[Dune] Alternative perspective

[Oswald Benedikt]

Hi Dune, after following the discussion on this mailing list
for some time, I dare propose an alternative perspective.

While Dune is a great framework, extremely useful for solving
PDE's, and developed by an impressive team effort, I think it
most important to reflect on its mission from an application
developers point of view.

It has certainly been impressive, let's say some ten years
ago, to treat problems with tens of thousands or even hundreds
of thousands of elements. For example the (commercial) High Frequency
Structure Solver (HFSS, frequency domain, electrodynamics, finite elements)
by Ansoft Inc. could only solve problems with a few tens
of thousands of Tets some 4 years ago; clearly, a small
electromagnetic problem.

However, the challenges faced by folks today, solving PDE's with methods
such as FE, FV etc. have grown by 2 to 3 orders of magnitude;
to cite HFSS again, they now solve problems with 60 millions
of unknowns (2.8 MTet, 2nd order) and they strive to 100 millions in
the very near future.

Our own problems here at PSI are of almost the same size,
i.e. we are now working hard on writing codes that can
solve FE problems with tens of millions of tets, both in frequency
and time domain. We are looking forward to run such problems
on novel supercomputers at LBNL with probably tens of thousands
of CPU's next year.

Of course, I clearly see that Dune has its own agenda and objectives,
nevertheless it should keep an eye on its competition, also in the
commercial world.

I definitely prefer working with a Dune based code, instead
of reverting to a closed source commercial code.

But to solve my upcoming large problems, I need to be sure
that my framework scales as well, not only in the size of the
linear systems but also in supporting functions, such as
grid managment, large-scale parallel storage and retrieval,
evaluation of solutions on large grids etc.

To keep its relevance, in the arena of competition,
Dune should also, apart from architectural discussions,
show its existing and prospective user base that it is indeed
the tool to rely on for the huge tasks that are ahead.

While I myself need not be convinved on that, I already am,
this could still be very good for others.

With my very best regards and wish for a great day, Benedikt




------------------------------------------------------------------------------------------------------------------
Benedikt Oswald, Dr. sc. techn., dipl. El. Ing. ETH, www.psi.ch, Computational Accelerator Scientist
Paul Scherrer  Institute (PSI), CH-5232 Villigen, Suisse, benedikt.oswald at psi.ch, +41(0)56 310 32 12
quamquam sint sub aqua, sub aqua maledicere temptant. http://maxwell.psi.ch/amaswiki/index.php/User:BenediktOswald 
------------------------------------------------------------------------------------------------------------------