Coordinate Systems in HDF5 presentation

• To: netcdf-hdf@xxxxxxxxxxxxxxxx
• Subject: Coordinate Systems in HDF5 presentation
• From: Quincey Koziol <koziol@xxxxxxxxxxxxx>
• Date: Wed, 31 Mar 2004 16:52:57 -0600 (CST)

Hi all,
    I've put my presentation about implementing coordinate systems in HDF5 up
on our FTP server at:
    ftp://hdf.ncsa.uiuc.edu/pub/outgoing/koziol/CoordSys.ppt

    One important thing to note about this presentation: it's got the actual
coordinate system extension embedded in the midst of some "infrastructure"
extensions for HDF5 (like the "Formula" data storage and shared dataspaces and
dimensions) that would be nice for various aspects of dealing with coordinate
systems in HDF5 files.  The infrastructure extensions are nice, but not
critical to the implementation of the coordinate system work in HDF5 (and are
especially not important to thinking about the data model of the new coordinate
system stuff).
    The core "objects" [classes, really] to think about for the coordinate
system extension are: Coordinate System, Axis, and Scale.

    Also, the slides about "Sharability in HDF5" in this presentation are mostly
intended for internal HDF5 library use and should be able to be skipped over
without much concern, until our telecon next week.

    Quincey

P.S. - Here's the list of issues which have been brought up, but I didn't
    change in the presentation yet:

        - Find a good description/picture to show "real space" & how it
            gets mapped to dataspace

        - Find a better way to describe the netCDF coord. system model (to
            better reflect how all the coordinate variables really compose
            a single coordinate system and each "normal" variable uses them to
            take a "slice" through the total space defined by the coordinate
            variables).

        - Possibly find a way to map sub-dataspaces onto another dataspace,
            including re-mapping indices and sub-setting, etc.  (to come more
            closely to the netCDF model of defining all the dimensions and
            choosing subset of them for each dataset)

        - Fix labels on Sierra use case to be different.

        - Allow dataspace rank to be greater than coordinate system rank and
            add use case (mapping a n-D coordinate to an offset on a space-
            filling curve, for example).

        - Change "Index Map" term to be "Index Array"

        - Possibly define a mechanism to "locate" one coordinate system
            within another coordinate system, in order to show that the
            elements for several dataspaces were located within a single,
            shared coordinate system.