The colloid parallel i/o requires updating in line with the MPI/IO for lattice quantities.
Background
In the existing method (colloids_io.c) the write is serialised at root; a parallel method is to be preferred.
On the read side, it is still necessary for each process to read the whole file in general. There is no relationship between position in the file and ownership in any particular decomposition.
What's needed
We need to retain the same data format at least. Appropriate abstraction is required to admit the addition of different formats (e.g., hdf5).
We need an MPI/IO implementation for collective write, but a read similar to the existing serialised read will be necessary. The record format is ascii or native binary as before.
Summary
The colloid parallel i/o requires updating in line with the MPI/IO for lattice quantities.
Background
In the existing method (
colloids_io.c) the write is serialised at root; a parallel method is to be preferred.On the read side, it is still necessary for each process to read the whole file in general. There is no relationship between position in the file and ownership in any particular decomposition.
What's needed