Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sat, 13 Mar 2010 17:23:27 +0100
From: "Magnum, P.I." <>
Subject: Re: Is JTR MPIrun can be optimized for more cores ?

I wrote:
> I just found something odd though. Can anyone explain this?
> The speed to /dev/null is bad, the one to file is awful.

Got it. Tried gprof but it didn't show much difference to non-mpi. So I 
tested recompiling with OpenMPI instead of MPICH2 and that put the 
figures back to more like normal. There still is a significant 
performance hit though, despite we're not actually doing any MPI stuff 
in this case. I'm glad I don't see this kind of performance hit when 

$ ./john -stdout -w=sd.lst -ru=s >file1
words: 62570773  time: 0:00:00:12 100.00%  w/s: 5171K  current: snip

$ ./john -stdout -w=sd.lst -ru=s >/dev/null
words: 62570773  time: 0:00:00:10 100.00%  w/s: 5804K  current: snip

john-1.7.5-jumbo2-fullmpi-3 using MPICH2
$ ./john -stdout -w=sd.lst -ru=s >file2
words: 62570773  time: 0:00:03:45 100.00%  w/s: 277021  current: snip
                              ^^^^               ^^^^^^
$ ./john -stdout -w=sd.lst -ru=s >/dev/null
words: 62570773  time: 0:00:00:29 100.00%  w/s: 2124K  current: snip

john-1.7.5-jumbo2-fullmpi-3 using OpenMPI
$ ./john -stdout -w=sd.lst -ru=s >file2
words: 62570773  time: 0:00:00:13 100.00%  w/s: 4693K  current: snip

$ ./john -stdout -w=sd.lst -ru=s >/dev/null
words: 62570773  time: 0:00:00:11 100.00%  w/s: 5266K  current: snip

I didn't change one single line of code, just recompiled it with openmpi 


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.