Do you have the Linux firewall running on either of your machines,
perchance? This can either block random socket connections between
nodes (which Open MPI's TCP communication will use) or eat the
connection requests in a black-hole fashion such that the connections
will timeout.
On Jan 16, 2008, at 5:35 AM, Roland Albrecht wrote:
Hello
I'm running an FDTD programm (meep) using open-mpi on a mini-cluster
consisting of 2 computers. Since the exchange of the mainbord on the
node (with an identical one as before) I have a problem. I can't
find the change in the configurations which is now causing the
problen.
Here's my problem:
I can start the meep application by mpi-run on each node
individually and the program runs without any problems.
However when I try to run the program distributed over both
computers I get at some point the following error message:
...[0,1,1][btl_tcp_endpoint.c:
572:mca_btl_tcp_endpoint_complete_connect] connect() failed with
errno=110
Which translates by Perl as: Connection timed out at -e line 1.
However I can't figure out where the problem lies in my network
configuration. SSH tunnels from one computer to another works. I
also can reach the internet from the node.
In the attached archive there's the config.log from the top open-mpi
tree, there's the output of ompi_info --all and there's the network
configuration of both computers.
I'm really greatfull for any help. Thank you!
Best regards
Roland Albrecht
--
___________________________________________
Roland Albrecht, Dipl. Phys. ETH
-------------------------------------------
Universität des Saarlandes
Fachrichtung 7.3 (Technische Physik)
AG Prof. Dr. Christoph Becher
Campus E2.6, Zimmer 2.04
D-66123 Saarbrücken
Germany
Phone:+49(0)681 302 3418
Fax: +49(0)681 302 4676
skype: roland_albrecht
<mpi.rar>_______________________________________________
users mailing list
us...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users
--
Jeff Squyres
Cisco Systems
