Open MPI opens random TCP sockets during the startup phase of MPI processes -- mostly
from the "orted" helper process that is started on each node (or VM) back to the initiating mpirun process.
Do you have firewalling or other TCP blocking software running? Or are the appropriate TCP routes setup between all your VMs?
On Apr 2, 2010, at 5:00 PM, rohan nigam wrote:
> I am trying to run a simple hello world program before actually launching some very heavy load testing over the Xen SMP set up that I have.
> I am trying to run this command over four different hosts, Dom0 being the host where i am launching mpirun and rest three being xen guest domains.
> I have password less login setup across all the hosts. These hosts are actually on AMD Opteron dual socket Quad core with 2 cores assigned to each host/domain.
> mpirun --prefix /root/xentools/openmpi-gcc/ -mca plm_rsh_agent ssh --host localhost, xenguest1 -np 4
> I am able to run mpirun successfully when I launch this command from one of the guests and also when i lauch this command on dom0 (localhost) alone. But when i launch mpirun from the Dom 0 on one or more guests there is no response from the guests and I am eventually having to kill the process which reports saying
> xenguest1 - daemon did not report back when launched
> Can someone point out where I am going wrong. I have seen people having similar problem in the list but no one posted how they got around this problem.
> Note: I also tried setting the default agent launcher to ssh. Also, on every host the directory structure is exactly the same and also the Library paths and paths are also set up properly and the executable is also present at the exact same location..
> Rohan Nigam
> Research Asst,
Univ. of Houston
> users mailing list
For corporate legal information go to:http://www.cisco.com/web/about/doing_business/legal/cri/
users mailing email@example.com://www.open-mpi.org/mailman/listinfo.cgi/users