Most likely problem is version confusion as Fedora comes with a version of OMPI pre-installed. Check you LD_LIBRARY_PATH and PATH on both ends to ensure they are pointing at the same OMPI installation. Remember, interactive login typically uses a different shell startup then non-interactive job launch, so be sure that the latter is getting the right path.

Also, be sure you don't have a firewall running between the two computers.

What version of OMPI are you trying to install?

On Mon, Jul 27, 2009 at 9:35 AM, jacob Balthazor <jcb344@gmail.com> wrote:

Hey,

      Please help me out as I cannot figure out from all the online documentation why my cluster wont work. 

1) I started with two computers with a fresh install of Fedora 10.
2) Created two user accounts on the two computers with the same user name.
3) Created an rsa keyed ssh between the two computers going both directions, so that each computer can log into one another without being prompted for a password.
4) Installed openmpi on both via yum with the command "yum install openmpi openmpi-devel openmpi-libs".
5) Compiled and ran hello.c with a host file containing the IP addresses of both computers with slot numbers. But openmpi only hangs for minutes, until I kill it.

      I am able to run hello.c locally and even with a hostfile containing only the local machines address, but am not able to started a multi node program from either computer. Have I done all the correct steps, am I missing something?

- Jacob B.




_______________________________________________
users mailing list
users@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users