Actually, I suspect the problem is that you don't have PMI running on the
machine. The processes have no reason to be opening sockets for the OOB on
a Cray XE6, and if you look at that platform file, it defines the location
of the PMI libraries that are required.
Since it built, I expect the libraries were found. However, you should
check to ensure your sys admin actually has PMI running.
On Wed, Oct 10, 2012 at 7:46 AM, Reuti <reuti_at_[hidden]> wrote:
> Am 10.10.2012 um 14:50 schrieb Christoph Niethammer:
> > I just tried to use Open MPI 1.7a1r27416 on a Cray XE6 system.
> Unfortunately I get the following error when I run a simple HelloWorldMPI
> > $ pirun HelloWorldMPI
> > App launch reported: 2 (out of 2) daemons - 0 (out of 32) procs
> > ...
> > [unset]:_pmi_alps_get_appLayout:pmi_alps_get_apid returned with error:
> Bad file descriptor
> > [nid01766:20603] mca_oob_tcp_init: unable to create IPv4 listen socket:
> Unable to open a TCP socket for out-of-band communications
> > ...
> > The installation was configured with the following command:
> > ./configure
> you are doing it local on the headnode only? It includes --with-alps by
> the platform file and there is the command `aprun` instead of `mpiexec` in
> the jobscript on a Cray. Maybe the allocation needs to be granted first.
> -- Reuti
> > I would be very happy if anybdy has an idea, what I could have missed
> during installation/runtime.
> > Thanks in advance.
> > Best regards
> > Christoph
> > --
> > Christoph Niethammer
> > High Performance Computing Center Stuttgart (HLRS)
> > Nobelstrasse 19
> > 70569 Stuttgart
> > Tel: ++49(0)711-685-87203
> > email: niethammer_at_[hidden]
> > http://www.hlrs.de/people/niethammer
> > _______________________________________________
> > users mailing list
> > users_at_[hidden]
> > http://www.open-mpi.org/mailman/listinfo.cgi/users
> users mailing list