Purpose: outline a short getting started guide for new CDF authors.

Audience: people getting started with creating CDF files who wish to contribute ideas.

See also: http://autoplot.org/developer.cdf which is focused on metadata to Autoplot display.

This is not a getting started with CDF guide, but instead it is notes for one that doesn't yet exist.

• rVar/zVar difference -- use zvars
• representing timetags: Epochs, TT2000 -- use TT2000 and be done with it.
• metadata that connects variables (DEPEND_0)
• valid data and plot hints
• ISTP metadata and skeleton files
• structure for common examples
• How to create quick CDF data files from IDL
• working with SPDF to make your data available to the community

• Introduction to CDF file format
  • History and goals
  • What is it? Binary format that contains a named arrays and associations between arrays and metadata.
  • Uses compared to ascii data, etc.
  • CDAWeb is like a database but data is ingested in chunks and can be copied out into chunks.
• Reading CDF files
  • Reading in Autoplot
  • IDL and Matlab
  • Using CDAWeb make your data available to the world
• Writing CDF files from scratch
  • creating and installing data.
  • first metadata to specify title, fill values and scale settings
  • installing times. TT2000.
  • linking data DEPEND_0
  • multidimensional datasets.
  • varying vs non-varying data.
  • what's not talked about
    • sparse data
    • blocking
    • compression
• Using skeletons.
  • Separation of roles. Let the scientist specify the metadata, and the programmers handle the data.
  • a complete processing system
  • validating product with skteditor.
• Ingesting data into CDAWeb
  • naming CDF files. Filenaming conventions, mission, instrument, dates, and versioning.
  • what about missing files? Should an empty file be made?
• Example schemes for data
  • spectrogram
  • sparse data
• Limitations -- people have asked a lot about how big can CDF files be? I suggest <300 MB, but I know Autoplot has handled 1GB files.
• WriteMyCDF
• Reading and Writing in Java.

• variables should be single blocks. This provides efficient access with NIO.
• or cdfs should be compressed by variable, but in this case we use chunks to provide random access.
• Timetags are stored in TT2000 format, and are not compressed. (Uh-oh, does this contradict last two?)

Does Bob McGuire have a powerpoint that shows some of these ideas that we could link to?

Tami (CDAWeb) has a page showing different types of data in CDF files: http://spdf.gsfc.nasa.gov/istp_guide/variables.html