By calling the collective I/O functions, you allow the MPI implementation to optimize the request based on the combined requests of all processes. The implementation can merge the requests of different processes and service the merged request efficiently. This is particularly effective when the accesses of different processes are noncontiguous and interleaved: