Openwall GNU/*/Linux - a small security-enhanced Linux distro for servers
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun, 23 Jun 2013 23:56:22 +0400
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: Re: Sayantan's weekly report #2

On Mon, Jun 24, 2013 at 01:14:33AM +0530, Sayantan Datta wrote:
> On Monday 24 June 2013 01:06 AM, Solar Designer wrote:
> >Also, I think we should include an optimization in some/all kernels,
> >where a consecutive range of ASCII codes (and beyond, for 8-bit chars)
> >would be specified.  Thus, there would be two ways to pass a charset
> >spec for a char position into the kernel: either list of chars (need one
> >byte per char), or range of chars (two bytes to specify the entire range).
> Yes indeed. This would no longer require looking into local memory.

Right.

> We 
> can increase 32 chars per range to 48 - 64 chars. Would that be sufficient ?

I think for the lists of chars (not consecutive ranges), it'd be sort of
OK to have the max at 62 chars (or you can make it 64 if it's same cost),
and the number of ranges at max 3.

> one example:

Thanks!

> sayantan@...n:~/JohnTheRipper/run$ ./john 
> --mask='[1234567m]?l?l?l[123a4567]?l[12345o67][1234567o]' 
> -format=descrypt-opencl testsingle
> Device 0: Tahiti (AMD Radeon HD 7900 Series)
> Loaded 1 password hash (descrypt-opencl, traditional crypt(3) [DES OpenCL])
> Press 'q' or Ctrl-C to abort, almost any other key for status
> missadoo         (u16-DES)
> 1g 0:00:00:11 0.08583g/s 16876p/s 540039c/s 540039C/s 1iss1doo..MAXLENGT
> Use the "--show" option to display all of the cracked passwords reliably
> Session completed
> 
> In this case range 0,4,6,7 is computed on GPU. Also the reported c/s is 
> 4096/32 = 128x slower.

You aren't properly updating *pcount yet?  Where do the numbers 4096 and
32 come from?  (I can try to guess, but I might as well guess wrong.)

Why does your code put ranges 0,4,6,7 on GPU, as opposed to the ranges
with "?l", which I'd expect to provide greater efficiency (since the
innermost loop would then test more candidate passwords per iteration).

Alexander

Powered by blists - more mailing lists

Your e-mail address:

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