This routine is conceptually similar to the serial case except that all writes are now done in parallel. It is described by Algorithm 2.
In contrast to the serial output case, here, the renumbering offset is calculated before any data is written. Once it is calculated, an MPI_Barrier call is issued to ensure synchronicity between processes prior to calling nc_put_vara which instructs processes to write (in parallel) to the output file.
Unlike the input files, parallel output files cannot be compressed at write time using netCDF, which is a limitation of the library. It is suggested to user that they make use of the netCDF operators to perform compression. See Section 5.3 for usage details.