Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun, 14 Mar 2010 19:57:20 +0300
From: Solar Designer <>
Subject: Re: problem with initializing mpirun extended

On Sat, Mar 13, 2010 at 09:47:27AM +0100, Magnum, P.I. wrote:
> The MPI patch seems to make the buffering of logging more noticable.

Why is that?  I think it should be taking about as much time (real time,
not CPU time) until the first log flush as it does without the patch.

> If you start a job in background, the check the logs, it may 
> very well stop after the "748892 preprocessed word mangling rules" line 
> you mention for a very long time. But if you issue a kill -HUP to that 
> process, it will flush its log buffer and you'll see it is in fact working.

I think the same would be happening with a single process.

There was a relevant change between 1.7.4 and - LOG_BUFFER_SIZE
was increased from 4 KB to 32 KB.  Maybe this confused you into thinking
that MPI had something to do with the change in behavior you've observed.

> I have considered flushing more often but I'm not sure how/when to do it 
> without risking introducing performance hits.

Rather than reduce LOG_BUFFER_SIZE, which may hurt performance with a
lot of passwords getting cracked quickly or with a lot of rules and a
small wordlist, you may lower the "Save" setting in john.conf - say,
going from 600 down to 60 seconds won't affect performance significantly.

> The current behaviour is as it comes from mpi10.

I don't see it in the code.


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.