I have some questions regarding large memory in Paraview (5.6.0 version) when load xmf+h5 data.
The mesh is unstructured prism mesh, with 826718 nodes and 1586508 cells, which is not crazy big. Paraview can load the original vtk mesh file pretty fast, using 1GB memory. The output numerical results are written in xmf+h5 format, based on PETSc/HDF5 interface running in parallel. Mesh file is separated from others. There is no problem in running the code and generate the results. The parallel I/O performance also looks reasonable. The output mesh file with some other information include is around 900MB (basin-3d_domain.h5). One of the dataset is around 360MB (basin-3d_8.gsp). I have no problem loading these data in a computer with 32GB memory. However, the loading of xmf+h5 data is really slow and takes a lot of memory. During my test (client only), it takes up to 20GB memory when the file is loading. After the file is loaded, it still takes more than 6GB memory. If I load some other data around 1.5GB, it crashes my computer as the total memory is well above 32GB.
The related dataset can be download via the link below. Please note the z direction in basin-3d_domain.h5 is scaled compared to the original basin-3d.vtk file.
Is this normal? If not, is there anything wrong in my h5 data file or Paraview configuration?