Date: Wed, 05 Dec 2012 01:40:40 +0100 From: magnum <john.magnum@...hmail.com> To: john-dev@...ts.openwall.com Subject: Re: self-test duration vs. GPU benchmarks On 07/17/2012 07:44 PM, Solar Designer wrote: > When testing GPU formats, especially for slow hashes, the self-test is > often taking multiple seconds, and it affects the c/s rate reported > while cracking. > > The attached patch corrects this to some extent by only running the > self-test exactly once for each format (previously, it'd be run twice > for one vs. many salts benchmarks, and also before the start of each > cracking mode when running in "batch mode") and by not including its > duration in total session time. > > These changes will appear in core and thus in bleeding. As core John development has come to a full halt I have now merged this patch to unstable-jumbo instead of just to bleeding. We'll want it in Jumbo-8 anyway. > Unfortunately, there may be some bugs that would be caught by the > repeated self-tests, but would not be caught now. There were some in > the past. So we'll need to be trying -DDEBUG builds on some occasions. > I also intend to improve the self-tests to make them faster yet less > likely to miss issues like that. But it's a separate task, for later. I would like to add a new format flag in Jumbo: FMT_GPU. With that in place, we can omit GPU formats from the --test [all] case (with a comment printed) right now, and we could also handle self-tests somewhat differently from CPU formats - like testing a good number of hashes in one call to crypt_all() instead of trying one at a time in a large number of long-duration calls. 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.