[petsc-dev] Question About Petsc ILU

[Dave Nystrom]

Jed Brown writes:
 > On Thu, Dec 29, 2011 at 19:30, Dave Nystrom
 > <Dave.Nystrom at tachyonlogic.com>wrote:
 > 
 > > Generally I use CG and LU from petsc.  Cholesky runs slower than LU and I
 > > was
 > > told in a previous email that it was because of the extra data movement for
 > > Cholesky versus LU.  I have also tried minres instead of CG and tend to get
 > > reduced iteration counts.  But I think minres may have an extra dot product
 > > and so is a little more expensive than cg.  But minres is usually cheaper
 > > over all because of the reduction in iterations.
 > >
 > 
 > If you are concerned about solve time for serial runs, I would either use
 > Cholesky from MUMPS or CHOLMOD (install suitesparse, then -pc_type cholesky
 > -pc_factor_mat_solver_type cholmod).

Thanks.  I am very interested in improving the direct solution time for my
Hall matrix problem for use in smaller runs.  My impression is that sparse
direct solves don't really scale to large problems but maybe it will be good
enough for this 2d code.

I will try suitesparse and cholmod.  I have been having problems with mumps
but will try it again and at least document the problem on petsc-maint.

So far, I have tried with success the following sparse direct solvers from
petsc: petsc, superlu, spooles, umfpack.  The best performer has been umfpack
and it gives a speedup in solving my Hall matrix of between 2x and 3x when
using sequential MKL for blas.  But I would definitely like to get a result
with mumps and cholmod.  Thanks for the suggestion.