On 08/29/2018 09:55 PM, Pai Liu wrote:
I learn form Prof. Wolfgang's lectures that parallel direct solver is
really competitive, when compared with the parallel distributed solver.
So I try to modify the solver in step-17 into a parallel direct solver,
and I found it *extremely solw*.
Direct solvers are only really competitive for relatively small problems
-- my rule of thumb is "less than 100,000 unknowns" in 2d. I don't know
how large your problem is, but your 40x40x40 mesh should have about
200,000 unknowns, so I'm not terribly surprised that the direct solver
is slower than the CG solver. Furthermore, 3d problems make life much
harder for direct solvers than 2d problems.
I am surprised that it is *that* much slower. My reference point is
UMFPACK, so it is possible that MUMPS is significantly slower than
UMFPACK. But, MUMPS can at least in principle run in parallel, so you
could use MPI to make things faster.
Interesting experiments would be:
* Try what happens if you run this in parallel
* Try how the run time of both the direct and iterative solvers
change as you increase the number of unknowns. (E.g., start with a
10x10x10 mesh, then try a 20x20x20, ... mesh.)
Best
W.
--
------------------------------------------------------------------------
Wolfgang Bangerth email: bange...@colostate.edu
www: http://www.math.colostate.edu/~bangerth/
--
The deal.II project is located at http://www.dealii.org/
For mailing list/forum options, see
https://groups.google.com/d/forum/dealii?hl=en
---
You received this message because you are subscribed to the Google Groups "deal.II User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to dealii+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
