[petsc-users] Petsc Performance Help

[Satish Balay]

On Fri, 16 Mar 2012, Jed Brown wrote:

> On Fri, Mar 16, 2012 at 14:52, <Nan.Jia at dartmouth.edu> wrote:
> 
> > Dear PETSc Group,
> > I am tuning the efficiency of my PETSc code for a while, but get very
> > little progress. So can anyone help me to analysis the log? Any suggestions
> > will be appreciated.
> >
> > My problem is time dependent. At every time step, two about 6000 by 6000
> > sparse matrices need to be solved, which come from a Poisson equation. I
> > use both sequential and parallel AIJ format to store matrices, but the
> > performances are both not very good.
> >
> 
> 1. You need to heed the huge warning
> 
>       ##########################################################
>       #                                                        #
>       #                          WARNING!!!                    #
>       #                                                        #
>       #   This code was compiled with a debugging option,      #
>       #   To get timing results run config/configure.py        #
>       #   using --with-debugging=no, the performance will      #
>       #   be generally two or three times faster.              #
>       #                                                        #
>       ##########################################################
> 
> 
> 
> 2. You only spend 18 of 880 seconds in the solver. What do you want?
> 
> 3. The problem is too small to get significant parallel speedup.
> 
> http://www.mcs.anl.gov/petsc/documentation/faq.html#slowerparallel

4. MatSetValues() is not logged by default - you should run with -info
and make sure there are no mallocs during assembly. Also check

http://www.mcs.anl.gov/petsc/documentation/faq.html#efficient-assembly

satish