NC_FILL (no plans to implement)

[Jianwei Li]

I think it's also a performance issue.
There may be some overhead using NC_FILL, depending on the density of
defined-value data in the file.

If pre-fill is performed when allocating the variables file space
and data is to be written later, NC_FILL is an obvious waist of
resources and I/O bandwidth.
Otherwise, it would be complecated for a multi-process program to
achieve this NC_FILL feature:)

Maybe we can implement it in the first approach as an optional feature
so that users have a chance if they really need to use it? Or we
can just go without it?

Jianwei

On Tue, 22 Jul 2003, Russ Rew wrote:

> Hi,
>
> >From the announcement to netcdfgroup, the "features not yet
> implemented" includes
>
>  - NC_FILL (no plan to implement)
>
> Is this because preserving fill value semantics makes it harder to do
> parallel I/O?  HDF5 supports fill-values for parallel I/O using
> MPI-IO, so I'm not sure why it's explicitly ruled out, or whether we
> should consider cutting it from our proposed feature list for the
> netCDF-4 on HDF5 project.
>
> --Russ
>
> _____________________________________________________________________
>
> Russ Rew                                         UCAR Unidata Program
> russ at unidata.ucar.edu                      http://my.unidata.ucar.edu
>