From: Rich C. <rc...@ll...> - 2008-02-14 01:35:21
|
On Feb 13, 2008, at 7:05 AM, Sean Ahern wrote: > I think all of us who will be doing development on blockbuster > should check that we are members of the blockbuster developer > mailing list at SourceForge. That will likely be the best place to > continue technical discussions about project direction. > > Rich Cook wrote: >> You convinced me. I'll move in that direction (MPI). This >> synchronization thing has become pretty high priority here. It's >> not that hard, actually, so let's see what happens this week. > > Okay. We're not going to get a chance to play with it for a few > weeks, like I said, so we're not in a hurry. Stay tuned > >> Agreed that staying on the same branch is desirable. At the same >> time, let's recognize that this will require more communication >> and thus a slightly higher level of project management on both ends. > > True enough. > >> Furthermore note that resources are scarcer around here than used to >> be, so there may be lags in progress occasionally on my end. > > That's okay. I understand. :-) Becky has mentioned we might be able to add someone to help me with graphics support to enable more blockbuster progress. > > -Sean > > __ > Sean Ahern > Oak Ridge National Laboratory > AIM: ornlsean > 865-241-3748 >> On Feb 12, 2008, at 1:45 PM, Sean Ahern wrote: >>> Rich Cook wrote: >>>> I am going to tighten up frame sync on the next iteration. I >>>> tentatively plan to create a barrier right before the slaves >>>> actually call glSwapBuffers, and to use TCP sockets instead of >>>> MPI, to avoid rewriting the code that launches the slaves, and >>>> to avoid the problems that arise from the variety of MPI >>>> launching schemes out there. I would actually like to rewrite >>>> the slave launching code, but it's a longer path. >>> >>> I definitely want to go down the MPI path. Our current wall has >>> 14 slave nodes. Our new cluster will have 27. That's just too >>> slow for a TCP-based method. Any communication method based on >>> MPI will be orders of magnitude faster than a TCP-based one. Not >>> just for launching, but for synchronization, too. >>> >>>> Let me know what you're planning. Perhaps we should branch or >>>> something? I'm not sure how to blend our efforts effectively >>>> but it will be nice to get your contributions, thanks! >>> >>> I'm not sure how to blend efforts either. Let me draft up an >>> outline of the work we hope to do, and we can see how it might >>> merge. At the worst, making an SVN branch would work. We can >>> set up an ORNL repo here, I suppose, and then sync up with you at >>> a later point. But I'd rather keep things synced up between LLNL >>> and ORNL. We can benefit from each other's modifications. >>> >>> -Sean >>> >>> __ >>> Sean Ahern >>> Oak Ridge National Laboratory >>> AIM: ornlsean >>> 865-241-3748 -- ✐Richard Cook ✇ Lawrence Livermore National Security Bldg-453 Rm-4037, Mail Stop L-557 7000 East Avenue, Livermore, CA, 94550, USA ☎ (office) (925) 423-9605 ☎ (fax) (925) 423-6961 --- Information Management & Graphics Grp., Services & Development Div., Integrated Computing & Communications Dept. (opinions expressed herein are mine and not those of LLNS) |