[Dune] parallel alugrid

[Andreas Dedner]

If ALU<2,2> is serial depends on which version you are using but quite
likely it is since there is no official version of the parallel grid yet
available. But also with ALUGrid you need to set up the grid only on
process zero (empty grids on other processes) and then loadbalance
distributes the elements
Andreas

On 09/09/15 14:09, Ganesh Diwan wrote:
>     Created serial ALUGrid<2,2,cube,nonconforming>.
> 
> 
> Is this the problem as the grid is still serial?
> 
> Thank you
> Ganesh
> 
> On Wed, Sep 9, 2015 at 2:07 PM, Ganesh Diwan <gcdiwan83 at gmail.com
> <mailto:gcdiwan83 at gmail.com>> wrote:
> 
>     Hi Andreas
> 
>     Here is the code:
> 
>     // Maybe initialize MPI
>             Dune::MPIHelper& helper = Dune::MPIHelper::instance(argc, argv);
> 
>             typedef Dune::ALUGrid<2, 2, Dune::cube, Dune::nonconforming>
>     GridType;
>            
>     //        typedef Dune::UGGrid<2> GridType;
>     //        std::auto_ptr<HGridType> grid(
>     GmshReader<HGridType>::read( "./inp/mymesh.msh", true, true ) );
>            
>             Dune::GridFactory<GridType> factory;
>             Dune::GmshReader<GridType>::read(factory,
>     "./inp/mymesh.msh", true, true);
>             GridType *grid = factory.createGrid();
>             typedef GridType::LeafGridView GV; // Define a typedef for a
>     Leaf Grid View
>             const GridType::LeafGridView gv = grid->leafGridView(); //
>     actually create a grid view object here
>            
> 
>             // now loadbalance
>             grid->loadBalance();
> 
>             //
>             int numElemesOnThisProcess = 0;
>             for (auto eIt = gv.begin<0, Interior_Partition>(); eIt !=
>     gv.end<0, Interior_Partition>(); ++eIt)
>             {
>                 numElemesOnThisProcess++;
>             }
>             std::cout << "you have " << numElemesOnThisProcess << "
>     elements on process  " << helper.rank() <<std::endl;
>             const std::string baseOutName = "Grid_";
>             VTKWriter<GV> vtkWriter(gv);
>             std::vector<int> rankField(gv.size(0));
>             std::fill(rankField.begin(), rankField.end(),
>     grid->comm().rank());
>             vtkWriter.addCellData(rankField,"rank");
>             vtkWriter.write(baseOutName+std::to_string(0));
> 
>             return 0;
> 
>     The output from my process 0 xterm window is:
> 
>     (gdb) run
>     Starting program:
>     /usr/local/home/gcd3/codes/vem/dune-part/src/dune-part
>     [Thread debugging using libthread_db enabled]
>     Using host libthread_db library
>     "/lib/x86_64-linux-gnu/libthread_db.so.1".
>     Reading 2d Gmsh grid...
>     version 2.2 Gmsh file detected
>     file contains 81 nodes
>     file contains 64 elements
>     number of real vertices = 81
>     number of boundary elements = 0
>     number of elements = 64
> 
>     Created serial ALUGrid<2,2,cube,nonconforming>.
> 
>     you have 64 elements on process  0
>     [Inferior 1 (process 31987) exited normally]
> 
>     and that from process 1 window:
>     (gdb) run
>     Starting program:
>     /usr/local/home/gcd3/codes/vem/dune-part/src/dune-part
>     [Thread debugging using libthread_db enabled]
>     Using host libthread_db library
>     "/lib/x86_64-linux-gnu/libthread_db.so.1".
>     Reading 2d Gmsh grid...
>     version 2.2 Gmsh file detected
>     file contains 81 nodes
>     file contains 64 elements
>     number of real vertices = 81
>     number of boundary elements = 0
>     number of elements = 64
> 
>     Created serial ALUGrid<2,2,cube,nonconforming>.
> 
>     you have 64 elements on process  1
>     [Inferior 1 (process 31958) exited normally]
> 
>     Since both of them show 64 elements (same as original grid) I
>     thought I am doing something wrong.
> 
>     Thank you
>     Ganesh
> 
> 
>     On Wed, Sep 9, 2015 at 1:17 PM, Andreas Dedner
>     <a.s.dedner at warwick.ac.uk <mailto:a.s.dedner at warwick.ac.uk>> wrote:
> 
>         Hi.
>         How do you read in your grid? Please post the code you are using
>         to do that.
>         Andreas
> 
> 
>         On 09/09/15 13:07, Ganesh Diwan wrote:
>         > Hi Dune list
>         >
>         > I am not able to distribute the grid with loadbalance when I
>         use alugrid
>         > as gridtype.  All processes contain the same number of
>         elements as the
>         > original grid. I think alugrid is not correctly configured for
>         parallel
>         > mode. However uggrid seems to work as loadbalance splits the
>         grid and I
>         > can see the rank partitions in vtk output fine. I configured with
>         > following contents in my opts file:
>         >
>         > CONFIGURE_FLAGS="CXX=g++-4.8 \
>         > --enable-parallel \
>         > --enable-experimental-grid-extensions \
>         > --with-metis=/usr \
>         > --with-metis-lib=metis \
>         > --with-parmetis=/usr \
>         > --with-parmetis-lib=parmetis \
>         > --with-alberta='/home/gcd3/dune/ext_build/alberta' \
>         > --with-zoltan='/home/gcd3/dune/ext_build/Zoltan-v3.82' \
>         > --with-ug='/home/gcd3/dune/ext_build/ug' \
>         > --prefix='/home/gcd3/dune/core-2.4.0/install/' \
>         > "
>         >
>         > Do I need to include any other flags, perhaps adding
>         CXXFLAGS=mpicc? I
>         > was under impression that --enable-parallel would invoke the
>         parallel
>         > flags for grid, is it correct?
>         >
>         > Thanks in advance for help,
>         > Ganesh
>         >
>         >
>         >
>         >
>         > _______________________________________________
>         > Dune mailing list
>         > Dune at dune-project.org <mailto:Dune at dune-project.org>
>         > http://lists.dune-project.org/mailman/listinfo/dune
>         >
> 
>         _______________________________________________
>         Dune mailing list
>         Dune at dune-project.org <mailto:Dune at dune-project.org>
>         http://lists.dune-project.org/mailman/listinfo/dune
> 
> 
>