<div dir="ltr"><div dir="ltr">In looking at this larger scale run ...<div><br></div><div>* Your eigen estimates are much lower than your tiny test problem.  But this is Stokes apparently and it should not work anyway. Maybe you have a small time step that adds a lot of mass that brings the eigen estimates down. And your min eigenvalue (not used) is positive. I would expect negative for Stokes ...</div><div><br></div><div>* You seem to be setting a threshold value of 0.1 -- that is very high</div><div><br></div><div>* v3.6 says "using nonzero initial guess" but this is not in v3.10. Maybe we just stopped printing that.</div><div><br></div><div>* There were some changes to coasening parameters in going from v3.6 but it does not look like your problem was effected. (The coarsening algo is non-deterministic by default and you can see small difference on different runs)</div><div><br></div><div>* We may have also added a "noisy" RHS for eigen estimates by default from v3.6.</div><div><br></div><div>* And for non-symetric problems you can try -pc_gamg_agg_nsmooths 0, but again GAMG is not built for Stokes anyway.</div><div><br></div></div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Tue, Mar 5, 2019 at 11:53 AM Myriam Peyrounette <<a href="mailto:myriam.peyrounette@idris.fr">myriam.peyrounette@idris.fr</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
  
    
  
  <div bgcolor="#FFFFFF">
    <p>I used PCView to display the size of the linear system in each
      level of the MG. You'll find the outputs attached to this mail
      (zip file) for both the default threshold value and a value of
      0.1, and for both 3.6 and 3.10 PETSc versions. <br>
    </p>
    <p>For convenience, I summarized the information in a graph, also
      attached (png file).</p>
    <p>As you can see, there are slight differences between the two
      versions but none is critical, in my opinion. Do you see anything
      suspicious in the outputs?</p>
    <p>+ I can't find the default threshold value. Do you know where I
      can find it?<br>
    </p>
    <p>Thanks for the follow-up</p>
    <p>Myriam<br>
    </p>
    <br>
    <div class="gmail-m_4553173887686987135moz-cite-prefix">Le 03/05/19 à 14:06, Matthew Knepley a
      écrit :<br>
    </div>
    <blockquote type="cite">
      
