From: jim bell <jimbell@pacifier.com>
Date: Sat, 27 Jul 1996 07:47:02 +0800
To: Remo Pini <cypherpunks@toad.com
Subject: Re: Distributed DES crack
Message-ID: <199607261723.KAA16305@mail.pacifier.com>
MIME-Version: 1.0
Content-Type: text/plain


At 01:28 PM 7/26/96 +0200, Remo Pini wrote:
>-----BEGIN PGP SIGNED MESSAGE-----
>
>To: cypherpunks@toad.com
>Date: Fri Jul 26 13:25:22 1996
>> At 10:30 7/23/96, Matt Blaze wrote:
>> 
>> >My estimate is that an FPGA-based machine that can do a single DES key
>> >every four months (eight months to exhaust the whole keyspace) could
>> >be built with off-the-shelf stuff for comfortably under $50k (plus
>> >labor, plus software development costs).  A prototype board should
>>  cost
>> >under $1000 and will help prove the concept and get a more accurate
>>  cost
>> >estimate.  I expect to build such a prototype machine myself, and, if
>>  it
>> >works as I expect, maybe the whole thing.
>> 
>> I am willing to financially contribute to the project.
>> 
>> 
>If this were to be a card (via RS232 or PC-bus), thousands of people would 
>be able to copy it, once the development process is finished. -> You'd have 
>hardware that all those people could use for a distributed crack, the 
>building cost would be distributed also (<$100), only development would 
>have to be at one place (sponsored of course). Now, that would be a scary 
>thought for DES-fans!

I've proposed that if it were done in this way, the circuit should be built 
external to a PC-clone or other computer, so that it can be easily tiled on 
a large pcb, in an "n by m" array.  The reason is that if an individual 
cracker module were as simple and cheap as it should be, a person could 
easily want to run dozens if not hundreds of them. Communication would 
probably be done with a single serial data bus, with each module 
individually addressed.  Due to the nature of the DES crack, communication 
would be rare, so it's likely that an ordinary '386 or '486-based computer 
could handle all the communication for a large number of such modules.

 I don't know if the figure of $10 thrown around for an FPGA is accurate, 
but if it is then a cost of $30 for each subsystem is probably doable, 
including pc board, assembly, and a few other components.   A cost of $3000 
for a 10-by-10 array seems reasonable to me, particularly since the 
throughput of each of those FPGA's ought to be at least 10x that of a 
general-purpose PC in this application. 

Jim Bell
jimbell@pacifier.com