Hi Gerhard and all, Thanks for those kind words of encouragement. I'll keep you guys posted on the progress. Who knows? It might actually work out and if it does, I'll make sure I document it properly on some wiki. BTW, does openbsd have an official wiki? There's wikia and openbsd-wiki, but they're both unofficial.
Also, I found a program for Linux/BSD called nvd: "The network-based virtual device driver for Linux/BSD systems. Now it can provide remote block device access just like 'nbd' but much more stable and efficiency. Charactor device and BSD support will be enabled in near future." I'm not sure if you can create block device out of pure RAM. Can you? If you can, then this path may be the way to go. Thanks, Vivek On Thu, Mar 26, 2009 at 11:35 PM, Gerhard Hoffmann <hoffm...@mathematik.tu-darmstadt.de> wrote: > Hi Vivek, > > On Thu, Mar 26, 2009 at 12:09:28PM -0700, Vivek Ayer wrote: >> I tried building open-mpi on those blade machines (1000 & 2000) which >> run openbsd/sparc64 with the -mv8plus flag for CFLAGS, CXXFLAGS, >> FFLAGS, and FCFLAGS. The configure script went well, but both systems >> crashed during make. MPICH2 seems to work better on these machines >> than Open-MPI. For MPICH2, I tried "./configure --with-pm=mpd:gforker >> --disable-romio --disable-mpe." gforker will only allow you run to run >> jobs on one node. It's good for SMP systems. Unfortunately, both blade >> systems have only one core and that's not my goal anyway. I wanted to >> add the " enable-fast" for the best performance, but just wanted to be >> safe, ./configure --with-pm=mpd:gforker --disable-romio --disable-mpe >> builds just fine. I believe having mpd means I may able to can > > yes, > > no --enable-fast: good idea, although I never experienced problems > --disable-romio: parallel filesystem: you don't need this at the moment > --disable-mpe: for debugging purposes. > mpd: the underlying demon, which creates the runtime MPI envionment > on the cluster > > Thanks for the hint with OpenMPI. Maybe I will talk to a guy who > is a member of the MPI group and is actively working on it. > > And you are really a tough guy. I would not dare to do this. > Just to tell Matlab/Mathematica to use this remote memory > will not be an easy task. And then all the rest ..., my god. > But good luck. > > Regards > Gerhard > >> distribute jobs between different architectures. My MATLAB/Mathematica >> Licenses live on ArchLinux/i686, so hopefully it'll work out! >> >> Cheers, >> Vivek >> >> On Thu, Mar 26, 2009 at 3:19 AM, Paul Irofti <bulib...@sdf.lonestar.org> >> wrote: >> > On Wed, Mar 25, 2009 at 06:43:54PM -0700, Vivek Ayer wrote: >> >> Yes please. I'd love to take a look at the code. >> > >> > My laptop and everything else is packed because I'm leaving in a few >> > hours so you'll have to do with an archive link. >> > >> > http://marc.info/?l=openbsd-ports&m=119480513327121&w=2
