ParaView

https://github.com/hpjeonGIT/training_slides

Introduction to ParaView

• Visualization program
• Developed by Kitware
• OpenSource
• Competing with VisIt (developed by LLNL)
• For POST-processing
  • Not for pre-processing: Not mesher!
• Multiple processor support
  • mpiexec is included in binary distribution
  • But not for multiple nodes
• Batch job capability
  • pvbatch/pvpython

Feature

• All kinds of plotting
  • 2D bar/line/histogram chart
  • 2D/3D Mesh
    • Mesh data from commercial/open source FEM/FVM
  • Particle Data
    • Not rendered as sphere image
• User customization
  • Paraview must be built from source package
• Can generation movie/animation

VTK file format

• Standard files for ParaView
• Supported in many other visualization programs as well
• ASCII or binary
• Can define particles, meshes of 2D/3D
• VTM
  • Multi-block version of VTK
    • This is different from legacy VTK
    • XML format
    • Actual data are stored in the linked VTU or other partitioned files
  • VTU
    • XML VTK file format for unstructured grid
• Examples
  • Read some data (xmf or CAS) then save as vtm in Paraview using ASCII format

$ more ccc.vtm
<VTKFile type="vtkMultiBlockDataSet" version="1.0" byte_order="LittleEndian" header_type="UInt64">
  <vtkMultiBlockDataSet>
    <Piece index="0" name="Grid_2">
      <DataSet index="0" file="ccc/ccc_0_0.vtu">
      </DataSet>
      <DataSet index="1" file="ccc/ccc_0_1.vtu">
      </DataSet>
    </Piece>
  </vtkMultiBlockDataSet>
</VTKFile>

$ more ccc/ccc_0_0.vtu
<VTKFile type="UnstructuredGrid" version="2.0" byte_order="LittleEndian" header_type="UInt64">
  <UnstructuredGrid>
    <Piece NumberOfPoints="5000" NumberOfCells="5000">
      <PointData Scalars="dmg" GlobalIds="___D3___GlobalNodeIds">
        <DataArray type="Float64" Name="dmg" format="ascii" RangeMin="0" RangeMax="0">
          0 0 0 0 0 0
          0 0
        </DataArray>
        <DataArray type="UInt8" Name="vtkGhostType" format="ascii" RangeMin="0" RangeMax="0">
…

Multi-core visualization

• Settings -> General -> Multi-core support -> Adjust core numbers
• Close paraview and restart
• Load file. Then go to Filters -> D3
• Some file formats might not be supported

Tracing

• Similar to VB in Excel or journaling in FEM packages
• Edit the script and rerun as pvbatch --use-offscreen-rendering batch.py
  • pvbatch is a customized python by ParaView, enclosed in the pre-built package
  • 4.0 has a bug and re-locate “WriteImage()” function to the bottom

Batch/distributed visualization

• Not recommended but use it at your own risk
  • VisIt would be recommended for batch distributed visualization
• Recompilation of source might be necessary
• Add server/connect allows only one node connection
• For multiple nodes connection, write a pvsc file and “load servers”
  • Submit a batch script running mpirun -np \$SLURM_NPROCS pvserver
  • Prepare .pvsc file and add the name of computing nodes. Then “load servers”

<Servers>
<Server name="case1" resource="cs://server1.aaa.bbb//server2.ccc.ddd">
    <ManualStartup />
 </Server>
</Servers>

• Open a paraview from a separate terminal. When paraview is terminated or disconnects the computing node, the batch job will be terminated as well

Running mpi from GUI

• Edit-> Setting -> adjust mpi rank number. Restart paraview
• Now paraview will use multiple ranks
• However, loading large scale models may trigger the violation of 32bit integer limit
  • Visualizing small models might be beneficial
  • For large scale models, use pvserver on localhost
  • From a separate terminal, run mpirun -np 8 pvserver
  • On Paraview GUI, connect server -> localhost
    • Now large scale model can be visualized successfully