Having downsampled bulk files would be useful for quicklooks. For one, vtkHyperTreeGrid does allow such things, but that also requires downsampled data (generation of which would add more MPI communications during runtime).
Having an option to create new sidecar files with e.g. the highest refinement level downsampled (via averaging) back to the second-highest refinement level would help developing data analysis for science with faster turnover.
To be thought a bit more about - at face value, substituting CellIDs from high-refinement cells with lower-level CellIDs (merging 8->1) and collecting all the child cell data for averaging to the same position in the new array should be biggest change (plus updating metadata).
Having downsampled bulk files would be useful for quicklooks. For one, vtkHyperTreeGrid does allow such things, but that also requires downsampled data (generation of which would add more MPI communications during runtime).
Having an option to create new sidecar files with e.g. the highest refinement level downsampled (via averaging) back to the second-highest refinement level would help developing data analysis for science with faster turnover.
To be thought a bit more about - at face value, substituting CellIDs from high-refinement cells with lower-level CellIDs (merging 8->1) and collecting all the child cell data for averaging to the same position in the new array should be biggest change (plus updating metadata).