[petsc-users] Tuning the parallel performance of a 3D FEM CFD code

[Barry Smith]

On May 15, 2011, at 8:59 AM, Jed Brown wrote:

> On Sun, May 15, 2011 at 15:23, Henning Sauerland <uerland at gmail.com> wrote:
> Unfortunately, it is not that easy to switch to BAIJ. I only have a local XFEM basis. That is, I have variable numbers of degrees of freedom throughout the domain (nodes of elements cut by the fluid interface have on additional "XFEM unknown"). At the moment, I don't know how or if I can handle the XFEM block when using BAIJ.
> 
> This is a problem, but you can still order unknowns in the vector so that velocity and pressure all come together, then AIJ/Inode will treat the components together which will tend to make ILU stronger.

    This may help a bit but I'm not sure how dramatically. What we could really use is a point block ILU for AIJ matrices with inodes. That is each inode set of rows is eliminated simultaneously (by factoring the little diagonal block associated with that inode).  This would possibly give the better convergence rate of point block ILU for this problem instead of point ILU. I don't think it is terribly difficult to code but it is a project and not something that could be whipped out in an hour.  Hmm, maybe one of our summer students?

    Barry