[gpaw-users] Questions about Parallel Runs

[Hongliang Xin]

Here is the output. The gpaw has been built with mpi support. The output
below does not seem right to me. We built gpaw with the system mpi, but it
shows below that the library is linked to our user mpi (/home/morabim). Is
this the problem? I will take a look and rebuild the gpaw.

thanks,

: ldd `which gpaw-python`
    libgfortran.so.1 => /usr/lib64/libgfortran.so.1 (0x00002aeb72229000)
    libpython2.4.so.1.0 => /usr/lib64/libpython2.4.so.1.0
(0x00002aeb724c0000)
    libpthread.so.0 => /lib64/libpthread.so.0 (0x00002aeb727f1000)
    libdl.so.2 => /lib64/libdl.so.2 (0x00002aeb72a0d000)
    libutil.so.1 => /lib64/libutil.so.1 (0x00002aeb72c11000)
    libm.so.6 => /lib64/libm.so.6 (0x00002aeb72e14000)
    libmpi.so.0 => /home/morabitm/usr/lib/libmpi.so.0 (0x00002aeb73098000)
    libopen-rte.so.0 => /home/morabitm/usr/lib/libopen-rte.so.0
(0x00002aeb7322c000)
    libopen-pal.so.0 => /home/morabitm/usr/lib/libopen-pal.so.0
(0x00002aeb73385000)
    libnsl.so.1 => /lib64/libnsl.so.1 (0x00002aeb734e5000)
    libc.so.6 => /lib64/libc.so.6 (0x00002aeb736fd000)
    /lib64/ld-linux-x86-64.so.2 (0x00002aeb7200c000)

On Mon, Apr 12, 2010 at 4:58 AM, Marcin Dulak <Marcin.Dulak at fysik.dtu.dk>wrote:

> Hi,
>
> is you gpaw built with mpi support?
> What does:
> ldd `which gpaw-python`
> say?
> To build the parallel version, mpicc must be in PATH at the build time,
> or the customize.py must include the relevant mpi information.
> Please look at the following example of  the
> https://trac.fysik.dtu.dk/projects/gpaw/browser/trunk/doc/install/Linux/Niflheim/customize-thul-acml.py
> (the scalapack settings and the platform_id variable at the end of the
> customize.py are not relevant to your case - anyway
> the example of using platform_id is described here
> https://wiki.fysik.dtu.dk/gpaw/install/Linux/Niflheim/Niflheim.html#niflheim
> ).
>
> Best regards,
>
> Marcin
>
> Hongliang Xin wrote:
>
>> Dear gpaw-users,
>>
>> We have been trying to figure out what is the problem with the parallel
>> gpaw calculations recently.
>> I know there is definitely the problem due to the following several
>> observations:
>> 1. There is no benefit for calculations with multiple processors compared
>> with one serial calculation.
>> 2. It seems it only runs several serial versions of the program without
>> any communication between nodes. 3. After calculation is done, there are
>> always .gpaw and another .old.gpaw files with exactly the same size.
>> 4. I did not see any indication that the job is paralleled.
>>
>> The gpaw is compiled without difficulty with the configuration as shown in
>> the bottom. The job is submitted according to the gpaw documentation using:
>> mpirun -np $NP -machinefile $PBS_NODEFILE gpaw-python Script.py
>>
>> Any suggestion will be very welcome.
>>
>> thanks,
>>
>> Hongliang
>>
>>
>> * Using standard lapack
>> * Architecture: linux-x86_64
>> * Building a custom interpreter
>> libraries ['acml', 'gfortran']
>> library_dirs ['/home/software/rhel5/acml/4.0.1-gcc/gfortran64/lib']
>> include_dirs
>> ['/home/software/rhel5/python/lib64/python2.4/site-packages/numpy/core/include',
>> 'c/libxc']
>> define_macros []
>> extra_link_args []
>> extra_compile_args ['-Wall', '-std=c99']
>> runtime_library_dirs []
>> extra_objects []
>>
>> Parallel configuration
>> mpicompiler mpicc
>> mpi_libraries []
>> mpi_library_dirs []
>> mpi_include_dirs []
>> mpi_define_macros []
>> mpi_runtime_library_dirs []
>>
>>
>>
>> --
>> Hongliang Xin
>> Ph.D. Candidate
>> Dept. of Chemical Engineering
>> University of Michigan
>> 3166 HH Dow
>> 2300 Hayward
>> Ann Arbor, MI 48109
>> Phone: (734) 647-8051
>> E-mail: hxin at umich.edu <mailto:hxin at umich.edu>
>> ------------------------------------------------------------------------
>>
>> _______________________________________________
>> gpaw-users mailing list
>> gpaw-users at listserv.fysik.dtu.dk
>> https://listserv.fysik.dtu.dk/mailman/listinfo/gpaw-users
>>
>
> --
> ***********************************
>
> Marcin Dulak
> Technical University of Denmark
> Department of Physics
> Building 307, Room 229
> DK-2800 Kongens Lyngby
> Denmark
> Tel.: (+45) 4525 3157
> Fax.: (+45) 4593 2399
> email: Marcin.Dulak at fysik.dtu.dk
>
> ***********************************
>
>


-- 
Hongliang Xin
Ph.D. Candidate
Dept. of Chemical Engineering
University of Michigan
3166 HH Dow
2300 Hayward
Ann Arbor, MI 48109
Phone: (734) 647-8051
E-mail: hxin at umich.edu
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://listserv.fysik.dtu.dk/pipermail/gpaw-users/attachments/20100412/9a36df40/attachment.html