john-users - Re: bitslice MD5

Openwall GNU/*/Linux 3.0 - a small security-enhanced Linux distro for servers

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date: Tue, 13 Nov 2007 09:39:31 +0100
From: Simon Marechal <simon@...quise.net>
To: john-users@...ts.openwall.com
Subject: Re: bitslice MD5

Solar Designer a écrit :
> On page 23, it has a very brief mention that bitslice MD5 "is slow" on
> Cell SPUs: "128 adds require 94 instructions."  To me, 94 instructions
> is not a lot, but perhaps a straightforward 4-way SIMD implementation is
> more efficient.

As MD4 and its successors have been designed with 32bit efficiency in 
mind it's no wonder that they work well that way. The CELL processor is 
a beast and a naive (using a lot of registers ...) MD4 implementation 
leads to 180M hashes/s. Using the "guess part of the password, reverse a 
part of the process" trick it's possible to go way past the 200M 
hashes/s symbolic limit.

-- 
To unsubscribe, e-mail john-users-unsubscribe@...ts.openwall.com and reply
to the automated confirmation request that will be sent to you.