      <div dir="ltr">
        <div dir="ltr">On Tue, Mar 5, 2019 at 7:14 AM Myriam Peyrounette
          <<a href="mailto:myriam.peyrounette@idris.fr" target="_blank">myriam.peyrounette@idris.fr</a>>
          wrote:<br>
        </div>
        <div class="gmail_quote">
          <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
            <div bgcolor="#FFFFFF">
              <p>Hi Matt,</p>
              <p>I plotted the memory scalings using different threshold
                values. The two scalings are slightly translated (from
                -22 to -88 mB) but this gain is neglectable. The
                3.6-scaling keeps being robust while the 3.10-scaling
                deteriorates.</p>
              <p>Do you have any other suggestion?</p>
            </div>
          </blockquote>
          <div>Mark, what is the option she can give to output all the
            GAMG data?</div>
          <div><br>
          </div>
          <div>Also, run using -ksp_view. GAMG will report all the sizes
            of its grids, so it should be easy to see</div>
          <div>if the coarse grid sizes are increasing, and also what
            the effect of the threshold value is.</div>
          <div><br>
          </div>
          <div>  Thanks,</div>
          <div><br>
          </div>
          <div>     Matt <br>
          </div>
          <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
            <div bgcolor="#FFFFFF">
              <p>Thanks<br>
              </p>
              Myriam <br>
              <br>
              <div class="gmail-m_4553173887686987135gmail-m_-3242500023102749998moz-cite-prefix">Le
                03/02/19 à 02:27, Matthew Knepley a écrit :<br>
              </div>
              <blockquote type="cite">
                <div dir="ltr">
                  <div dir="ltr">
                    <div dir="ltr">On Fri, Mar 1, 2019 at 10:53 AM
                      Myriam Peyrounette via petsc-users <<a href="mailto:petsc-users@mcs.anl.gov" target="_blank">petsc-users@mcs.anl.gov</a>>
                      wrote:<br>
                    </div>
                    <div class="gmail_quote">
                      <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">Hi,<br>
                        <br>
                        I used to run my code with PETSc 3.6. Since I
                        upgraded the PETSc version<br>
                        to 3.10, this code has a bad memory scaling.<br>
                        <br>
                        To report this issue, I took the PETSc script
                        ex42.c and slightly<br>
                        modified it so that the KSP and PC
                        configurations are the same as in my<br>
                        code. In particular, I use a "personnalised"
                        multi-grid method. The<br>
                        modifications are indicated by the keyword
                        "TopBridge" in the attached<br>
                        scripts.<br>
                        <br>
                        To plot the memory (weak) scaling, I ran four
                        calculations for each<br>
                        script with increasing problem sizes and
                        computations cores:<br>
                        <br>
                        1. 100,000 elts on 4 cores<br>
                        2. 1 million elts on 40 cores<br>
                        3. 10 millions elts on 400 cores<br>
                        4. 100 millions elts on 4,000 cores<br>
                        <br>
                        The resulting graph is also attached. The
                        scaling using PETSc 3.10<br>
                        clearly deteriorates for large cases, while the
                        one using PETSc 3.6 is<br>
                        robust.<br>
                        <br>
                        After a few tests, I found that the scaling is
                        mostly sensitive to the<br>
                        use of the AMG method for the coarse grid (line
                        1780 in<br>
                        main_ex42_petsc36.cc). In particular, the
                        performance strongly<br>
                        deteriorates when commenting lines 1777 to 1790
                        (in main_ex42_petsc36.cc).<br>
                        <br>
                        Do you have any idea of what changed between
                        version 3.6 and version<br>
                        3.10 that may imply such degradation?<br>
                      </blockquote>
                      <div><br>
                      </div>
                      <div>I believe the default values for PCGAMG
                        changed between versions. It sounds like the
                        coarsening rate</div>
                      <div>is not great enough, so that these grids are
                        too large. This can be set using:</div>
                      <div><br>
                      </div>
                      <div>  <a href="https://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/PC/PCGAMGSetThreshold.html" target="_blank">https://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/PC/PCGAMGSetThreshold.html</a></div>
                      <div><br>
                      </div>
                      <div>There is some explanation of this effect on
                        that page. Let us know if setting this does not
                        correct the situation.</div>
                      <div><br>
                      </div>
                      <div>  Thanks,</div>
                      <div><br>
                      </div>
                      <div>     Matt</div>
                      <div> </div>
                      <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"> Let me know
                        if you need further information.<br>
                        <br>
                        Best,<br>
                        <br>
                        Myriam Peyrounette<br>
                        <br>
                        <br>
                        -- <br>
                        Myriam Peyrounette<br>
                        CNRS/IDRIS - HLST<br>
                        --<br>
                        <br>
                      </blockquote>
                    </div>
                    <br clear="all">
                    <div><br>
                    </div>
                    -- <br>
                    <div dir="ltr" class="gmail-m_4553173887686987135gmail-m_-3242500023102749998gmail_signature">
                      <div dir="ltr">
                        <div>
                          <div dir="ltr">
                            <div>
                              <div dir="ltr">
                                <div>What most experimenters take for
                                  granted before they begin their
                                  experiments is infinitely more
                                  interesting than any results to which
                                  their experiments lead.<br>
                                  -- Norbert Wiener</div>
                                <div><br>
                                </div>
                                <div><a href="http://www.cse.buffalo.edu/%7Eknepley/" target="_blank">https://www.cse.buffalo.edu/~knepley/</a><br>
                                </div>
                              </div>
                            </div>
                          </div>
                        </div>
                      </div>
                    </div>
                  </div>
                </div>
              </blockquote>
              <br>
              <pre class="gmail-m_4553173887686987135gmail-m_-3242500023102749998moz-signature" cols="72">-- 
Myriam Peyrounette
CNRS/IDRIS - HLST
--
</pre>
            </div>
          </blockquote>
        </div>
        <br clear="all">
        <div><br>
        </div>
        -- <br>
        <div dir="ltr" class="gmail-m_4553173887686987135gmail_signature">
          <div dir="ltr">
            <div>
              <div dir="ltr">
                <div>
                  <div dir="ltr">
                    <div>What most experimenters take for granted before
                      they begin their experiments is infinitely more
                      interesting than any results to which their
                      experiments lead.<br>
                      -- Norbert Wiener</div>
                    <div><br>
                    </div>
                    <div><a href="http://www.cse.buffalo.edu/%7Eknepley/" target="_blank">https://www.cse.buffalo.edu/~knepley/</a><br>
                    </div>
                  </div>
                </div>
              </div>
            </div>
          </div>
        </div>
      </div>
    </blockquote>
    <br>
    <pre class="gmail-m_4553173887686987135moz-signature" cols="72">-- 
Myriam Peyrounette
CNRS/IDRIS - HLST
--
</pre>
  </div>

</blockquote></div>