[Dune] AMG and performance

Arne Rekdal arne at stud.ntnu.no
Thu Apr 30 16:52:45 CEST 2009


> BTW: I was really surprised that our hierarchy building is slower than
> SAMG. Now I see that you are using ILU0
>
> Is that on purpose (or just because I did this in the tests to confuse
> our users:)? Have you tries SeqSSOR instead?

I have tried both SeqSSOR and ILU0, and found the ILU0 to be faster  
with the relaxation factor set to 0.8.

criterion.setDefaultValuesAnisotropic(3,2) seem not to make any impact  
of the solution time.

If you are interested you can download my program from
http://folk.ntnu.no/arne/upscaling_cube.tar.gz

This is a complete Dune module, and the syntax for my code is
./upscaling_cube eclipsefile [ writeOutput relaxfactor ]

Example: ./upscaling_cube ../grids/fine.grdecl 0 0.8

I have included two eclipsefiles in the folder 'grids'.


Thanks for looking into this Markus!


Best regards
Arne Rekdal





More information about the Dune mailing list