A netCDF classic or 64-bit offset file is stored in three parts:

1. The header, containing information about dimensions, attributes, and variables
2. The fixed-size data, containing data values for variables that don't have an unlimited dimension
3. The record data, containing data values for variables that have an unlimited dimension

By default, the header has almost no extra space; it is just large enough to contain the dimensions, variables, and attributes (including all the attribute values) rounded up to a whole number of disk blocks.

• Advantage: netCDF files are compact, requiring very little overhead
• Disadvantage: operations that require the header to grow (such as adding new dimensions or variables) requires moving all the data by copying it

To avoid copying data when the file schema changes

• Either create all necessary dimensions, variables, and attributes before writing data, or
• reserve extra space in the file header for later additions (using nc__enddef() in C, NF__ENDDEF() or NF90__ENDDEF() for Fortran, not available in Java interface)
• There is also an NCO program that adds extra space in a netCDF header, if you forget to do it when the file is created.
• NetCDF-4 files do not have a contiguous header for metadata, so none of this is necessary for netCDF-4.

2007 Unidata NetCDF Workshop for Developers and Data Providers > Performance