[torqueusers] Segmentation fault when using OpenMPI -pernode option

Gus Correa gus at ldeo.columbia.edu
Mon Jun 4 11:09:14 MDT 2012


Hi Damian

I am not a Debian user,
but I would guess it is unlikely that
the Debian packages will add Torque support.
This is because OpenMPI can support SGE, Slurm, and other
resource managers, and you probably cannot add support for
all of them at the same time.
Which resource manager one chooses is a matter of taste.

It may be easier to just uninstall the Debian OpenMPI packages
and install OpenMPI from source, with Torque support,
in a non-system directory [/usr/local/openmpi,
/opt/openmpi-X.Y.Z, ...]

Something like this:

./configure --prefix=/a/non-system/directory 
--with-tm=/your/torque/directory ...
make
make install

Uninstalling the existent OpenMPI packages will save
you headaches with inconsistent/duplicate/mixed binaries,
libraries, paths, etc.

Then add the new OpenMPI directories to the
appropriate environment variables
[PATH and LD_LIBRARY_PATH] the way you prefer [say, via
.bashrc, .tcshrc]

I hope this helps,
Gus Correa

On 06/04/2012 12:44 PM, Damian Montaldo wrote:
> On Mon, Jun 4, 2012 at 12:58 PM, Damian Montaldo
> <damianmontaldo at gmail.com>  wrote:
>> On Mon, Jun 4, 2012 at 12:48 PM, Gus Correa<gus at ldeo.columbia.edu>  wrote:
>>> Hi Damian
>>>
>>> Did you build your OpenMPI with Torque support?
>>> Or did you install it from a Debian package?
>>
>> Hig Gus, I just install de Debian package of torque an openmpi (both
>> form the Debian repository)
>>
>>> The Debian OpenMPI package [if it exists] may not
>>> have Torque support.
>>> In this case, the mpiexec/mpirun probably won't
>>> know how to coordinate with Torque regarding
>>> nodes, cores, resources, etc.
>>>
>>> You can do
>>> 'mpicc --showme'
>>> to see if
>>> '-ltorque'
>>> appears there.
>>>
>>> I hope this helps,
>>> Gus Correa
>>
>> $ mpicc --showme
>> gcc -I/usr/lib/openmpi/include -I/usr/lib/openmpi/include/openmpi
>> -pthread -L/usr/lib/openmpi/lib -lmpi -lopen-rte -lopen-pal -ldl
>> -Wl,--export-dynamic -lnsl -lutil -lm -ldl
>>
>> You're right, thanks a lot!
>>
>> I'll try build it from source code instead of using the package from Debian.
>> If I cloud solve it I'll post it here.
>
> I was looking forward to report this bug but it was reported before here
>
> http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=592887
> Build with support for Torque (except on HURD).
>
> I've installed that version (1.4.2-4) but it seems to lack of support.
> I'll continue this issue in the debian related lists or in the OpemMPI list.
>
> Thanks.
> _______________________________________________
> torqueusers mailing list
> torqueusers at supercluster.org
> http://www.supercluster.org/mailman/listinfo/torqueusers



More information about the torqueusers mailing list