Date: Wed, 16 Sep 2015 02:05:30 +0300 From: Solar Designer <solar@...nwall.com> To: john-dev@...ts.openwall.com Subject: Re: Judy array On Wed, Sep 16, 2015 at 12:43:44AM +0200, magnum wrote: > On 2015-09-15 09:57, magnum wrote: > >On 2015-09-15 05:40, Solar Designer wrote: > >>This saves 4 seconds, but only when -mem=999999999 is also used: > >> > >>-WordlistMemoryMap = Y > >>+WordlistMemoryMap = N > > > >We could probably default to not mmap when that other buffer is used. > > I had a look, that's a no-go. When mmap was introduced the code was > simplified a bit. I'm not going to make it worse again. We still need to > re-write it from scratch (with "scratch" meaning the original non-Jumbo > file). > > Also, I don't observe any gain from disabling mmap and only minimal gain > from using --mem=0 when mmap is enabled (I stopped using -mem after mmap > was implemented). I'm surprised by your results, especially given you > were short on memory... or is it *because* of just that..? No, I wasn't that short on memory. The machine has 16 GB, and after my copy-on-write reducing fixes these test runs allocate at most 9 GB total across the 8 processes. So there's still enough memory to have the wordlist, etc. cached in RAM across the runs, etc. Let me re-test... First, I took the latest cracker.c. This got my prefetching disabled. And, surprise, it's now 2 seconds slower (I still have SHR=0): real 0m58.804s user 4m26.433s sys 0m18.934s Maybe the prefetching is sometimes helping, after all, even though for our current raw-md5 it's very limited (max_keys_per_crypt is only 12, so those loops don't actually go to the hard-coded maximum of 64 outstanding prefetches). Now, enabling mmap (still no prefetch, etc.): real 1m0.847s user 4m51.136s sys 0m15.579s In both cases, I also had -mem=0 on the command line. Alexander
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.