On Thu, Sep 13, 2012 at 3:46 PM, John Peterson <[email protected]> wrote:

> On Thu, Sep 13, 2012 at 2:40 PM, Paul T. Bauman <[email protected]>
> wrote:
> > If you have superlu installed, I've found -pc_type bjacobi -sub_pc_type
> ilu
> > -sub_pc_mat_solver_package superlu to be much better than the standard
> ILU.
> > Note needs to be PETSc 3.3-p2 at least because there was a bug where
> > superlu only got used in the first iteration.
>
> Is superlu doing full (dense) LU or ILU?  If it's the former, that
> would explain why it's better...
>

I'm pretty certain it's ILU. This capability was added in the not too
distant past to PETSc. In the eyeball norm, the performance was comparable
(actually superlu seemed much faster where the work per processor was not
super small).


>
> Anyway, it's probably fine for small examples like this one to use
> -pc_type lu.  That way our examples don't require superlu/hypre just
> to run.
>

This is what I do usually, but with the external packages. The internal
PETSc lu doesn't handle pivoting.
------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
_______________________________________________
Libmesh-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-users

Reply via email to