I think Dave has probably hit the problem - that node may well not have a “numa” object. You also might check that node “leo” has libnuma on it
> On Oct 30, 2015, at 6:48 AM, Dave Love <d.l...@liverpool.ac.uk> wrote: > > Fabian Wein <fabian.w...@fau.de> writes: > >> Is this a valid test? >> >> >> /opt/openmpi-1.10.0-gcc/bin/mpiexec -n 4 hostname >> leo >> leo >> leo >> leo > > So, unless you turned off the default binding -- to socket? check the > mpirun man page -- it worked, but the "numa" level failed. I don't know > if that level has to exist, and there have been bugs in that area > before. Running lstopo might be useful, and checking that you're > picking up the right hwloc dynamic library. > > What happens if you try to bind to sockets, assuming you don't want to > bind to cores? [I don't understand why the default isn't to cores when > you have only one process per core.] > >> /opt/openmpi-1.10.0-gcc/bin/mpiexec -bind-to numa -n 4 hostname >> -------------------------------------------------------------------------- >> A request was made to bind a process, but at least one node does NOT >> support binding processes to cpus. >> >> Node: leo >> This usually is due to not having libnumactl and libnumactl-devel >> installed on the node. > > By the way, you can check the binding done, independently to what > openmpi says, with > mpirun ... hwloc-ps > _______________________________________________ > users mailing list > us...@open-mpi.org > Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/users > Link to this post: > http://www.open-mpi.org/community/lists/users/2015/10/27957.php
