[Dune] GridFactory and parallel grids (and boundary adaptation)

[Aleksejs Fomins]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Also, I have already tried constructing Alugrid in parallel using our
reader, it works no problem

On 11/02/15 13:34, Aleksejs Fomins wrote:
> Dear Jo,
> 
> CurvilinearGMSHReader is currently capable of reading meshes in 
> parallel. That is, each process reads only the part of the mesh
> which it needs. Adapting this reader to non-curvilinear grids
> should be almost as straightforward as only reading linear gmsh
> files. It uses a slightly extended version of a GridFactory which
> allows to insert additional information into the Grid, however,
> adapting this to use standard Dune GridFactory and disregard
> curvilinear information should not be too much work. Please be a
> little patient, we will definitely make this tool public before
> summer.
> 
> Best, Aleksejs
> 
> On 11/02/15 12:52, Jö Fahlke wrote:
>> Am Tue, 10. Feb 2015, 19:02:12 +0100 schrieb Jö Fahlke:
>>> With regard to loading a parallel grid with the GridFactory,
>>> the current informal standard seems to be inserting on rank 0
>>> only, inserting nothing on the other ranks, and following that
>>> by a loadBalance().  I know that worked with the old ALUGrid,
>>> and apparently UG supports that too[1].  Is that still the way
>>> to go with dune-alugrid or are there better ways (so long as
>>> you want to keep using the GridFactory?).
> 
>> As Robert pointed out to me, there is indeed a better way in 
>> ALUGrid, see[1]: insert each vertex on each rank it is needed
>> with a globally unique identifier, and optionally identify
>> process and domain boundaries.
> 
>> [1] http://arxiv.org/abs/1407.6954, page 11-12
> 
>> There is no mention about GmshReader however.  I am pretty
>> certain that there is no specialization of the GmshReader that
>> takes advantage of the parallel GridFactory, though it should be
>> pretty straight-forward to add one (GSoC anyone?).  Still the
>> method mentioned above should work as long as the grid data fits
>> into rank 0's memory.  Please correct me if I'm wrong.
> 
>> There is also no mention about parallel boundary adaptation, so
>> I'd still be interested if there is some ALUGrid-specific way to
>> do that.
> 
>> Regards, Jö.
> 
> 
> 
>> _______________________________________________ Dune mailing list
>>  Dune at dune-project.org 
>> http://lists.dune-project.org/mailman/listinfo/dune
> 
> 
> _______________________________________________ Dune mailing list 
> Dune at dune-project.org 
> http://lists.dune-project.org/mailman/listinfo/dune
> 
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.12 (GNU/Linux)
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=N06z
-----END PGP SIGNATURE-----