<div dir="ltr">I see. Thanks for the info.</div><div class="gmail_extra"><br><br><div class="gmail_quote">On Mon, May 5, 2014 at 3:04 PM, Burlen Loring <span dir="ltr"><<a href="mailto:burlen.loring@gmail.com" target="_blank">burlen.loring@gmail.com</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">
    If you care at all about performance you're going to want to switch
    to appended unencoded. This is the fastest most efficient mode and
    generally gives smaller files than the others.<div><div class="h5"><br>
    <br>
    On 05/05/2014 02:41 PM, Mohammad Mirzadeh wrote:<br>
    <blockquote type="cite">
      <div dir="ltr">Thanks for the reference. What exactly is the
        benefit of having the data in appended mode? I guess right now
        I'm using  binary mode</div>
      <div class="gmail_extra"><br>
        <br>
        <div class="gmail_quote">On Mon, May 5, 2014 at 11:46 AM, Burlen
          Loring <span dir="ltr"><<a href="mailto:bloring@lbl.gov" target="_blank">bloring@lbl.gov</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"> Ahh, right, I assumed
              you were using VTK's classes to write the data, although
              if you're writing them yourself, you'll still want to
              emulate the "unencoded appended" format to get the best
              performance. see SetModeTo* and EncodeAppendedData* funcs
              (<a href="http://www.vtk.org/doc/nightly/html/classvtkXMLWriter.html" target="_blank">http://www.vtk.org/doc/nightly/html/classvtkXMLWriter.html</a>)<br>
              <br>
              There's also some essential info in the file format doc( <a href="http://www.vtk.org/VTK/img/file-formats.pdf" target="_blank">http://www.vtk.org/VTK/img/file-formats.pdf</a>).

