[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <0f932a0db2fa26b7a09db25d3d102704@smtp.hushmail.com>
Date: Thu, 26 Sep 2013 00:14:16 +0200
From: magnum <john.magnum@...hmail.com>
To: "john-dev@...ts.openwall.com" <john-dev@...ts.openwall.com>
Subject: External node distribution (call for help)
Solar, all,
The enclosed patch avoids the modulo operation in External mode when
distributing work using node/fork/MPI. The speedup depends on several
parameters but I've seen 10-15% for 8 nodes. This is the same scheme
that I now use in mask mode.
The downside is that the patch (once you enable it by uncommenting the
#define) changes the session file format for External mode so it will be
incompatible with core.
However, there is a way to avoid changing the file format: We could read
a "seq" number from file and convert it to "my_words" and "their_words".
And vice versa when saving. That should be fairly easy (although I
haven't really tried it yet) BUT what if the session was started on core
John, and "seq" wrapped around it's 32 bits once or several times? And
only then we resume it using the new code... and let's say our number of
nodes is a prime... Would this mean our conversion might not be valid?
Or can I totaly ignore such wrapping? I haven't figured that out yet %-)
magnum
View attachment "0001-Faster-node-fork-MPI-distribution-for-External-mode..patch" of type "text/plain" (3464 bytes)
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.
