Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20110923013448.GI21439@openwall.com>
Date: Fri, 23 Sep 2011 05:34:48 +0400
From: Solar Designer <solar@...nwall.com>
To: john-dev@...ts.openwall.com
Subject: Re: Fork=n

magnum, Rich -

On Wed, Sep 21, 2011 at 04:42:05PM -0400, Rich Rumble wrote:
> On Wed, Sep 21, 2011 at 4:18 PM, magnum <rawsmooth@...dband.net> wrote:
> > I just recently got hold of that code and I don't even have any
> > documentation at all. Is there any? How did you use -fork?
> I did:   john.exe hashes.txt --fork=8
> I never tried node, though others did.

Both --fork and --node mostly worked fine for their purpose in the
contest.

We actually used --node to an extent where we felt its scalability
limitations (same as those of MPI support in -jumbo), at over 100 nodes
when also using a narrow lengths range and a filter() (focusing on
specific previously identified patterns).  In some cases, there were
simply not enough incremental mode order[] entries meeting the
MinLen/MaxLen and filter() criteria at once - that is, we had more nodes
than suitable entries.  But this was more of an occasional exception
even under contest conditions.

> An of course you could apply
> all the other augments too (format,rules,chr...)

Actually, as implemented for the contest, --fork and --node only worked
right either with --rules or with incremental mode.  They were not
supported for plain wordlist mode (without rules), although this was not
explicitly detected as an error (another reason not to release this hack).

> --node=MIN[-MAX]/COUNT    this node's number range out of total count
> --fork=COUNT              fork this many processes
> --node=1/8 --fork=4   <- on one quad-core machine, uses node nums 1-4
> --node=5/8 --fork=4   <- on another quad-core machine, uses nums 5-8
> 
> That is about all I know of how to USE fork, no idea how it was coded,
> but it was coded in a short time.

Right.

> The issue was that not all the threads
> (if that's the right word) were updating the Pot file unless a sighup was
> sent, or the cracking ended, even then I don't think it wrote all the ones
> it cracked, at least not on win32.

That's one of many issues, not the only one, and this one was only
reported by you - I've never seen it.  So I think it was in fact
platform specific.

> Restore was a problem too, and that I believe was to be expected.

Exactly.

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.