Date: Sat, 26 Mar 2011 04:40:47 +0300 From: Solar Designer <solar@...nwall.com> To: john-dev@...ts.openwall.com Subject: Re: GSoC John the Ripper project Douglas, all - On Thu, Mar 24, 2011 at 12:10:56PM -0600, Douglas Hains wrote: > I'm very interested in the John the Ripper tasks, specifically > parallelization of the hashing on GPUs. I was hoping to get in contact with > someone from your project who is interested in mentoring these tasks or > would be able to chat with me more about the tasks. I guess that would be me at this time, although I welcome other volunteer mentors as my own mentoring capacity is too limited (it looks like I'll be the mentor for several other tasks as well, not only JtR). I suggest that we discuss the JtR tasks on this mailing list. > I have experience in programming for GPUs on a variety of models of NVIDIA > cards and have worked with multi-GPU programming on clusters of Fermi and > Tesla cards. Will you have access to such clusters during GSoC, for testing? > I also have experience using both OpenMP and MPI and am > interested in combining both GPU parallelization There was a recent discussion on making use of GPUs here: http://thread.gmane.org/gmane.comp.security.openwall.john.user/3531 The multi-GPU aspect was not touched yet, though. > and host side parallelization through MPI. It was previously determined that MPI was not such a good match for JtR, or at least the existing attempts sort of misused MPI: http://openwall.info/wiki/john/parallelization#Extended-efforts There were posting on john-users (including by me) on why these patches could have easily achieved the same without the dependency on MPI. > To me, the most interesting aspect of these > kinds of projects is finding the optimal granularity of parallelization and > performing a kind of dynamic load balancing through allocation of tasks to > the available compute nodes and/or GPUs. This would be the main focus of > my proposal. Yes, this is one of the most interesting aspects in my opinion too. Also, it is desirable to have a distributed setup working over the Internet at large, and over unreliable links and with unreliable (crashing/disappearing) nodes in general. This is one reason why I think we need to design a custom protocol, not use a generic parallel processing library. > I have questions about the expected scope of a JtR tasks proposal and the > platforms, architectures and compilers that would be acceptable to be used > for this project. There are several sub-projects, and different approaches can be attempted simultaneously. I'd appreciate it if you post your specific questions in here. We can also schedule an IRC meeting (just e-mail me off-list for that), although I'd prefer the discussion to be visible to others on john-dev. Thanks, 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.