Skip to content

better handle distribution of partioned datasets in XML reader when reading in parallel

Utkarsh Ayachit requested to merge utkarsh.ayachit/vtk:xml-paritioning into master

Improve parallel load balancing when reading MBs

When reading a VTM file that is a multi-block dataset comprising of solely of multi-pieces, the vtkXMLMultiBlockDataReader can now distribute each multi-piece across ranks based on the PieceDistribution mode set on the reader.

If the multiblock has any non-multipiece, it reverts to the legacy mode of simply treating all leaf nodes as a collection to be distributed across all ranks.

In the following test image, the each group is a separate reader reading different files in different modes on 3 ranks, identified by the 3 colors..

  • top/first: a mixed multiblock, interleave (or round-robin) mode
  • second: a mixed multiblock, block (or contiguous) mode
  • third: a multiblock of 3 multipieces (each row of spheres is a multi-piece), interleave (or round-robin) mode
  • bottom/fourth: a multiblock of 3 multipieces (each row of spheres is a multi-piece), block (or contiguous) mode

Untitled_drawing__2_

Backport: paraview/release

Edited by Utkarsh Ayachit

Merge request reports