[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: NetCDF/HDF5 Proposal Approved

>To: address@hidden
>From: Éaddress@hidden
>Subject: Re: 20030919: NetCDF/HDF5 Proposal Approved
>Organization: CNRC
>Keywords: 200309191351.h8JDpWk1005725

Hi Éric,

> Do you know if the proposed netCDF-4 will still support the netCDFv2
> interface? And with the HDF5 lower layer, is random-access (direct
> disk-access) still going to work?

Yes, it is our intention to support the netCDF version 2 interfaces,
since these are just a layer on top of netCDF-3, which will be
supported. 

Direct access is still available in HDF5, and it may take either more
or fewer seeks to access the same data as with netCDF, depending on
how your the data is written.  As an example, when you declare an
unlimited dimension, the data is "chunked" or tiled, which may mean
fewer disk accesses to get various cross sections of data than with
the netCDF format.  So we already know that using HDF5 may result in
slower or faster I/O performance than netCDF, depending on
circumstances.  What we don't know yet is how close we can get to
current netCDF performance.

By the way, we also intend to try to support read and write access to
netCDF format data as well as netCDF-4 written HDF5 data through the
netCDF-4 interface.  And we are currently implementing a netCDF-3
interface on HDF5 which will include netCDFv2 interfaces, as a proof
of concept for our backward-compatibility approach ...

--Russ