Read single Exodus II file in parallel
In order to read data in parallel with the current Exodus II reader, the data has to be broken up into multiple files so that each file is read by exactly one process. This was predicated on the assumption that parallel simulations would write out one file per process. The consequence was that if the files were joined into one, ParaView could only read that into a single process. The solution so far has been to train users to leave Exodus files broken up.
However, things are changing in the IOSS libraries used by Sandia simulations to write data. The parallel IO library is internally joining the data and writing, in parallel, to a single file. This actually allows the simulation to write out faster than if every process writes to separate files. However, it causes a problem for ParaView reading the data back in.
We need to modify the Exodus reader to be able to read data from a single Exodus file in parallel. In some brief conversations with Greg Sjaardema, we are told that the existing Exodus library is capable of reading parts of the file in parallel so (presumably) it is just a matter of choosing which process reads what.