Cheap Brute Force Attacks

The use of idle cycles on workstations or PCs for brute force attacks
made me think about the economics of this problem. How cheaply could
a massively parallel system be built for key cracking? Much of the
cost and complexity of current massively parallel systems is in
the inter-processor communications and memory systems. Why not build
a system with a large number of single chip processors (ROM/RAM/IO on
one chip) and a simple/slow/cheap communication bus? I'm assuming that
each processor can attack a small chunk of the keyspace without needing
any external support. One or more PC boards could be populated with
an array of cheap processors. There would be a master controller to
assign chunks of key space and check for results. It could also download
the software if it was to be stored in on-chip RAM. A PC board could
be designed that would provide 5V power, a shared serial I/O bus and
a wired-or interrupt for the microprocessors to signal the master
controller that a key has been cracked. What would be a good microprocessor
for this task? It would have to be cheap, reasonably low power with
lots of integer MIPS. A decent amount of on-chip RAM would allow the
software to be downloaded instead of being masked or burned into ROM.
1000 processors could be put into a relatively small box, lets say
10 boards each containing 100 single chip microprocessors. It shouldn't
cost more than $10-$20 per processor, about the same total cost as
one decent workstation ($10K-$20K).


--
John A. Limpert
johnl@Radix.Net