<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    <p>Hello,<br>
    </p>
    <p>thanks again for your efforts. <br>
    </p>
    <p>
      <blockquote type="cite"><span
style="font-size:small;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">Boundary
          processors have less nonzeros than interior processors.</span></blockquote>
      Doesn't that mean that boundary processors have less work to do
      than the others? Or does this affect the size of the coarse grids?</p>
    <p>I know that defining the outermost nodes as boundary is not the
      most efficient way in this particular case (using Dirichlet
      boundary conditions on a smaller grid would do the same), but I
      need the solver to be able to handle arbitrarily shaped boundaries
      inside the domain, e.g. to calculate the potential inside a
      spherical capacitor (constant potential on the boundaries, charge
      distribution inside). Is there a better way to do that?</p>
    <p>Michael<br>
    </p>
    <p><br>
    </p>
    <p><br>
    </p>
    <br>
    <div class="moz-cite-prefix">Am 12.06.2018 um 22:07 schrieb Junchao
      Zhang:<br>
    </div>
    <blockquote type="cite"
cite="mid:CA+MQGp8Ty0sho_qzVsa30WSabqd6cLtUOp_b3-kG70Q=m7SOFg@mail.gmail.com">
      <div dir="ltr">Hello, Michael, 
        <div>  Sorry for the delay. I am actively doing experiments with
          your example code. I tested it on a cluster with 36
          cores/node. To distribute MPI ranks evenly among nodes, I used
          216  and 1728 ranks instead of 125, 1000.  So far I have these
          findings:<br>
        </div>
        <div> 1) It is not a strict weak scaling test since with <span
style="font-size:small;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">1728
            ranks it needs more KPS iterations, and more calls to MatSOR
            etc functions.</span></div>
        <div><span
style="font-size:small;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline"> 2)
            If I use half cores per node but double the nodes (keep MPI
            ranks the same), the performance is 60~70% better. It
            implies memory bandwidth plays an important role in
            performance.</span></div>
        <div><span
style="font-size:small;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline"> 3)
            I find you define the outermost two layers of nodes of the
            grid as boundary. Boundary processors have less nonzeros
            than interior processors. It is a source of load imbalance.
            At coarser grids, it gets worse. But I need to confirm this
            caused the poor scaling and big vecscatter delays in the
            experiment. </span></div>
        <div><span
style="font-size:small;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline"><br>
          </span></div>
        <div><span
style="font-size:small;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline"> Thanks.</span></div>
        <div><span
style="font-size:small;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline"><br>
          </span></div>
      </div>
      <div class="gmail_extra"><br clear="all">
        <div>
          <div class="gmail_signature" data-smartmail="gmail_signature">
            <div dir="ltr">--Junchao Zhang</div>
          </div>
        </div>
        <br>
        <div class="gmail_quote">On Tue, Jun 12, 2018 at 12:42 AM,
          Michael Becker <span dir="ltr"><<a
              href="mailto:michael.becker@physik.uni-giessen.de"
              target="_blank" moz-do-not-send="true">michael.becker@physik.uni-giessen.de</a>></span>
          wrote:<br>
          <blockquote class="gmail_quote" style="margin:0 0 0
            .8ex;border-left:1px #ccc solid;padding-left:1ex">
            <div text="#000000" bgcolor="#FFFFFF">
              <p>Hello,</p>
              <p>any new insights yet?</p>
              <span class="HOEnZb"><font color="#888888">
                  <p>Michael<br>
                  </p>
                </font></span><span class=""> <br>
                <br>
                <br>
                <div class="m_4694688645138750811moz-cite-prefix">Am
                  04.06.2018 um 21:56 schrieb Junchao Zhang:<br>
                </div>
                <blockquote type="cite">
                  <div dir="ltr">Miachael,  I can compile and run you
                    test.  I am now profiling it. Thanks.</div>
                  <div class="gmail_extra"><br clear="all">
                    <div>
                      <div class="m_4694688645138750811gmail_signature"
                        data-smartmail="gmail_signature">
                        <div dir="ltr">--Junchao Zhang</div>
                      </div>
                    </div>
                    <br>
                  </div>
                </blockquote>
                <br>
              </span></div>
          </blockquote>
        </div>
        <br>
      </div>
    </blockquote>
    <br>
  </body>
</html>