Restart parallel computation with MPI.
With the current code, restart parallel computation with MPI can only be available for non-deformation processes. Moreover, the restart parallel computation needs two intermediate data processing steps for the solution data of the previous computation:
- using
RemoveGhostData
to merge the partition VTU files into one VTU file. - using
partmesh
to partition the mesh and solutions data contained in the merged vtu file.
Since the integration point data, such as stress, strain, plastic variables and so on are not handled by that intermediate processing of data, the existing restart parallel computation procedure is invalid for deformation related processes.
My idea is to use MPI parallel I/O functions to dump all raw result data of all partitions to a single file at the end of a parallel computation for a restart parallel computation.
Pros and cons of the suggested approach:
Pros:
- parallel I/O, partition to partition data transport, fast,
- no intermediate data processing.
Cons: The raw data file is binary, and it is platform dependent. If we keep in mind that the restart parallel computation should always be performed on the same platform, the binary file I/O is not an issue.
If all are agree with my proposal, I will start to implement it.