[petsc-users] Using direct solvers in parallel

[Anton Popov]

On 5/15/12 9:25 AM, Thomas Witkowski wrote:
> I made some comparisons of using umfpack, superlu, superlu_dist and 
> mumps to solve systems with sparse matrices arising from finite 
> element method. The size of the matrices range from around 50000 to 
> more than 3 million unknowns. I used 1, 2, 4, 8 and 16 nodes to make 
> the benchmark. Now, I wonder that in all cases the sequential umfpack 
> was the fastest one. So even with 16 cores, superlu_dist and mumps are 
> slower. Can anybody of you confirm this observation? Are there any 
> other parallel direct solvers around which are more efficient?
>
> Thomas
You should definitely try PARDISO. On shared memory machine is 
unbeatable by any kind of MPI-based solver, and offers functionality for 
both symmetric and non-symmetric matrices. There is a version available 
from MKL library.

If you're fine enough with symmetric matrices, then this thing - HSL 
MA87 - is even better than PARDISO. It's a commercial solver, but it's 
free for academic use.

The only "problem" I see here is that both of them are difficult to use 
together with PETSc, since they're OMP-based. But difficult probably 
doesn't mean impossible.

Cheers,

Anton