              Search for "Appended".<br>
              <br>
              One way to get a handle on what's happening with the
              various modes and options is to examine the files you can
              produce in PV. For example open up PV, create a sphere
              source(or if you prefer some unstructured data), under
              file menu save the data and chose one of the pvt* options
              Compare the files produced for binary and appended modes
              etc...
              <div>
                <div><br>
                  <br>
                  <div>On 05/05/2014 11:34 AM, Mohammad Mirzadeh wrote:<br>
                  </div>
                  <blockquote type="cite">
                    <div dir="ltr">
                      <div class="gmail_extra">
                        <div class="gmail_quote">
                          <div>Burlen,</div>
                          <div><br>
                          </div>
                          <div>Thanks a lot for your comments. </div>
                          <div> </div>
                          <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
                            <div text="#000000" bgcolor="#FFFFFF"> This
                              is not an answer to your question but
                              there is a usage caveat w/ VTK XML files
                              that I want to make sure you're aware of. 
                              When you use that format make sure you set
                              mode to "appended" and "encode" off. This
                              is the combination to produce binary files
                              which are going to be faster and very
                              likely smaller too. You probably already
                              know that, but just in case ...<br>
                              <br>
                            </div>
                          </blockquote>
                          <div> </div>
                          <div>I write the data itself as binary inside
                            the .vtu file itself. Is this what you mean
                            by appended mode? I cannot see any 'mode'
                            keyword the xml file. Same as for encode; I
                            don't have it in the xml file.</div>
                          <div><br>
                          </div>
                          <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
                            <div text="#000000" bgcolor="#FFFFFF"> now
                              to get to your question:
                              <div><br>
                                <blockquote type="cite">1) Go with a
                                  single parallel HDF5 file that
                                  includes data for all time-steps. This
                                  makes it all nice and portable except
                                  there are two issues. i) It looks like
                                  doing MPI-IO might not be as efficient
                                  as separate POSIX IO, especially on
                                  large number of processors. ii)
                                  ParaView does not seem to be able to
                                  read HDF5 files in parallel</blockquote>
                              </div>
                              comment: If I were you I'd avoid putting
                              all time steps in a single file, or any
                              solution where files get too big. Once
                              files occupy more than ~80% of a tape
                              drive you'll have very hard time getting
                              them on and off archive systems. see this:
                              <a href="http://www.nersc.gov/users/data-and-file-systems/hpss/storing-and-retrieving-data/mistakes-to-avoid/" target="_blank">http://www.nersc.gov/users/data-and-file-systems/hpss/storing-and-retrieving-data/mistakes-to-avoid/</a>
                              My comment assumes that you actually use
                              such systems. But you probably will need
                              to if you generate large datasets at
                              common HPC centers.<br>
                              <br>
                            </div>
                          </blockquote>
                          <div><br>
                          </div>
                          <div>That's actually a very good point I was
                            not thinking of! Thanks for sharing.</div>
                          <div> </div>
                          <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
                            <div text="#000000" bgcolor="#FFFFFF"> I've
                              seen some AMR codes get elaborate in their
                              HDF5 formats and run into serious
                              performance issues as a result. So my
                              comment here is that if you go with HDF5,
                              keep the format as simple as possible! and
                              of course file sizes small enough to be
                              archived ;-)<span><font color="#888888"><br>
                                  <br>
                                  Burlen</font></span>
                              <div>
                                <div><br>
                                  <br>
                                  <br>
                                  <div>On 05/05/2014 10:48 AM, Mohammad
                                    Mirzadeh wrote:<br>
                                  </div>
                                  <blockquote type="cite">
                                    <div dir="ltr">They are represented
                                      as unstructured grid. As a sample
                                      run, a 100M grid point on 256 proc
                                      produces almost 8.5G file. We
                                      intent to push the limits close to
                                      1B at most at this time with #
                                      processors up to a few thousands.
                                      However, it would be good to have
                                      something that could scale to
                                      larger problems as well</div>
                                    <div class="gmail_extra"><br>
                                      <br>
                                      <div class="gmail_quote">On Sat,
                                        May 3, 2014 at 1:28 AM, Stephen
                                        Wornom <span dir="ltr"><<a href="mailto:stephen.wornom@inria.fr" target="_blank">stephen.wornom@inria.fr</a>></span>
                                        wrote:<br>
                                        <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Mohammad

