Hi Ben:
On Sep 24, 2008, at 6:37 AM, Ben Domenico wrote:
But my impression is that others think that we should abandon the CF-
netCDF encoding spec. and ONLY be proposing CF-OPeNDAP. Is that
the heart of the suggestion that's on the table in terms of the
OPeNDAP part of the discussion?
-- Ben
This is an interesting question, and it is related to the larger
question of whether it is better to transfer files as mime-types or
better to have a binary protocol to transfer the information, in
particular if we envision very large amounts of data being
transferred. I know the OOI people are very concerned about this
issue, in fact are pushing the idea of a modified OPeNDAP that uses
the Advanced Messaging Queueing Protocol (AMQP see http://jira.amqp.org/confluence/display/AMQP/Advanced+Message+Queuing+Protocol)
as what gets put over the wire, a messaging protocol that is in
heavy use in the financial industry to transfer real-time their
transaction data (maybe we can get a $700 million bailout also!).
Seriously, one downside of netcdf delivered as a mime-type is that you
can't stream the service - you have to hold the file in cache until it
is complete and then send the result. OpeNDAP can be streamed, though
not many OpeNDAP servers do that. John Caron implied that at the GO-
ESSP meeting that he had code that could stream netcdf-3, but that it
would be unlikely that netcdf4 (or hdf5) could be streamed.
Interesting discussion.
-Roy
