[dune-pdelab] Parallel implementation of MFEM

Bernd Flemisch bernd at iws.uni-stuttgart.de
Tue May 14 17:40:43 CEST 2013


Hi Jizhou,

that is great to hear!

Sorry for not responding to your issue when using UGGrid. To me, this 
sounds like you have not installed UG for parallel usage. Like for DUNE, 
you need to use the --enable-parallel option when configuring UG.

Kind regards
Bernd

On 05/14/2013 04:18 PM, Jizhou Li wrote:
> Hi Bernd,
> I am writing to thank you for pointing me to the right direction. I first tried UGGrid, but it still have communication problem. Then I switch to SPGrid and it worked.
>
> Thanks
>
> Jizhou
>
> On May 3, 2013, at 4:50 PM, Jizhou Li wrote:
>
>> Hi Bernd,
>> Thank you for responding so quickly. I think you are right that I am using the yaspgrid and MFEM does requires passing out the edge information.
>> So, the first thing I tried is to parallelize uggrid just on a example problem. This is what I did:
>>
>>         typedef Dune::UGGrid<2> GridType;
>>         GridType grid;
>>
>>         std::string grid_file = "cube.msh";
>>         Dune::GridFactory<GridType> factory(&grid);
>>         if(helper.rank()==0) {
>>           Dune::GmshReader<GridType>::read(factory,grid_file,true,false);
>>         }
>>         factory.createGrid();
>>         if(!Dune::MPIHelper::isFake)  grid.loadBalance();
>>         grid.globalRefine(level);
>>
>>         typedef GridType::LeafGridView GV;
>>         const GV& gv=grid.leafView();
>>
>>         example02_DG_parallel_nonoverlaping(gv);
>>
>> But, it doesn't seems like it was parallelizing the code. It seems to me it's just trying to execute the solver #n times.
>> Also, this is just a DG example which doesn't really need to passing the edge information at the time.
>> Do you think MFEM will just work once I witch the grid? For the solver for the MFEM, I didn't do anything fancy. It's almost exactly like the rt0main.cc except I change the RT0Constraint and the iterative solver.
>>
>> Thanks
>>
>> Jizhou
>>
>>
>>
>> On May 3, 2013, at 5:00 AM, dune-pdelab-request at dune-project.org wrote:
>>
>>> Send dune-pdelab mailing list submissions to
>>> 	dune-pdelab at dune-project.org
>>>
>>> To subscribe or unsubscribe via the World Wide Web, visit
>>> 	http://lists.dune-project.org/mailman/listinfo/dune-pdelab
>>> or, via email, send a message with subject or body 'help' to
>>> 	dune-pdelab-request at dune-project.org
>>>
>>> You can reach the person managing the list at
>>> 	dune-pdelab-owner at dune-project.org
>>>
>>> When replying, please edit your Subject line so it is more specific
>>> than "Re: Contents of dune-pdelab digest..."
>>>
>>>
>>> Today's Topics:
>>>
>>>   1. Parallel implementation of MFEM (Jizhou Li)
>>>   2. Re: Parallel implementation of MFEM (Bernd Flemisch)
>>>
>>>
>>> ----------------------------------------------------------------------
>>>
>>> Message: 1
>>> Date: Thu, 2 May 2013 15:13:26 -0500
>>> From: Jizhou Li <jl48 at rice.edu>
>>> To: dune-pdelab at dune-project.org
>>> Subject: [dune-pdelab] Parallel implementation of MFEM
>>> Message-ID: <DAE84CC4-212F-4AF4-9481-F114E43D333E at rice.edu>
>>> Content-Type: text/plain; charset=us-ascii
>>>
>>> Hi,
>>> Does anyone know how to implement the parallel version of mixed finite element method? I tried to get everything set up the same as the sample code with overlapping and the code compiles. But, I got the error: "interface communication not implemented" when running the code. I wonder if it was caused by using the compositegridfunctionspace, where the constraints on two finite element spaces for the mixed method are not the same.
>>>
>>> Thanks
>>>
>>> Jizhou Li
>>>
>>>
>>> ------------------------------
>>>
>>> Message: 2
>>> Date: Thu, 02 May 2013 22:28:18 +0200
>>> From: Bernd Flemisch <bernd at iws.uni-stuttgart.de>
>>> To: dune-pdelab at dune-project.org
>>> Subject: Re: [dune-pdelab] Parallel implementation of MFEM
>>> Message-ID: <web-168744896 at uni-stuttgart.de>
>>> Content-Type: text/plain;charset=iso-8859-1; format="flowed"
>>>
>>> Hi Jizhou,
>>>
>>> I am only guessing, but maybe you use YaspGrid which is
>>> only able to communicate on elements and vertices. If you
>>> have dofs on edges or faces like in MFEM, you have to use
>>> a grid manager that is able to communicate on those
>>> entities. That would be UGGrid or SPGrid,
>>> http://dune.mathematik.uni-freiburg.de/grids/dune-spgrid/,
>>> or, in 3d, ALUGrid.
>>>
>>> If that is not it, please provide a more detailed report,
>>> favorably with an example code that is stripped down as
>>> much as possible, so that people can help you more easily.
>>>
>>> Kind regards
>>> Bernd
>>>
>>> On Thu, 2 May 2013 15:13:26 -0500
>>> Jizhou Li <jl48 at rice.edu> wrote:
>>>> Hi,
>>>> Does anyone know how to implement the parallel version
>>>> of mixed finite element method? I tried to get everything
>>>> set up the same as the sample code with overlapping and
>>>> the code compiles. But, I got the error: "interface
>>>> communication not implemented" when running the code. I
>>>> wonder if it was caused by using the
>>>> compositegridfunctionspace, where the constraints on two
>>>> finite element spaces for the mixed method are not the
>>>> same.
>>>>
>>>> Thanks
>>>>
>>>> Jizhou Li
>>>> _______________________________________________
>>>> dune-pdelab mailing list
>>>> dune-pdelab at dune-project.org
>>>> http://lists.dune-project.org/mailman/listinfo/dune-pdelab
>>> ___________________________________________________________
>>>
>>> !!!! CMWR 2014: 9th - 13th June 2014 in Stuttgart !!!!
>>> 		Please visit www.cmwr14.de
>>> ___________________________________________________________
>>>
>>> Bernd Flemisch                     phone: +49 711 685
>>> 69162
>>> IWS, Universitaet Stuttgart          fax: +49 711 685
>>> 67020
>>> Pfaffenwaldring 61        email:
>>> bernd at iws.uni-stuttgart.de
>>> D-70569 Stuttgart        url:
>>> www.hydrosys.uni-stuttgart.de
>>> ___________________________________________________________
>>>
>>>
>>>
>>> ------------------------------
>>>
>>> _______________________________________________
>>> dune-pdelab mailing list
>>> dune-pdelab at dune-project.org
>>> http://lists.dune-project.org/mailman/listinfo/dune-pdelab
>>>
>>>
>>> End of dune-pdelab Digest, Vol 42, Issue 1
>>> ******************************************
>>>


-- 
______________________________________________________________

     !!!! CMWR 2014: 9th - 13th June 2014 in Stuttgart !!!!
                   Please visit www.cmwr14.de
______________________________________________________________

Bernd Flemisch                        phone: +49 711 685 69162
IWS, Universität Stuttgart            fax:   +49 711 685 60430
Pfaffenwaldring 61           email: bernd at iws.uni-stuttgart.de
D-70569 Stuttgart           url: www.hydrosys.uni-stuttgart.de
______________________________________________________________





More information about the dune-pdelab mailing list