On Fri, Jun 01, 2012, Ross Walker wrote:
> AAARRRGHHHH!!! This is why "I hate Ubuntu"(tm). Ubuntu installs the mpi
> libraries, binaries and header files in non-standard paths,
But why would you use ubuntu's (or any distribution's) MPI? We have no way of
knowing how it was configured.
I really think that our standard response to MPI problems is not to tell
people to change their operating system. Rather, the first thing should be
to have them use the configure_openmpi or configure_mpich2 scripts that we
provide. That way, we know that MPI has the right compilers, the right
options requested, and that the include files are in the proper locations.
After a user has a working installation, if she wants to play around, trying
other MPI's, that is certainly a possiblity. In some situations, that might
lead to performance improvements (although I kind of doubt it for
pmemd.cuda.MPI, where the scalaing problems don't really have to do with MPI
limitations, as I understand it.)
And, if users have problems with our configuation scripts, we at least have a
chance in hell of fixing it, and we can improve the scripts for everyone.
[Aside: I have basically the same opinion about netcdf: I don't see the
advantage in trying to use a vendor's pre-installed netcdf...seems' like it is
making us more reliant than we need to be on some one else's code.]
....dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Jun 01 2012 - 10:00:04 PDT
