On Jun 24, 2008, at 9:13 PM, Joshua Bernstein wrote:
Exactly, but for now, if I make sure the NODES envar is setup correctly and make sure the OpenMPI is NFS mounted, and I actually have to copy out the mca libraries (libcache doesn't seem to work), I actually end up with something running!
FWIW, you might be getting tripped up on the fact that OMPI uses dlopen to find all of its DSO's (i.e., the stuff in $libdir/openmpi). If scyld proxies dlopen back to the head node, it could work, but if not, you might want to build OMPI with the "--disable-dlopen" configure option. This puts all the plugins inside libmpi.so (and friends), and disables the use of dlopen. This might make operation via your magic .so transporting to back-end nodes work properly.
-- Jeff Squyres Cisco Systems
