On 03/27/2014 07:19 AM, Dave Love wrote:
Rob Latham <[email protected]> writes:
Ugh, thanks for the link to the ticket. Edgar Gabriel's approach is
not exactly what I would have done (or did do in MPICH). Hope I am
not too late to set things straight.
Actually it's not clear whether or not there will be another 1.6 release
-- 1.6.5 was supposed to be the last -- but I guess it's worth
sanitizing for 1.7+. Presumably people who need pvfs with 1.6 can just
patch it, and at least there's something in list archives now. Thanks
for looking at it.
Also, I didn't read/check vithanousek's reply closely enough. I'd taken
a patch from openmpi 1.7 that changed ADIOI_PVFS2_IReadContig to
ADIOI_FAKE_IreadContig rather than NULL.
setting to NULL seems like a horrible idea.
==rob
I wondered why it was different in OMPI 1.6 and 1.7, but I haven't
understood the interaction between OMPI and ROMIO.
The way I think it usually works is folks develop ROMIO in MPICH, then
every so often OpenMPI syncs up with us. Edgar Gabriel, bless his
heart, is maybe the only guy who works on OpenMPI-ROMIO directly. I did
not realize he had been doing that, and we blew away his changes last
time we re-synced ROMIO.
An additional complication is that there's "OpenMPI's MPI-IO", again
that Edgar Gabriel and his students wrote. That MPI-IO implementation
also has some PVFS2 support.
==rob
--
Rob Latham
Mathematics and Computer Science Division
Argonne National Lab, IL USA
_______________________________________________
Pvfs2-users mailing list
[email protected]
http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
