[Dune] Performace of ISTL

[Patrick Leidenberger]

Hi all,

thanks a lot for the quick response.

> I do not know the Aztek solver, but a short glance at the user guide
> revealed that it uses GMRes to solver you system.
> 
> As you use the GradientSolver of Dune you are kind of comparing a
> sandwiches with three-course meals.
That's right, I am sorry. I make a better comparison with the same mesh 
like yesterday:
Trilinos with Jacobi and CG:  2 sec, 40 iteratrions residual: 1.05e-8
Dune     with Jacobi and CG: 30 sec, 40 iteratrions residual: 1.05e-8
And I don't forget a '0' for trilinos!

> Why don't you use other combinations of the solvers and
> preconditioners?
The original idea was to parallelize the istl solver, and the Jacobi 
seems to be the most easy one. But I have to say that it is not easy (a 
howto would be fine).

>> An other application for such an index will be, to write the solution
>> from different processes in the same hdf5 file.
>> 
> Does it really have to be consecutive for this task?
I think so, but I am not an hdf5 expert.

> On a locally refined and distributed grid (like ALU) it is very time
> consuming to construct a consecutive global index set. This cost
> is only reasonable if e.g. an expensive linear solver has to be used
> between grid refinement steps - or only globally refined grids are used.
I have only static, globally refined ALU grids at the moment.

> For a static grid (between the adaptation steps) you can easily
> compute global index. You communicate the local size and then you can
> compute a global index based on the process id, the sizes of all
> preceeding process and the local index.
That's a perfect idea.
I am interested in the edges in a ALU 3d simplex grid.
Is the border on one process unique? Is the border on process ghost on 
the other processes?

Have a nice day,

Patrick