[petsc-users] The "DMComplexDistribute" is time-consuming

[Jed Brown]

      ##########################################################
      #                                                        #
      #                          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.              #
      #                                                        #
      ##########################################################


 Matt, why don't the mesh operations have events? I should ask myself the
same question about PetscSF (and not forget to write the two-sided
implementation).

On Sat, Aug 18, 2012 at 12:22 PM, fdkong <fd.kong at siat.ac.cn> wrote:

> Hi  Jed Brown,
>
> >> Hi developers,
> >>
> >> I use DMComplex represent the unstructured mesh. Everything is OK, but
> the
> >> function "DMComplexDistribute" is always time-consuming. I guess that it
> >> results from using the PetscSF. I don't know why we use this
> communication
> >> model. Could you please make some tests on the function?  Or could we
> use
> >> other communication functions ?
> >>
>
> >1. Always send -log_summary
>
> >2. Distributing a mesh is a fairly expensive operation.
>
> >3. DMComplex is new and we are more interested in getting the right
> capability than in micro-tuning performance. Mesh distribution is usually
> not done frequently.
>
> ------------------
> Fande Kong
> ShenZhen Institutes of Advanced Technology
> Chinese Academy of Sciences
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20120818/c32a09b0/attachment.html>