Open MPI logo

Open MPI User's Mailing List Archives

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

From: James Conway (jxc100+_at_[hidden])
Date: 2006-02-13 12:25:35


Jeff, and George,

I have success to report: --mca btl_tcp_if_include was needed.

>>> - Have you absolutely entirely disabled all firewalling between
>>> the two hosts?
>>
>> As far as I know - simply, hit the "Stop" button on Mac OSX
>> Sharing pref-panel for Firewall, on the local and remote systems
>> both (one is my PowerBook G4, the other my PowerMac G5).
>
> Sounds like this should be sufficient.

Indeed, I can confirm that it is.

>>> - Do you have only one TCP interface on both machines? If you
>>> have more than one, we can try telling Open MPI to ignore one of
>>> them.
>>
>> Interesting idea. The remote machine has two ethernet ports
>> (PowerMac) and the local machine has ethernet and airport. Only
>> one port should be enabled on each, but the PowerBook airport is
>> what I use at home so maybe it didn't get properly disabled when I
>> switched to my work settings. Since the call to MPI-send seems to
>> hand on the local host, it may be an attempt to use the airport
>> (wireless) connection. How to I tell Open MPI to ignore a
>> particular interface?
>
> Check out http://www.open-mpi.org/faq/?category=tcp#tcp-selection
> -- it talks about the MCA parameters you can use to specify
> different networks. For example, on my powerbook, en0 appears to
> be my wired connection.
>
> Let us know what happens.

By including --mca btl_tcp_if_include on the command line, the ring
program continues past the first round to completion. So even though
my non-ethernet interfaces were disabled (airport, firewire), one of
them seems to have been sufficiently active to get in the way. (In
fact, about a week ago I started to be suspicious of a hardware fault
on my PowerBook's airport card, and I have seen it attempting but
failing to make connections when it was supposedly disabled). The
Open MPI command line then is:

   mpirun --hostfile mpi_hosts --mca btl_tcp_if_include en0 --np 2
mpi_test1

where en0 goes to the ethernet port on my PowerBook G4, and on to the
remote PowerMac.

On Feb 13, 2006, at 12:14 AM, George Bosilca wrote:

> I not 100% sure but I think I might know what's wrong. I can
> reproduce something similar (oddly it does not happens all the
> time) if I activate my firewall and let all the trafic through (ie.
> accept all connections). In few words, I think the firewall (even
> when disabled) introduce some delays in the setup stage of the TCP
> connection and we "kind of" lose one of the messages. Let me find a
> high delay cluster and I will take a look.

It may be related - I would be interested to know if you made any
progress on this. For now, I have the Firewall disabled (stopped) for
testing, and I am sure to be OK since my fingers are crossed, right?
(My test systems are behind a departmental firewall, so as long as I
can trust my co-workers - and of course I do - the fingers-crossed
method should suffice until I wire up a private network.)

Thanks for the insight, Jeff. I look forward to progressing to real
MPI software (which I have kindly been given).

Best regards,

James Conway
----------------------------------------------------------------------
James Conway, PhD.,
Department of Structural Biology
University of Pittsburgh School of Medicine
Biomedical Science Tower 3, Room 2047
3501 5th Ave
Pittsburgh, PA 15260
U.S.A.
Phone: +1-412-383-9847
Fax: +1-412-648-8998
Email: jxc100_at_[hidden]
Web: <http://www.pitt.edu/~jxc100/> (under construction)
----------------------------------------------------------------------