Date: Thu, 24 May 2012 23:07:12 +0200 From: magnum <john.magnum@...hmail.com> To: john-dev@...ts.openwall.com Subject: Timing issues When I tried the high-rounds crypt-SHA512 hash (as seen on john-users) on bull just now, I stumbled onto two issues, not really new to me but I think they need to be looked at. 1. I tested using --max-time=60 but it was not obeyed. This option is designed so that if the self-test takes 40 seconds the expected total run time would be about 1:40 instead of 60 seconds, but this does not explain it (it ran for several minutes before I killed it). I will look into this later but I have a feeling it's something with our OS_TIMER=0 - does that make sense? I have seen other weird things that started to happen after we did this change (can't remember exactly what right now, I think it was something like "reported time not reflecting wall time"). 2. When cracking (as opposed to benchmarking) the time before cracking actually starts is counted: guesses: 0/7680 time: 0:00:00:43 c/s: 175 trying: HC - �� guesses: 0/76800 time: 0:00:01:00 c/s: 1279 trying: W� - �> guesses: 0/176640 time: 0:00:01:23 c/s: 2119 trying: #Z� - #�� The real speed is constantly at something like 4300 c/s considering it did not even start until about 0:40. This makes quick tests like this one hard to do (need to run a lot longer, or re-calculate manually which is what I did, hopefully not totally wrong). Do we want to change this? It could be considered cheating in a way but I think the current behaviour is just confusing (say for example you try to follow the suggested homing for markov mode - you will end up with a too low level, and stop long before your targeted runtime) This would of course be mitigated a little by fixing the underlying issue that the self-test takes 40 seconds. 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.