dolfin team mailing list archive

["Johan Hoffman" <jhoffman@xxxxxxxxxx>]

> On Mon, Feb 18, 2008 at 04:35:54PM +0100, Johan Hoffman wrote:
>> > This is good juncture at which re-focus on important goals for a
>> hopefully not too distant next release. Three priorities that I see are:
>> >
>> > - New build system.
>> >
>> > - Improved support for parallel assembly and solve. The class Function
>> will probably need the most work. Good goals would be to get Poisson and
>> elasticity demos running in parallel.
>>
>> I thought the Poisson problem was running, and showing good parallel
>> scaling? Or maybe that was a test outside the dolfin-dev demos?
>
> The demo is in src/demo/fem/assembly/.

Ok, so the matrix assembly is working and is implemented as a demo, but
solve is missing.

>> We (Nicklas) have now a preliminary distributed parallel implementation
>> running for Poisson and a Unicorn flow solver. The structure is still
>> quite messy and the partitioning is based on Metis, but we expect it to
>> be
>> integrated with DOLFIN in the coming months, hopefully before the
>> summer.
>
> Could this be done in small pieces? I'm a little worried about seeing
> big changesets that add things to the mesh classes.

Yes, we need to figure out how this should be done to best fit with the
existing mesh classes. When we have cleaned up the code we will take that
discussion here at the dolfin mailing list so that we can agree on a
solution.

>> When the structure is more clear we can discuss on the mailing list how
>> to
>> best incorporate this into DOLFIN. Apart from the structure of the
>> implementation we are also focusing on distributed local mesh
>> refinement.
>
> Great.
>
> What is the difference from what we have now? Is it the same but with
> a distributed mesh (so each processor just sees the local part, not
> the whole mesh)? Or is it something different?

I think the main issues are to handle refinement over
partition-boundaries, and efficient re-partitioning after refinement (for
load-balancing).

> I think what we have now can be fairly easily extended to a
> distributed mesh without any (significant) changes to the mesh
> classes. The key point is to build the DofMap in parallel. Take a look
> here, in particular Section 6 and Algorithm 5:
>
>   http://simula.no/research/scientific/publications/Simula.SC.75
>

Seems I need a username/password for this.

/Johan

>> > - Testing of QuadratureFunction and adding corresponding demos.
>> >
>> > Another point is to add Trilinos as a supported linear algebra back
>> end.
>> How does this tie up with closer links to PyCC?
>> >
>> > Garth
>>
>> - Something else that I think is important is to include geometry in
>> DOLFIN. We were to start such a project before Christmas, but we lost
>> the
>> msc student that was supposed to work on this. I would expect that we
>> restart this later this spring. Basically we need to import CAD-geometry
>> (or other geometry descriptions), create an internal DOLFIN format
>> (typically based on NURBS or similar) with a Geometry-class. Typical
>> functionality needed is then to project a point onto the surface
>> geometry.
>
> That would be nice.
>
> --
> Anders
>
>
>> /Johan
>>
>>
>> > _______________________________________________
>> > DOLFIN-dev mailing list
>> > DOLFIN-dev@xxxxxxxxxx
>> > http://www.fenics.org/mailman/listinfo/dolfin-dev
>> >
>>
>>
>>
>>
>> _______________________________________________
>> DOLFIN-dev mailing list
>> DOLFIN-dev@xxxxxxxxxx
>> http://www.fenics.org/mailman/listinfo/dolfin-dev
> _______________________________________________
> DOLFIN-dev mailing list
> DOLFIN-dev@xxxxxxxxxx
> http://www.fenics.org/mailman/listinfo/dolfin-dev
>