[Libmesh-users] Problem: reading mesh data in parallel

[Jens O. <oe...@st...>]

Hi,

I'm having a little bit trouble. The problem is the execution of my
code in parallel.

In the current implementation processor 0 reads in the mesh. After that
the mesh is distributed to every other processor. In the read method of
the class UnvMeshInterface is a call for close_foreign_id_maps() so that
the id_maps are closed after that, but only on processor 0. 
If you now try to read in mesh data in parallel, the programm would tell
you something like this:

src/mesh/mesh_data.C:257: void MeshData::read(const std::string&):
Assertion `_elem_id_map_closed && _node_id_map_closed' failed.

This is correct, since on processor 1, 2 and so on are the id_maps not
closed. If now try to close the id_maps with a little hack in Mesh::read
you will see, that there are no _elem_id, _id_elem, _node_id and
_id_node on these processors.
A solution for this could be to distribute these things with the mesh
and to call close_foreign_id_maps() in Mesh::read. Or to change the
read implementation Mesh and MeshData classes. In my opinion this should
be reading in the mesh on processor 0 and than close the id_maps. After
this processor 0 reads in the mesh data. At least the mesh and
mesh data should be distributed to every processor.

Am I'm right or did i misunderstood something?

I've tried to run example 12 on 2 or 3 processor. I got nearly the same
error message so I think that it's not my implementation of the TetGen
interface, which is more or less like UnvMeshInterface class.

Any hints or comments are welcome.

Best regards
Jens.

PS: Would you like to add the TetGen methods after some more tests from
my side?


--