Having a simple DistributedMesh partitioner to compare vs ParMETIS would
certainly be very useful in driving the original issue forward.

The LinearPartitioner assert trip does appear reproducible with
gcc7.2/mpich3.2 when running the refinement example
<https://github.com/bboutkov/grins/tree/master_refine_test>from before.
The initial mesh can be made as small as 4x4 for easier debugging, and to
be totally explicit im building off libMesh master and manually attaching
the partitioner right before partition() in prepare_for_use().

Thanks for the help -

On Thu, Nov 9, 2017 at 10:50 AM, Roy Stogner <royst...@ices.utexas.edu>
wrote:

>
> On Thu, 9 Nov 2017, Boris Boutkov wrote:
>
> Well, I eliminated PETSc and have been linking to MPI using
>> --with-mpi=$MPI_DIR and playing with the refinement example I had
>> mentioned earlier to try and eliminate ParMETIS due the the
>> hang/crash issue. In these configs I attach either the
>> LinearPartitioner or an SFC in prepare_for_use right before calling
>> partition(). This causes assert trips in MeshComm::Redistribute
>> where elem.proc_id != proc_id while unpacking elems (stack below).
>>
>
> Shoot - I don't think either of those partitioners have been upgraded
> to be compatible with DistributedMesh use.  Just glancing at
> LinearPartitioner, it looks like it'll do fine for an *initial*
> partitioning, but then it'll scramble everything if it's ever asked to
> do a *repartitioning* on an already-distributed mesh.
>
> I could probably fix that pretty quickly, if you've got a test case I
> can replicate.
>
> SFC, on the other hand, I don't know about.  We do distributed
> space-filling-curve stuff elsewhere in the library with libHilbert,
> and it's not trivial.
> ---
> Roy
>
------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
Libmesh-devel mailing list
Libmesh-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/libmesh-devel

Reply via email to