Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Fri, 07 Mar 2014 08:31:14 +0100
From: magnum <john.magnum@...hmail.com>
To: john-users@...ts.openwall.com
Subject: Re: Help - mpi ocl restore session

On 2014-03-07 07:26, Anthony Tanoury wrote:
> I use John the Ripper, version 1.8.0.2-bleeding-jumbo_mpi
> [linux-x86-64-opencl]

Is this a very recent snapshot or an older one? Some timer oddities has 
changed (for the better, hopefully) very recently. Like yesterday...

> I can run an mpirun opencl session just fine and all sessions
> complete just fine. My only  trouble is with session restore and only
> if it involves remote hosts. I can resume a session if there is not a
> remote host. However, if I terminate a session with a one or more
> remote hosts using "killall mpirun", "kill HUP" or Ctrl-c and try to
> restore the session, only one core or one GPU will resume.

When you say "one" I assume you mean only the root node resumes?

Does this happen even if the session had been running for 30 minutes or 
more? Did you set "Save = 60" in john.conf per the instructions? Before 
killing an MPI session, you should "kill -USR1" the mpirun that controls 
it. This should trigger a session save. Then wait at least 30 seconds 
before aborting them.

What happens with the other nodes? Do they silently just not resume or 
are there any errors or other clues?

Can you see any clues in the log files?

If it's a very slow (unresponsive) format, try running with lower GWS 
(using eg. "mpirun -x GWS=2048" or whatever number is a lot lower than 
what is othereise used) when testing.

magnum

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.