[MOAB-dev] Mesh partitioning question(s)

Dmitry Karpeev karpeev at mcs.anl.gov
Tue Mar 9 21:07:57 CST 2010


On Tue, Mar 9, 2010 at 7:35 PM, Tim Tautges <tautges at mcs.anl.gov> wrote:
>
>
> Dmitry Karpeev wrote:
>>
>> A couple of questions (any clues appreciated):
>>
>> 1) What are the appropriate mesh file formats that support partial file
>> read-in?
>> For example, when I try to read in a .cub mesh, I can't have each
>> process just read
>> its appropriate piece in.  In terms of parallel/mbparallelcomm_test, I
>> can't use
>> readpar_option == -2.  I have meshes in the VTK format here, as well
>> as some .h5m
>> and .smf mesh files.
>>
>
> If you're referring to the parallel read option, only h5m, I believe.  When
> you try other file formats, do they fail or just not behave as advertised?
>  That should be logged in a ticket...

.cub definitely doesn't work with the parallel read option READ_PARALLEL.
I got a lot of different files from Iulian to try it on, though (.h5m,
.vtk, which I can convert with mbconvert).
I guess I'm not exactly sure how some of the files should behave, but
in one case I definitely
have something that should be considered a bug (or documented):
appearance of negative values
in MPI_Bcast, where they are not expected.  I also have a lot of cases
of a parallel reads "hanging".
I'll submit those tickets tomorrow morning.

>
>> 2) In absence of GEOM_DIMENSION tag, what is the appropriate tag to define
>> a parallel mesh partition?
>
> I've used MATERIAL_SET in the past, as well as PARALLEL_PARTITION (which is
> the tag used by mbzoltan to write the partition).

Thanks.  I've been thinking about partitioning files before running these tests.
mbconvert+mbzoltan should prepare enough files for testing.

Dmitry.
>
> - tim
>
>>
>> Thanks.
>>
>> Dmitry.
>>
>
> --
> ================================================================
> "You will keep in perfect peace him whose mind is
>  steadfast, because he trusts in you."               Isaiah 26:3
>
>             Tim Tautges            Argonne National Laboratory
>         (tautges at mcs.anl.gov)      (telecommuting from UW-Madison)
>         phone: (608) 263-8485      1500 Engineering Dr.
>           fax: (608) 263-4499      Madison, WI 53706
>
>


More information about the moab-dev mailing list