I should say that on FC5, where I see the error, mpi runs just fine. Brian
On 10/9/07, Dirk Eddelbuettel <e...@debian.org> wrote: > > On 8 October 2007 at 22:06, Brian Granger wrote: > | Also seeing this problem on Fedora Core 5. Any resolution yet? > > No, none. With the exact same configuration (encoded in the Debian package > build 'recipe' for the package), I get > > -- on Debian: 'unable to open osc pt2pt' verbosity but a working Open MPI > setup > > -- on Ubuntu: no verbosity, but Open MPI hangs > > Very puzzling. I have rebuilt many other Debian packages on my Ubuntu > systems and yet to see any divergence or regression in behaviour. > > Needless to say, I'd like to get this work, but I do not know what to try > next. > > Dirk > > > > | Brian > | > | On 10/6/07, Dirk Eddelbuettel <e...@debian.org> wrote: > | > > | > On 6 October 2007 at 09:36, Dirk Eddelbuettel wrote: > | > | > | > | On 5 October 2007 at 21:31, Brian Barrett wrote: > | > | | On Oct 5, 2007, at 8:48 PM, Dirk Eddelbuettel wrote: > | > | | > | > | | > With the (Debian package of the) current 1.2.4 release, I am seeing > | > | | > a lot of > | > | | > > | > | | > mca: base: component_find: unable to open osc pt2pt: file not > | > | | > found (ignored) > | > | | > > | > | | > that I'd like to suppress. > | > | | > > | > | | > For these Debian packages, we added a (commented-out by default) > | > | | > entry to > | > | | > suppress the Infiniband noise when no Infiniband hardware is to be > | > | | > found. I > | > | | > would like to suppress this 'osc pt2pt' message too. > | > | | > > | > | | > But all attempts at guestimating parameters for > | > | | > /etc/openmpi/openmpi-mca-params.conf > | > | | > based on what > | > | | > ompi_info all all > | > | | > shows failed. Could someone help me along? > | > | | > | > | | This is a bit different, and points to something bad going on. The > | > | | error message is that for some reason, a library that the pt2pt > | > | | component depends on was not found. The pt2pt osc component is > | > | > | > | Uh-oh. Doesn't sound good. > | > | > | > | | entirely built on the MPI layer -- it shouldn't have any external > | > | | dependencies. Can you run ldd on the library and see if there's > | > | | anything obvious? > | > | > | > | edd@ron:~> ldd /usr/lib/openmpi/mca_osc_pt2pt.so > | > | libnsl.so.1 => /lib/i686/cmov/libnsl.so.1 (0xb7f9a000) > | > | libutil.so.1 => /lib/i686/cmov/libutil.so.1 (0xb7f96000) > | > | libm.so.6 => /lib/i686/cmov/libm.so.6 (0xb7f70000) > | > | libpthread.so.0 => /lib/i686/cmov/libpthread.so.0 (0xb7f59000) > | > | libc.so.6 => /lib/i686/cmov/libc.so.6 (0xb7e11000) > | > | /lib/ld-linux.so.2 (0x80000000) > | > | > | > | Not that I can tell. What else could I test? The build-logs don't > reveal > | > | anything fishy -- all pt2pt occurrences look fine. The build itself > | > | proceeded fine (and this was the Debian package build I then uplod) > | > > | > Two more observations: > | > -- the message does not appear on my Ubuntu system > | > -- but it appears on Hao's Debian machine which does not use the Debian > package > | > > | > Could this be some dynamic loading issue? How can we go about solving > this? > | > > | > Dirk > | > > | > -- > | > Three out of two people have difficulties with fractions. > | > _______________________________________________ > | > users mailing list > | > us...@open-mpi.org > | > http://www.open-mpi.org/mailman/listinfo.cgi/users > | > > | _______________________________________________ > | users mailing list > | us...@open-mpi.org > | http://www.open-mpi.org/mailman/listinfo.cgi/users > > -- > Three out of two people have difficulties with fractions. > _______________________________________________ > users mailing list > us...@open-mpi.org > http://www.open-mpi.org/mailman/listinfo.cgi/users >
