Open MPI logo

Open MPI User's Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Open MPI User's mailing list

Subject: [OMPI users] tcp connectivity OS X and 1.3.3
From: Jody Klymak (jklymak_at_[hidden])
Date: 2009-08-11 15:54:29


Hello,

On Aug 11, 2009, at 8:15 AM, Ralph Castain wrote:

> You can turn off those mca params I gave you as you are now past
> that point. I know there are others that can help debug that TCP btl
> error, but they can help you there.

Just to eliminate the mitgcm from the debugging I compiled example/
hello_c.c and run as:

  /usr/local/openmpi/bin/mpirun --debug-daemons -n 8 -host xserve01
hello_c >& hello_c4_1host.txt

There is no ostensible problem. If I run as:

/usr/local/openmpi/bin/mpirun --debug-daemons -n 8 -host
xserve01,xserve02 hello_c >& hello_c4_2host.txt

The process says Hello, but hangs at the end, and needs to be killed
with ^C.

I then modified connectivity_c to include a printf as MPI is
initialized, and hardwired verbose=1. This completes, and appears to
work fine..

/usr/local/openmpi/bin/mpirun --debug-daemons -n 8 -host xserve01
connectivity_c >& connectivity_c8_1host.txt

However, again, two hosts sours the mix:

/usr/local/openmpi/bin/mpirun --debug-daemons -n 8 -host
xserve01,xserve02 connectivity_c >& connectivity_c8_2host.txt

This hangs, and after waiting a minute or so we see that rank 0--4 on
xserve01 cannot contact rank 5 (presumably on xserve02).

It seems that I have something wrong in my tcp setup, but
communication between these servers worked yesterday using 1.1.5, and
ping etc all work fine, so something else is up. Some sort of port
permissions?

Th most glaring error I see in these is:

[xserve02.local:43625] [[28627,0],2] orte:daemon:send_relay -
recipient list is empty!

I see reference in the archives to a similar error where
"contacts.txt" could not be found. I've had trouble with 10.5.7 with
temporary directories, so maybe that is the issue?

Thanks Jody

--
Jody Klymak
http://web.uvic.ca/~jklymak/