I was wondering is there is a way to obtain a global numbering of the nodes=
that does not change with partitions.
The point of that question is to perform a study of the influence of the=20
partitioning on the computed solution for a given problem.
I saw that system and mesh can be dumped to file, but with the understandin=
have reordering always occurs and then does not allow that kind of=20
Maybe there is a possibility to sort nodes by spatial criterion (based on=20
boxes for example) ?
Does any one have a clew ?
Guillaume ANCIAUX Tel : 06 89 29 11 50=20
LaBRI, Universit=E9 Bordeaux I (+33)5 40 00 38 21 (bureau)=20
351, Cours de la Lib=E9ration anciaux@...=20
33405 TALENCE Cedex, FRANCE http://dept-info.labri.fr/~anciaux
From: Roy Stogner <roystgnr@ic...> - 2006-06-19 16:02:29
On Mon, 19 Jun 2006, Guillaume ANCIAUX wrote:
> I was wondering is there is a way to obtain a global numbering of the nodes
> that does not change with partitions.
The node numbering should stay the same unless you remove nodes,
shouldn't it? I know we've got more than a little code in some of the
finite element classes that assumes node ordering is never changed.
Keep in mind that the node numbering and the degree of freedom
numbering aren't the same thing.
I don't think there's a natural way to keep the degree of freedom
numbering constant through a repartitioning, since we want to keep a
contiguous range of degree of freedom indices on each processor, and
trying to repartition while maintaining that state and without
renumbering would be too limiting.
Can you create such a numbering yourself? If you're using
isoparametric Lagrange variables, then add another one, and assign to
each of its indices a unique value. After any repartitioning, the
System::project_vector calls should make sure that your lookup
variable gets updated correctly.