dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Mon, Sep 14, 2009 at 05:47:34AM -0500, Matthew Knepley wrote:
> On Mon, Sep 14, 2009 at 4:34 AM, Anders Logg <logg@xxxxxxxxx> wrote:
>
>     On Mon, Sep 14, 2009 at 11:19:58AM +0200, Garth N. Wells wrote:
>     >
>     >
>     > Anders Logg wrote:
>     > > On Mon, Sep 14, 2009 at 11:15:57AM +0200, Garth N. Wells wrote:
>     > >>
>     > >> Anders Logg wrote:
>     > >>> Which is the preferred parallel direct solver? MUMPS or SPOOLES?
>     > >>>
>     > >> Probably depends on the system, but in my experiences MUMPS has been
>     > >> better. However, the Debian/Ubuntu PETSc packages are configured with
>     > >> SPOOLES, but not MUMPS.
>     > >
>     > > ok, I'm trying out MUMPS now.
>     > >
>     > > Harish: Perhaps --download-mumps=yes should be added to petsc.package
>     > > in Dorsal?
>     >
>     > It requires a few more flags than just '--download-mumps=yes' as you'll
>     > soon see ;).
>
>     Yes...
>
>     So far, I've added
>
>      --with-parmetis=1 --with-parmetis-dir=/usr/include/parmetis
>     --download-mumps=1
>
>     But PETSc is complaining about ScaLAPACK. Is it possible to use the
>     Debian package? I'm using OpenMPI (from Debian) but
>     libscalapack-mpi-dev seems to require mpich.
>
>     Which is the simplest way to get MUMPS running?
>
>
> --download-scalalpack --download-blacs --download-mumps
>
> It is really depressing that MUMPS is built on such as crappy package. However,
> it is the
> best parallel direct solver, and SPOOLES is not actively supported anymore.
>
> Also, you can only use the packages if you buy into whatever compiler/flag/
> package combination
> they happened to use. One reason we need the "crazy download stuff".
>
>    Matt

Thanks, I'll try that.

I'm very happy PETSc handles this (and Hypre) for me. In an ideal
world, there would be decent Debian packages for all this.

--
Anders

Attachment: signature.asc
Description: Digital signature