On May 6, 2013, at 2:10 AM, Angel de Vicente <angelv_at_[hidden]> wrote:
> Ralph Castain <rhc_at_[hidden]> writes:
>> On May 4, 2013, at 4:54 PM, Angel de Vicente <angelv_at_[hidden]> wrote:
>>> Is there any way to dump details of what OpenMPI is trying to do in each
>>> node, so I can see if it is looking for different libraries in each
>>> node, or something similar?
> thanks for the suggestions, but I'm still stuck:
>> What I do is simply "ssh ompi_info -V" to each remote node and compare
>> results - you should get the same answer everywhere.
> exactly the same information in the three connected machines
So you should then be getting the same libraries
>> Another option in these situations is to configure
>> --enable-orterun-prefix-by-default. If you install in the same
>> location on each node (e.g., on an NSF mount), then this will ensure
>> you get that same library.
> Re-configured and re-compiled OpenMPI, but I get the same behaviour.
> I'm starting to think that perhaps is a firewall issue? I don't have
> root access in these machines but I'll try to investigate.
Given that result, then yes - check iptables. I suspect they are running and TCP socket comm is being blocked.
> Ángel de Vicente
> users mailing list