I’m trying to revisit an old problem of artifacts at processor boundaries (simulation data source: openfoam). In our VTK handling we are using vtkMultiPieceDataSet, with one piece per MPI rank. Since each individual piece is a standalone representation, there doesn’t seem to be an obvious way of applying global point renumbering and have it do something sensible. Ghost cells are completely out of scope (way too complicated) and ghost points probably aren’t enough (and also feel like an odd fit). What alternatives are there for producing globally consistent data apart from bypassing the multi-piece representation and transferring everything to the master rank? I feel like I must be missing something rather obvious.
/mark