting the winner forward
the successful key, and broadcasting the "you can stop now" message.

Back when the AT&T DSP-32 was new and hundred-CPU machines were exciting,
AT&T made a box that had 128 DSPs in a binary tree; each one had a couple
flavors of RAM and some communications glue.  It would work just fine for this
sort of applications (except that the DSP32 was a floating-point DSP
and what you really need here is big integers.)

And the Inmos Transputer, with its 4 communications channels, would do
just fine as well - you could either build the things into large grids
or build a large ternary tree, or get fancy and build cubes of 6 chips
that you make hypercubes out of, or whatever.  A modern version of the same
chip, built out of FPGAs or ASICs with enough memory and program on chip,
would blaze just fine.


#--
#				Thanks;  Bill
# Bill Stewart, stewarts at ix.netcom.com / billstewart at attmail.com +1-415-442-2215
# http://www.idiom.com/~wcs     Pager +1-408-787-1281