This information can be found elsewhere, but I figured that it's worth reposting. The bottom line here are errors that look like:
librdmacm: Warning: couldn't read ABI version.
librdmacm: Warning: assuming: 4
librdmacm: Fatal: unable to get RDMA device list
--------------------------------------------------------------------------
[[13473,1],0]: A high-performance Open MPI point-to-point messaging module
was unable to find any relevant network interfaces:
Module: OpenFabrics (openib)
Host: a_server
Another transport will be used instead, although this may result in
lower performance.
--------------------------------------------------------------------------
librdmacm: Warning: couldn't read ABI version.
librdmacm: Warning: assuming: 4
librdmacm: Fatal: unable to get RDMA device list
Are a direct result of your openmpi implementation looking for infiniband hardware. You can turn off this by plugging the equally criptic option -mca btl ^openib
into your command line call of whatever programme you are running.
The option -mca btl ^openib
specifies that the option ^openib
is passed to the "MCA" module btl
. This is a specification of what type of communication to use between nodes. The man page for mpiexec
is here, here is another useful post about this and here is some more additional information about mca parameters.
Enjoy!