                                          Mirzadeh wrote:<br>
                                          <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
                                            <div>
                                              <div> Hi I am at a
                                                critical point in
                                                deciding I/O format for
                                                my application. So far
                                                my conclusion is to use
                                                parallel HDF5 for
                                                restart files as they
                                                are quite flexible and
                                                portable across systems.<br>
                                                <br>
                                                When it comes to
                                                visualization, however,
                                                i'm not quite sure. Up
                                                until now I've been
                                                using pvtu along with
                                                vtu files and although
                                                they generally work
                                                fine, one easily gets in
                                                trouble when running big
                                                simulations on large
                                                number of processors as
                                                the number of files can
                                                easily get out of
                                                control and even
                                                simplest utility
                                                commands (e.g. ls) takes
                                                minutes to finish!<br>
                                                <br>
                                                After many thinking I've
                                                come to a point to
                                                decide between two
                                                strategies:<br>
                                                <br>
                                                1) Go with a single
                                                parallel HDF5 file that
                                                includes data for all
                                                time-steps. This makes
                                                it all nice and portable
                                                except there are two
                                                issues. i) It looks like
                                                doing MPI-IO might not
                                                be as efficient as
                                                separate POSIX IO,
                                                especially on large
                                                number of processors.
                                                ii) ParaView does not
                                                seem to be able to read
                                                HDF5 files in parallel<br>
                                                <br>
                                                2) Go with the same
                                                pvtu+vtu strategy except
                                                take precautions to
                                                avoid file explosions. I
                                                can think of two
                                                strategies here: i) use
                                                nested folders to
                                                separate vtu files from
                                                pvtu and also each time
                                                step ii) create an IO
                                                group communicator with
                                                much less processors
                                                that do the actual IO.<br>
                                                <br>
                                                My questions are 1) Is
                                                the second approach
                                                necessarily more
                                                efficient than MPI-IO
                                                used in HDF5? and 2) Is
                                                there any plan to
                                                support parallel IO for
                                                HDF5 files in paraview?<br>
                                                <br>
                                                <br>
                                              </div>
                                            </div>
                                            <div>
                                              _______________________________________________<br>
                                              Powered by <a href="http://www.kitware.com" target="_blank">www.kitware.com</a><br>
                                              <br>
                                              Visit other Kitware
                                              open-source projects at <a href="http://www.kitware.com/opensource/opensource.html" target="_blank">http://www.kitware.com/opensource/opensource.html</a><br>
                                              <br>
                                              Please keep messages
                                              on-topic and check the
                                              ParaView Wiki at: <a href="http://paraview.org/Wiki/ParaView" target="_blank">http://paraview.org/Wiki/ParaView</a><br>
                                              <br>
                                              Follow this link to
                                              subscribe/unsubscribe:<br>
                                              <a href="http://www.paraview.org/mailman/listinfo/paraview" target="_blank">http://www.paraview.org/mailman/listinfo/paraview</a><br>
                                            </div>
                                          </blockquote>
                                          Are your meshes structured or
                                          unstructured? How many
                                          vertices in your meshes?<span><font color="#888888"><br>
                                              <br>
                                              Stephen<br>
                                              <br>
                                              -- <br>
                                              <a href="mailto:stephen.wornom@inria.fr" target="_blank">stephen.wornom@inria.fr</a><br>
                                              2004 route des lucioles -
                                              BP93<br>
                                              Sophia Antipolis<br>
                                              06902 CEDEX<br>
                                                              <br>
                                              Tel: 04 92 38 50 54<br>
                                              Fax: 04 97 15 53 51<br>
                                              <br>
                                            </font></span></blockquote>
                                      </div>
                                      <br>
                                    </div>
                                    <br>
                                    <fieldset></fieldset>
                                    <br>
                                    <pre>_______________________________________________
Powered by <a href="http://www.kitware.com" target="_blank">www.kitware.com</a>

Visit other Kitware open-source projects at <a href="http://www.kitware.com/opensource/opensource.html" target="_blank">http://www.kitware.com/opensource/opensource.html</a>

Please keep messages on-topic and check the ParaView Wiki at: <a href="http://paraview.org/Wiki/ParaView" target="_blank">http://paraview.org/Wiki/ParaView</a>

Follow this link to subscribe/unsubscribe:
<a href="http://www.paraview.org/mailman/listinfo/paraview" target="_blank">http://www.paraview.org/mailman/listinfo/paraview</a>
</pre>
                                  </blockquote>
                                  <br>
                                </div>
                              </div>
                            </div>
                          </blockquote>
                        </div>
                        <br>
                      </div>
                    </div>
                    <br>
                    <fieldset></fieldset>
                    <br>
                    <pre>_______________________________________________
Powered by <a href="http://www.kitware.com" target="_blank">www.kitware.com</a>

Visit other Kitware open-source projects at <a href="http://www.kitware.com/opensource/opensource.html" target="_blank">http://www.kitware.com/opensource/opensource.html</a>

Please keep messages on-topic and check the ParaView Wiki at: <a href="http://paraview.org/Wiki/ParaView" target="_blank">http://paraview.org/Wiki/ParaView</a>

Follow this link to subscribe/unsubscribe:
<a href="http://www.paraview.org/mailman/listinfo/paraview" target="_blank">http://www.paraview.org/mailman/listinfo/paraview</a>
</pre>
                  </blockquote>
                  <br>
                </div>
              </div>
            </div>
          </blockquote>
        </div>
        <br>
      </div>
    </blockquote>
    <br>
  </div></div></div>

</blockquote></div><br></div>