Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Tue, 2 Aug 2011 11:31:56 +0400
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: GPU patches (was: cryptmd5cuda)

Lukas,

On Tue, Aug 02, 2011 at 09:11:08AM +0200, Lukas Odzioba wrote:
> > This is puzzling. As far as I'm aware, they're supposed to be faster on
> > NVidia as well. How did this change affect code size? Did the code
> > size reduce with the fewer-ops F and G functions?
> Yes, there is reduction in ptx code (http://ideone.com/X6RsH)
> 
> > Perhaps you have some other bottleneck that you're hitting.
> I tried to test speed only F and G, and the difference in speed was even bigger.

My guess is that you should increase the number of threads in order to
cover the latencies when you optimize F and G to have fewer instructions.

> > That's OK, but it's 3 times slower than the benchmark you have posted on
> > the wiki, right? So is there a 3x performance hit for actual cracking
> > as compared to --test?
> Yes, and I need to find why.

OK.  Since KoreLogic's contest starts in just 2 days, can you perhaps
re-focus on SHA-crypt now, especially the SHA-512 flavor of it?  My
guess is that we'll see some of these hashes in the contest, and the
only alternative to your implementation appears to be to use JtR's
generic crypt(3) code, which is relatively slow (since the crypt_r()
code in glibc cannot use SIMD).  So you can make a bigger difference for
the contest by optimizing the two SHA-crypt's than by optimizing
MD5-crypt or phpass (where we have more/better alternatives).  Well,
phpass is also good to optimize since it's simpler and thus your code's
efficiency is greater.

So perhaps focus on more extensive testing of your patches for
sha512crypt, phpass, and sha256crypt (in that order, although I am just
guessing what we'll see in the contest, and I might be wrong).  For
testing phpass, use the hashes from the JtR test suite that JimF and
magnum have been working on.  For testing SHA-crypt, perhaps generate a
thousand of test hashes of your own by calling crypt(3) on a Linux
system with recent glibc.

Thanks,

Alexander

Powered by blists - more mailing lists

Your e-mail address:

Powered by Openwall GNU/*/Linux - Powered by OpenVZ