Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20111129205007.GB29519@openwall.com>
Date: Wed, 30 Nov 2011 00:50:07 +0400
From: Solar Designer <solar@...nwall.com>
To: john-users@...ts.openwall.com
Subject: Re: AMD X6 multicore issue

My comments are inline:

On Tue, Nov 29, 2011 at 06:59:09PM +0200, Caesarmv wrote:
> * c@cd:~/pass/john-1.7.8-jumbo-8/run$ mpiexec -np 6 ./john ../../pass.txt*
> 
> Loaded 2 password hashes with 2 different salts (generic crypt(3) [?/64])
> Loaded 2 password hashes with 2 different salts (generic crypt(3) [?/64])
> Remaining 1 password hash
> Loaded 2 password hashes with 2 different salts (generic crypt(3) [?/64])
> Remaining 1 password hash
> Remaining 1 password hash
> Node 0@cd: Crash recovery file is locked: ./john.rec
> Node 0@cd: Crash recovery file is locked: ./john.rec
> Loaded 2 password hashes with 2 different salts (generic crypt(3) [?/64])
> Node 0@cd: Crash recovery file is locked: ./john.rec
> Remaining 1 password hash
> Loaded 2 password hashes with 2 different salts (generic crypt(3) [?/64])
> Remaining 1 password hash
> Loaded 2 password hashes with 2 different salts (generic crypt(3) [?/64])
> Remaining 1 password hash

Looks like you're running mpiexec on a non-MPI build.  Maybe you did not
enable MPI in the Makefile, or maybe you forgot to do a "make clean"
between different builds.

> *c@cd:~/pass/john-1.7.8-jumbo-8/run$ mpiexec -np 6 ./john --test*
> 
>  Benchmarking: Raw SHA-512 [64/64]... DONE
> Raw:    1235K c/s real, 1235K c/s virtual
> 
> Benchmarking: Raw SHA-512 [64/64]... DONE
> Raw:    1216K c/s real, 1228K c/s virtual
> 
> Benchmarking: Raw SHA-512 [64/64]... DONE
> Raw:    1213K c/s real, 1237K c/s virtual
> 
> Benchmarking: Raw SHA-512 [64/64]... DONE
> Raw:    1194K c/s real, 1231K c/s virtual
> 
> Benchmarking: Raw SHA-512 [64/64]... DONE
> Raw:    1186K c/s real, 1235K c/s virtual
> 
> Benchmarking: Raw SHA-512 [64/64]... DONE
> Raw:    1786K c/s real, 1804K c/s virtual

This is similarly wrong.  With an MPI-enabled build, you'd get just one
benchmark per hash type.  (Ditto for OpenMP.)

> > Compilation type is  *make linux-x86-64.*
> >
> > Did you enable MPI or/and OpenMP in the Makefile?
> 
> Yes.

What exactly did you do?

> I'm interested in *SHA512* (ubuntu default pass file).

You're confused here.  crypt(3) based on SHA-512 is one thing (slow),
raw SHA-512 is another (fast).  You need the former.

Anyway, John includes not only MPI, but also OpenMP support in its
"generic crypt(3)" module (which is what you should be using for this
hash type currently).  So my advice is that you don't bother with MPI
(which is a jumbo-specific feature, and which you somehow misconfigured)
and instead enable OpenMP in the Makefile (should be easier to do and
use).  You may use a clean 1.7.9 tree (no need for -jumbo for your hash
type).  Please don't forget to "make clean" before making a new build.

I hope this helps.

Alexander

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.