On Mon, 28 Jan 2008, Umar Farooq wrote:

> I was wondering if anyone has an example of how to use the ParallelMesh

Ideally, all the examples will eventually work with ParallelMesh, and
the only concern for the user will be to decide whether their code is
better run with the memory overhead of SerialMesh or the runtime
overhead (at least on small numbers of processors) of ParallelMesh.

At the moment, we've got example 8 turned off when ParallelMesh is the
default Mesh implementation, and ParallelMesh is so untested that (at
least for adaptive codes) I wouldn't yet trust it with any critical
results.

> with an assemble function that loops through element nodes for
> imposing boundary conditions, after the element matrix has been
> added to the system matrix.

The only important concern here is that not all element nodes will
exist in your local chunk of the mesh, so you'd probably want to have
each processor either loop only over local nodes, or loop only over
local elements (and then loop over nodes on each element).

> Ex9 was a good template for my problem a while ago, but Im not sure if the
> current svn version of Ex9.C is correct.  For example, should the system
> matrix look diferent for seq/2/4 procs?

I don't think it should look different (up to some multiple of FPU
precision, thanks to entries being summed in different orders).  Does
it look significantly different?  I'll try it on my own workstation
now and see if I can replicate the problem.
---
Roy

-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
Libmesh-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-devel

Reply via email to