Open MPI logo

Open MPI User's Mailing List Archives

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

Subject: Re: [OMPI users] deadlock on intercommunicator after MPI_Comm_spawn_multiple (OS X / Linux)
From: Ralph Castain (rhc_at_[hidden])
Date: 2012-12-05 00:03:56


Great! If you get tired of putting them on the cmd line, just remember
that, like all MCA params, you can instead express them in your
environment. For example,

export OMPI_MCA_oob_tcp_if_exclude=en0

for bash. Then you don't have to add them on the cmd line anymore. Just
FWIW.

On Tue, Dec 4, 2012 at 8:32 PM, Valentin Clement
<valentin.clement_at_[hidden]>wrote:

> Hi,
>
> Thanks, with this the problem is gone. In fact, no interface is posing
> problem if there is only one of them active. Anyway, using the options to
> mpiexec are just fine for the moment for me.
>
> Regards
>
> Valentin
>
>
> On Dec 5, 2012, at 11:19 AM, Ralph Castain <rhc_at_[hidden]> wrote:
>
> Strange - that shouldn't be happening. However, to get thru it, just
> restrict the interfaces OMPI uses. If it's the hardwired Ethernet that is
> causing the problem, then add
>
> -mca oob_tcp_if_exclude en0 -mca btl_tcp_if_exclude en0
>
> to your cmd line. If it's the wireless, then substitute en1 for en0 in the
> above.
>
>
>
> On Tue, Dec 4, 2012 at 5:16 PM, Valentin Clement <
> valentin.clement_at_[hidden]> wrote:
>
>> Hi,
>>
>> It seems, the problem is happening if I have two active interfaces on my
>> computer. Is there any configuration to use MPI_Comm_spawn_multiple on a
>> machine with multiple interfaces ?
>>
>> Regards,
>>
>> Valentin
>>
>> On Dec 3, 2012, at 3:00 PM, Valentin Clement <valentin.clement_at_[hidden]>
>> wrote:
>>
>> Hi,
>>
>> I'm using call to MPI_Comm_spawn_multiple in a quite big application.
>> I've seen a deadlock occurred in a very strange situation. If I'm running
>> my application on my Ubuntu 12.10 with OpenMPI 1.6.3 there is absolutely no
>> problem.
>>
>> On my Mac OS X 10.8.2 with also OpenMPI 1.6.3, I'm experiencing a dead
>> lock on an intrecommunicator resulting from the MPI_Comm_spawn_multiple
>> only if my ethernet interface is enable. If I disable it, the deadlock is
>> gone.
>>
>> Anyone has an idea of what is happening ? I joined the output of
>> ompi_info on both OS X and Linux.
>>
>> Regards,
>>
>> Valentin
>>
>>
>>
>> -----------------------------------------------------------------------------------------
>> Valentin Clement - Student trainee at RIKEN AICS
>> Programming Environment Research Team
>> valentin.clement_at_[hidden]
>> valentin.clement_at_[hidden]
>> Master thesis project
>> POP-C++ on the K Computer
>> Project homepage: https://forge.tic.eia-fr.ch/projects/poponk
>> Project board: https://forge.tic.eia-fr.ch/projects/poponk/wiki/Wiki
>>
>> -----------------------------------------------------------------------------------------
>> <ompi_info.tar.bz2>
>> _______________________________________________
>> users mailing list
>> users_at_[hidden]
>> http://www.open-mpi.org/mailman/listinfo.cgi/users
>>
>>
>> On Dec 3, 2012, at 3:00 PM, Valentin Clement <valentin.clement_at_[hidden]>
>> wrote:
>>
>> Hi,
>>
>> I'm using call to MPI_Comm_spawn_multiple in a quite big application.
>> I've seen a deadlock occurred in a very strange situation. If I'm running
>> my application on my Ubuntu 12.10 with OpenMPI 1.6.3 there is absolutely no
>> problem.
>>
>> On my Mac OS X 10.8.2 with also OpenMPI 1.6.3, I'm experiencing a dead
>> lock on an intrecommunicator resulting from the MPI_Comm_spawn_multiple
>> only if my ethernet interface is enable. If I disable it, the deadlock is
>> gone.
>>
>> Anyone has an idea of what is happening ? I joined the output of
>> ompi_info on both OS X and Linux.
>>
>> Regards,
>>
>> Valentin
>>
>>
>>
>> -----------------------------------------------------------------------------------------
>> Valentin Clement - Student trainee at RIKEN AICS
>> Programming Environment Research Team
>> valentin.clement_at_[hidden]
>> valentin.clement_at_[hidden]
>> Master thesis project
>> POP-C++ on the K Computer
>> Project homepage: https://forge.tic.eia-fr.ch/projects/poponk
>> Project board: https://forge.tic.eia-fr.ch/projects/poponk/wiki/Wiki
>>
>> -----------------------------------------------------------------------------------------
>> <ompi_info.tar.bz2>
>> _______________________________________________
>> users mailing list
>> users_at_[hidden]
>> http://www.open-mpi.org/mailman/listinfo.cgi/users
>>
>>
>>
>> -----------------------------------------------------------------------------------------
>> Valentin Clement - Student trainee at RIKEN AICS
>> Programming Environment Research Team
>> valentin.clement_at_[hidden]
>> valentin.clement_at_[hidden]
>> Master thesis project
>> POP-C++ on the K Computer
>> Project homepage: https://forge.tic.eia-fr.ch/projects/poponk
>> Project board: https://forge.tic.eia-fr.ch/projects/poponk/wiki/Wiki
>>
>> -----------------------------------------------------------------------------------------
>>
>>
>> _______________________________________________
>> users mailing list
>> users_at_[hidden]
>> http://www.open-mpi.org/mailman/listinfo.cgi/users
>>
>
> _______________________________________________
> users mailing list
> users_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/users
>
>
>
> -----------------------------------------------------------------------------------------
> Valentin Clement - Student trainee at RIKEN AICS
> Programming Environment Research Team
> valentin.clement_at_[hidden]
> valentin.clement_at_[hidden]
> Master thesis project
> POP-C++ on the K Computer
> Project homepage: https://forge.tic.eia-fr.ch/projects/poponk
> Project board: https://forge.tic.eia-fr.ch/projects/poponk/wiki/Wiki
>
> -----------------------------------------------------------------------------------------
>
>
> _______________________________________________
> users mailing list
> users_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/users
>