[Dune] ALUGrid loadbalance

[Andreas Dedner]

Hi.
Looks weird - a bit of increase is expected due to the ghost elements 
but not to that extend.
Have you simply taken a look at the resulting partitioned grid in paraview?
Andreas

On 02/07/15 16:00, Arya Fallahi wrote:
> Dear Dune,
>
> Sometime ago I upgraded my alugrid and started using the one 
> integrated into dune library, namely dune-alugrid. I was always 
> suffering from slow computation time that I suddenly realized today 
> that something is going wrong with my load balancing. In the 
> following, I am sending you an example of what I do in my code:
>
> typedef GridSelector::GridType Grid;
>
> MPI_Init(&argc,&argv);
> MPI_Comm_rank(MPI_COMM_WORLD,&rank);
> MPI_Comm_size(MPI_COMM_WORLD,&size);
>
> GridPtr<Grid> ag ( "./grid1.dgf" , MPI_COMM_WORLD );
> Grid& g = *ag;
> g.loadBalance();
> typedef LeafMultipleCodimMultipleGeomTypeMapper<Grid,P0Layout> emap;
> emap em(g);
> std::cout << em.size() << std::endl;
>
> Without load balancing I get the following output on 4 processors:
>
> 0
> 0
> 0
> 15468
>
> meaning that the grid has 15648 elements. However, with load balancing 
> I get the following output:
>
> 11235
> 11869
> 12906
> 11244
>
> It seems that the total number of elements are getting considerably 
> higher. Do you have any guess what I am doing wrong?
>
> Thank you very much for your help,
> Arya
>
>
> _______________________________________________
> Dune mailing list
> Dune at dune-project.org
> http://lists.dune-project.org/mailman/listinfo/dune

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.dune-project.org/pipermail/dune/attachments/20150702/1f16ceaa/attachment.htm>