Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Sun, 14 Mar 2010 23:05:33 +0100
From: "Magnum, P.I." <>
Subject: Re: weird behavior JTR 1.7.5 64 bits + fullMPI 3 + Jumbo
 patch 2 wrote:
> Hi
>  I compiled JTR 1.7.5 + john-1.7.5-jumbo-2.diff then 
> john-1.7.5-fullmpi-3-after-jumbo2.diff (make clean macosx-x86-64 )
>  I ran JTR with a big worlidst (64 gigas) (see the log)

This is not logs, it's screen output.

>  Then I try same JTR with "-rules" (see log)
>  As you can the cracking session start with cores from 1 tp 7 at 100% !
>  Only 1 core working 
>  What is wrong ???

You have the logs, check what happened. I've included verbose logging 
messages so you can see what happens on each core.

If the large buffers are confusing, try setting "Save = 1" in john.conf 
and re-run the job. But don't forget setting it back afterwards.

My guess is there was only one rule. The MPI code currently does not 
detect that there are fewer rules than nodes, so it will hand out this 
rule to node 0 and then run out of rules. The others will not get any 
job. In this situation it would be better to distribute words instead of 
rules. But there are so many different cases.

MPI is really best for Markov and Incremental. Maybe Single, if you have 
lots of users and valid user info. The Wordlist modes are just an 
experiment and they do not always do the right thing. They always do the 
same job, but not always faster.

Given the "right" jobs (either lots of rules, or no rules but very large 
wordlist) it does pretty well, but an intelligent, manual (sometimes 
tedious) split, after choosing a method depending on the situation, 
would almost always perform better.


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.