[Trilinos-Users] ML, IFPack and performance

[Duncan Karnitz]

I've been experimenting with Epetra, AztecOO, and ML pre-conditioners for solving a laplacian matrix for a fluid simulation. I've noticed in my profiling that Ifpack seems to perform very poorly when using Gauss-Seidel, compared to ML using AztecOO & Epetra methods for other smoothers.

In the documentation for ML 5.0 I also noticed this: "Ifpack implementation is automatically used (if available) for matrices of type Epetra CrsMatrix."

I will likely look to using the newer solver stack with Tpetra, Belos, and MueLu in the near future, but I was wondering if there are ways I can configure ML, etc to improve performance for this textbook problem.