john-users - Re: beta MPI john patch against 1.7.2.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Fri, 2 Jun 2006 13:11:14 -0400
From: Erik Winkler <ewinkler@...ls.com>
To: john-users@...ts.openwall.com
Subject: Re: beta MPI john patch against 1.7.2.

The patch seems to be working.  Here are a few sample tests:

1) mpirun -np 1 /JTR/john -test -format:DES (master - 1.3 Ghz G4)
Benchmarking: Traditional DES [128/128 BS AltiVec]... DONE
Many salts:     620108 c/s real, 635357 c/s virtual
Only one salt:  566553 c/s real, 571122 c/s virtual

2) mpirun -nolocal -np 1 /JTR/john -test -format:DES (single 1.8 Ghz  
G5 CPU)
Benchmarking: Traditional DES [128/128 BS AltiVec]... DONE
Many salts:     1011K c/s real, 1013K c/s virtual
Only one salt:  877721 c/s real, 883019 c/s virtual

3) mpirun -np 4 /JTR/john -test -format:DES (3-1.8 Ghz G5, 1-1.3 Ghz  
G4 CPUs)
Benchmarking: Traditional DES [128/128 BS AltiVec]... DONE
Many salts:     3636K c/s real, 3680K c/s virtual
Only one salt:  3185K c/s real, 3217K c/s virtual

mpirun -np 4 /JTR/john -test -format:LM (3-1.8 Ghz G5, 1-1.3 Ghz G4  
CPUs)
Benchmarking: NT LM DES [128/128 BS AltiVec]... DONE
Raw:    21537K c/s real, 21876K c/s virtual

mpirun -np 4 /JTR/john -test -format:MD5 (3-1.8 Ghz G5, 1-1.3 Ghz G4  
CPUs)
Benchmarking: FreeBSD MD5 [32/32 X2]... DONE
Raw:    13610 c/s real, 13824 c/s virtual

Really makes a difference on the MD5 crunching.  I have tested this  
version of john running against hash files with no success.  While  
the hash crack faster, the master instance of john doesn't shut down  
until I do a kill -9 on the process.  The password is then revealed.   
This might be a bug, but I am not sure.  I am running MPICH.

Erik

On May 31, 2006, at 11:34 PM, Otheus (aka Timothy J. Shelling) wrote:

> This is a (beta) patch for John 1.7.2 for use with MPI.  All of the  
> MPI-specific code is encapsulated by #ifdef MPI/#endif, so in  
> theory this can be integrated into the main code branch. However, a  
> few generic changes to the code are also made.
>
> Please let me know of any rejects while performing the patch. (Use - 
> l to ignore whitespace to avoid problems with bench.c.)
>
> Changes:
>   o  internalized computing the checksum for ext_word. 15% increase  
> in speed (at least)
>   o  makes sure options required for MPI are present
>   o  computes MPI performance in --test mode.
>
> Notes:
>   o  benchmark computes results by summing cracks across all nodes  
> and dividing that by the AVERAGE of the crack-times. (in the code,  
> cracks are multiplied by the MPI-size, which is the same thing).
>
> Known bugs:
>   o   john.pot gets littered with extraneous output, usually low  
> integers like "4" or "1" and a newline. Very strange. Still trying  
> to track it down.
>   o   successfull cracks do not get posted to the other tasks
>   o   use of external filter with internal checksum causes a slight  
> (1-2%) slowdown
>
>
> Incorporate the john.conf file into yours (or the default)... then  
> run with:
>
> mpirun <mpirun-args> john <mode> --external=MPIbyInternalKeySum -- 
> session=john%d.rec
>
> For a restore:
>
> mpirun <mpirun-args> john --restore=john%d.rec
>
> You can also benchmark it:
>
> mpirun <mpirun-args> john --test
>
>
> -- 
> Otheus
> otheus@...il.com
> +43.650.790.2571
> -- 
> To unsubscribe, e-mail john-users-unsubscribe@...ts.openwall.com  
> and reply
> to the automated confirmation request that will be sent to you.