[petsc-users] Slow speed when using PETSc multigrid

[Jed Brown]

On Thu, Jun 7, 2012 at 4:56 PM, TAY wee-beng <zonexo at gmail.com> wrote:

> I looked at the manual but I'm not sure how to get better performance.
>
> *      ##########################################################*
*      #                                                        #*
*      #                          WARNING!!!                    #*
*      #                                                        #*
*      #   This code was compiled with a debugging option,      #*
*      #   To get timing results run ./configure                #*
*      #   using --with-debugging=no, the performance will      #*
*      #   be generally two or three times faster.              #*
*      #                                                        #*
*      ##########################################################*

I wonder if making this text bigger, red, bold, and blinking (which I can't
seem to do in this email) would increase the chances that you read it.


>  Also, what are the more common options to start with. Is there an appropriate C example? Some options are:
> *PCMGSetLevels*  - how many lvls are appropriate?
>
> PCMGSetCycleType -
>
> PCMGSetNumberSmoothUp/down etc
>
> 1. Control this stuff using command line options. Recompiling to run a
different method was obsolete decades ago.

2. You need to understand a little about the math and a little about the
performance of each component in isolation. I recommend seeing how time
moves between different operations as you modify the algorithm.

3. Experiment with your machine and the methods. Learn which configurations
have good algorithmic performance (few iterations) and which execute most
efficiently.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20120607/463034ed/attachment.html>