Re: [Libmesh-users] Question for using LibMesh Library with my own parallel method

[Stogner, R. H <roy...@ic...>]

On Wed, 6 Mar 2019, 서승진 wrote:

> In my problem, I will solve revised poisson equation ({Ke}[u] = [Fe]) for ten mesh files with commands
> 
>   mpiexec -np 10 ./example-opt -d 2 -n 15
> 
>   so that ten processors will solve the physics;
>    processor 0 = mesh0.msh
>    processor 1 = mesh1.msh
>    processor 2 = mesh2.msh
>    ... (to processor 9 and mesh9.msh)
> 
> However, the LibMesh will automatically partition and assign a mesh
> into several mesh to conduct parallel communication, and some
> LibMesh function is quitely related with the parallel communicator,
> such as;
> 
>   Mesh mesh (init.comm());

It's automatic, but it *only* automatically parallelizes work over the
communicator you use to construct the Mesh (which then gets passed
along to the EquationSystems and so forth).

> What I want to ask is that how can I just use LibMesh Library
> without LibMesh's function for parallel calculation? (for applying
> my own parallel logic with multi-meshes by using one precessor by
> one mesh concept)

If you want each Mesh to only exist on one processor, try
Mesh mesh (Parallel::Communicator(MPI_COMM_SELF));

Please let us know if you run into any problems!  We've had more than
a few users of this capability but I don't know if we have enough
*test coverage* of this capability, so it's possible there have been
regressions.

> Thanks for reading my question.

No problem, but in the future please send questions to the libmesh
users mailing list.  Anything sent just to me risks getting buried in
my inbox (so you might not get an answer) and doesn't get added to the
mailing list archives (so future users with the same question won't
get an answer quickly).
---
Roy