[MOAB-dev] [Fathom] Parallel read issues: mbconvert vs mbparallelcomm_test (PARALLEL_RESOLVE_SHARED_ENTS)

[Tim Tautges]

Resolving shared entities for a broadcast doesn't make any sense.  In this read method, all processors have the same 
exact mesh, so all entities will be shared.  Maybe in principal this should work anyway, but it's pretty far outside the 
mainstream.

- tim

Dmitry Karpeev wrote:
> Hi,
> It appears that some of the problems I had earlier with reading meshes
> in parallel have to do with
> the resolution of shared entities.  mbconvert typically read the files
> in without a problem, while mbparallelcomm_test
> refused to read some files and not others, etc (it was confusing).
> 
> I stripped off most of the mbconvert code to make a simple
> benchmarking tool (tools/mbparread).
> Unlike mbparallelcomm_test, mbparread (and mbconvert, from which it
> derives) does not set any parallel reader
> options by default.  When reintroducing them one by one, I stumbled
> upon the following issue:
> feeding PARALLEL_RESOLVE_SHARED_ENTS can cause mbparread to fail.
> It appears that problems are more complicated when reading
> tjunc6RIB_16.h5m, but with blocks-64.h5m
> it is more straightforward (this is a 2 proc version on my laptop):
> 
> With PARALLEL=BCAST:
> ------------------------------------------------------------------------------------------------------------------------------------------------------
> [hal Mon Apr 05 05:37 PM] ~/fathom/moab/trunk
>> $MPI2_PREFIX/bin/mpiexec -n 2
> /home/karpeev/fathom/moab/unstable/build/tools/mbparread -O
> PARTITION_DISTRIBUTE -O CPUTIME -O PARTITION=MATERIAL_SET -O
> PARALLEL=BCAST /home/karpeev/fathom/moab/data/blocks-64.h5m
> Read times: 0.448433 0.00270009 5.96046e-06 0.892087 (PARALLEL
> READ/PARALLEL CHECK_GIDS_SERIAL/PARALLEL GET_FILESET_ENTS/PARALLEL
> BROADCAST/)
> [proc 0 of 2]: Read "/home/karpeev/fathom/moab/data/blocks-64.h5m"
> [proc 1 of 2]: Read "/home/karpeev/fathom/moab/data/blocks-64.h5m"
> 
> With PARALLEL=BCAST and PARALLEL_RESOLVE_SHARED_ENTS
> ------------------------------------------------------------------------------------------------------------------------------------------------------
> [hal Mon Apr 05 05:38 PM] ~/fathom/moab/trunk
>> $MPI2_PREFIX/bin/mpiexec -n 2
> /home/karpeev/fathom/moab/unstable/build/tools/mbparread -O
> PARTITION_DISTRIBUTE -O CPUTIME -O PARTITION=MATERIAL_SET -O
> PARALLEL=BCAST -O PARALLEL_RESOLVE_SHARED_ENTS
> /home/karpeev/fathom/moab/data/blocks-64.h5m
> Failed to load "/home/karpeev/fathom/moab/data/blocks-64.h5m".
> Error code: MB_ENTITY_NOT_FOUND (4)
> Error message: Failed in step PARALLEL RESOLVE_SHARED_ENTS
> 
> 
> Failed to load "/home/karpeev/fathom/moab/data/blocks-64.h5m".
> Error code: MB_ENTITY_NOT_FOUND (4)
> Error message: Failed in step PARALLEL RESOLVE_SHARED_ENTS
> ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
> 
> The other read modes (BCAST_DELETE, READ_DELETE, READ_PART) appear to
> work fine with and without PARALLEL_RESOLVE_SHARED_ENTS.
> Any ideas what may be going on?
> The mesh files I used can be found in ~karpeev/fathom/moab/data on MCS
> machines and on cosmea.
> 
> Thanks.
> Dmitry.
> _______________________________________________
> Fathom mailing list
> Fathom at lists.mcs.anl.gov
> https://lists.mcs.anl.gov/mailman/listinfo/fathom
> 

-- 
================================================================
"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