Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Sat, 07 Apr 2012 15:22:10 -0500
From: "Jimmy Blakeney" <jblakeney@...phisemail.com>
To: <john-users@...ts.openwall.com>
Subject: Re: Restore to mpi

Yes. Traditional DES.
 
using linux-x86-64 to build on the new system.
 
Old system(single proc):
Benchmarking: Traditional DES [128/128 BS SSE2-16]... DONE
Many salts:     837094 c/s real, 838771 c/s virtual
Only one salt:  757606 c/s real, 757606 c/s virtual
New System(vmware 6 virtual procs):
Benchmarking: Traditional DES [128/128 BS SSE2-16]... DONE
Many salts:     14367K c/s real, 2403K c/s virtual
Only one salt:  11801K c/s real, 1973K c/s virtual
 
Nearly 20x increase!!
 
Thanks for all the help.  Being new to OpenMP I thought I needed to install some additional software to make it work.  Everything was already there.
 
Jimmy
 
 
>>> 


From: Solar Designer <solar@...nwall.com>
To:<john-users@...ts.openwall.com>
Date: 4/7/2012 2:04 PM
Subject: Re: [john-users] Restore to mpi
On Sat, Apr 07, 2012 at 01:37:45PM -0500, Jimmy Blakeney wrote:
> Thanks for the info.  It is for a basic Unix crypt password set.  Looks like 1.7.9 supports it??

Yes.  You mean "Traditional DES", right?

> I recompiled and it appears to run.  Looks much easier than MPI.

Sure.  The efficiency of OpenMP for that hash type is around 90%, though
(and that's assuming a 64-bit build).

> It looks like the status is cumulative for c/s??  I will keep an eye on it to see if it update.

Yes, it is cumulative, so it should start increasing very slowly.

I suggest that you run "john --test --format=des" for both builds (old
and new) to get an idea of how the performance changed.  Also, you may
need to tune GOMP_SPINCOUNT (this is an environment variable) - e.g., a
value of 3000000 (up from gcc 4.6+'s default of 300000) works better for
some machines.

You may want to make sure your numbers agree with the posted benchmarks:

http://openwall.info/wiki/john/benchmarks

(if not, there may be something for you to tune).

> Is there a verbose switch to show what is being tried.  I thought an older version would show what is currently being tested.

You just press a key to see the status.  It works the same in either
build, although with OpenMP the buffers are larger (so with OpenMP it is
more likely to see the same "range" of passwords being tested on two
consecutive keypresses).

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.