[Dune] AMG and performance

[Arne Rekdal]

Hello,

I have attached some timing results of the SAMG and  
Dune::CGSolver<Dune::AMG>. I turned on the verbose switch in SAMG, and  
I now see that the number of cycles seems to be constant wrt. the  
problem size, while number of iterations in CG increase with the  
problem size.

The timings is performed on an AMD Opteron 2216 with 32 GB memory.

Siterer Markus Blatt <Markus.Blatt at iwr.uni-heidelberg.de>:
> This information is a little bit too sparse. What is S (scalar or system,
> 2D/3D, problem type etc.)

S is a matrix coming from a 3D Mixed-Hybrid Element formulation of an  
anisotropic Poisson problem. S is the result of performing two  
Schur-complement reductions of the original system. The unknowns in  
the transformed system (Sr=b), is the pressure on the element's faces.

> So you probably have to choose, what you need more: less memory usage
> or faster computation. One cannot get the best of both worlds.

Since I have a more memory available, are there anything I can do to  
shorten the computation time? I don't want to use the SAMG solver :)


Best regards
Arne Rekdal
-------------- next part --------------
A non-text attachment was scrubbed...
Name: timings_dune.ods
Type: application/vnd.oasis.opendocument.spreadsheet
Size: 7708 bytes
Desc: not available
URL: <https://lists.dune-project.org/pipermail/dune/attachments/20090430/58233704/attachment.ods>