[doc] describe the block packing algorithm used by our various Keep clients (arv-put, python sdk, keep-web, arv-mount)
Describe the differences, the reason for them (e.g. information is not available due to access patterns), which write patterns would be most affected (many small files, presumably?).
Are there possible mitigation strategies to avoid duplication, e.g.:
- can we improve some of our clients to write normalized collections from the get-go?
- can we measure/predict which collections could be normalized?
- can we make a normalization tool?