Hi,
This question is very broad-based and if I'm asking to compare apples to oranges, let me know, that is definitely possible.
In the User's Guide, for io_type, the options are (1) pnetcdf (2) pnetcdf,cdf5, (3) netcdf (I am using this currently) and (4) netcdf4. Is there an optimal io_type to employ as far as faster file read/write leading to faster model runtime? It definitely sounds like if I'm running in parallel (the model itself) and I am, then pnetcdf is optimal since by definition, with pnetcdf, one can read/write to netCDF files in parallel. Given that, is there a gain to using 'pnetcdf,cdf5'? I am not totally clear on the definition of 'large-variable files' (CDF-5). Does this pertain to the actual number of variables in your input/output files? If so, how many variables constitutes large-variable files then? Not sure if timings have been done with the various io_type's, but wanted to find out what your recommendations are.
This question is very broad-based and if I'm asking to compare apples to oranges, let me know, that is definitely possible.
In the User's Guide, for io_type, the options are (1) pnetcdf (2) pnetcdf,cdf5, (3) netcdf (I am using this currently) and (4) netcdf4. Is there an optimal io_type to employ as far as faster file read/write leading to faster model runtime? It definitely sounds like if I'm running in parallel (the model itself) and I am, then pnetcdf is optimal since by definition, with pnetcdf, one can read/write to netCDF files in parallel. Given that, is there a gain to using 'pnetcdf,cdf5'? I am not totally clear on the definition of 'large-variable files' (CDF-5). Does this pertain to the actual number of variables in your input/output files? If so, how many variables constitutes large-variable files then? Not sure if timings have been done with the various io_type's, but wanted to find out what your recommendations are.