DAP 2 Operational Experience

We have installed an OpeNDAP netCDF server, and use this routinely to share
moored time series data and large 4D model output data sets with our
collaborators (and with the general public). We have also converted several of
our previous netCDF clients into OPeNDAP clients by relinking with the OpenDAP
netcdf libraries. Relinking the Matalb/NetCDF interface has proved hugely
successful, as we can now use all of our previously developed Matlab code that
accessed NetCDF to work with OpenDAP data. Thus we can use all our old existing
software to work with OpenDAP data instead of learning and installing new
software (e.g. The Matlab/OpenDAP gui, command line tools, etc). We have used
DODS/OpenDAP for 5 years.
We tried using JGOFS and simple FTP/Web data distribution before DODS/OpenDAP.
One of the big benefits of OpenDAP has been how easy it is for the data provider
to serve data. As long as you have a UNIX web server, you just download the
OpenDAP binaries , plunk them in the cgi-bin, and then drop a bunch of netcdf
files onto your web server and you are serving data via OpenDAP. The whole
thing literally takes about 2 hours max. Developing clients that do
something useful with the OpenDAP data you've supplied of course takes longer.
This can be eased by making the NetCDF files conform to certain standards,
such as the CF conventions for NetCDF. In this way, clients can be developed
that access a number of different types of NetCDF data in a consistent manner,
for example, a single toolkit can access many different types of numerical model
output via OpenDAP without knowing what type of model generated the information,
only that the data is CF-compliant. The Integrated Data Viewer from Unidata
is an example of a OpenDAP client that takes advantage of CF-compliant OpenDAP
-served filed.
I don't know the answers to how much bandwidth and data are downloaded , or the
other questions. I would be happy to chat about any of these issues.