Problem it solves
Currently the IB state is written in src/simulation/m_data_output.fpp by having rank 0 (only one MPI rank) write the IB state serially. While this works, it is very inefficient as all MPI ranks must wait on rank to finish a serial operation.
Proposed solution
The rest of MFC makes use of the MPI file writing primitives, which allow each rank to write a chunk of the data output to a subset of the output file in a distributed way. This would greatly improve out file IO time and make it consistent with the rest of MFC.
Alternatives
No other alternatives should be considered.
Additional context
One should review the rest of the file IO in the same data_output file.
Problem it solves
Currently the IB state is written in
src/simulation/m_data_output.fppby having rank 0 (only one MPI rank) write the IB state serially. While this works, it is very inefficient as all MPI ranks must wait on rank to finish a serial operation.Proposed solution
The rest of MFC makes use of the MPI file writing primitives, which allow each rank to write a chunk of the data output to a subset of the output file in a distributed way. This would greatly improve out file IO time and make it consistent with the rest of MFC.
Alternatives
No other alternatives should be considered.
Additional context
One should review the rest of the file IO in the same data_output file.