ParaView/ParaView Readers and Parallel Data Distribution

From KitwarePublic
< ParaView
Revision as of 04:44, 8 March 2013 by Utkarsh (talk | contribs)
Jump to navigationJump to search
Under Development. Editors, try to keep this list sorted by the extension


File Extension Format Description Reader Notes on Data Parallelism Time Support
*.ex2,.exo Exodus Files ExodusIIReader

This file format supports parallel distribution of data by splitting data across many files. It also supports mesh adaptation by splitting files at simulation times where adaptation occurs. Hence ParaView does the following:

The root node scans the directory for files in the set and reads metadata (blocks and variables defined on them) from a single file in the set. It then broadcasts this information to all processes. Each reads a different subset of files.

Time steps may be contained in a file as long as the mesh topology does not change. Time may also be split across a file series.

*.vti XML VTI (Image) Files XMLImageDataReader

This file format does not support saving parallel distribution of data (refer to *.pvti file format for saving partitions). When loading the dataset in parallel, the structured extent of the dataset is distributed among all processes in a load-balanced way. All processes open the data file, however only the data corresponding to the subextent corresponding to the process are read in.

The file format support saving timesteps in the file, however, the more common mechanism is to use a file series.

*.vtk Legacy VTK Files LegacyVTKReader

This is the Legacy VTK file format, supporting binary or ASCII, for datasets of various types including structured and unstructured. This file format, however, has no support for saving parallel distribution of data. Hence ParaView does the following:


Structured Data


For structured data, this reader reads the entire file on all processes and then crops the structured extents on each process so that for each process, the filters downstream process a block of the structured data alone.


Unstructured Data


The root node reads the entire file and then the data is distributed to all processes (using MPI) after an internal, fairly naive, partitioning algorithm (vtkTransmitUnstructuredGridPiece or vtkTransmitPolyDataPiece).

Time supported only as a file series.