Update the content of chunking_introduction.ipynb

[guillaumeeb]

The goal of this issue is to discuss what we should put in chunking introduction notebook.

This comes from comments and discussions in #45.

I believe we should do the following, but this needs discussion (especially with @tinaok):

• Either remove the word compression or talk about it somewhere, a short chapter should be fine. Some subjects to talk about
  • Generalities (why compression matters, how it is a gain in processing time despite the overcost of compression -- loading big files in memory is slow, different algorithms currently used in scientific domains, Python blosc, etc.),
  • A section showing the difference between a NetCDF or zarr file with and witout compression and the resulting size
  • Default of Xarray / Various data formats (is NetCDF compression activated by default?).
  • Compressions is often apply on chunks for optimized access.
• Add more context about Xarray and Chunking before introducing kerchunk. Begin with open_dataset and the chunks attribute, introduce Dask Arrays, talk about lazyness and sequential processing. Talk also of native chunks in files here, if possible without kerchunk (with h5py maybe?).
• Kerchunk part: introduce it by talking about the difficulties of reading a dataset composed of several files in a optimal way. Then explain how it reads file metadata and detect chunks, and construct a Zarr compatible metadatafiles that allows for optimized access of whole dataset with Xarray. Should we keep the main metadata file creation in the notebook? I'm not sure of that.

[tinaok]

Compression part will be moved #53 and I'll update the title of chunking_introduction.ipynb.

[annefou]

A picture to show chunks will be added by @tinaok

[guillaumeeb]

About compression, as said on my email, I'll add a few things if I found the time, but not a priority.

[guillaumeeb]

About native chunks vs Xarray chunks=auto:

• Check that Xarray method does not uses native file chunks, but say that it is still some optimisation on data loading
• See if we can obtain native chunks with something simpler than kerchunk,
• Use these native chunks in Xarray to speed up access on a single file.
• Which introduces Kerchunk, especially in multiple file context.

[tinaok]

@annefou plz merge #56 ? @ guillaumeeb you can join us the discussion at Gitter?