Research possible effect of the use of ZDDs on internal data size and processing time

https://en.wikipedia.org/wiki/Zero-suppressed_decision_diagram

Zero-suppressed decisions diagrams represent sets of sets of elements of some universe. Their notable feature is that they compress representation as much as possible. For example, representing Cartesian product of two sets A and B requires size(A) + size(B) space instead of size(A)*size(B). Our join algorithm does the latter and it would be nice to have space savings of ZDDs with materialization as late as possible.

There is a join operation over ZDDs: https://www.jstage.jst.go.jp/article/transinf/advpub/0/advpub_2025EDP7062/_pdf

Finally, if we allow for several ZDDs roots with same compressed representation, we can associate i-th root with the 2^i multiplier. This way we can represent multisets and they will be compressed as well - subsets for 2^i and 2^j will share data between them as much as possible.

The task here is the following:

1. Create TPC-DS data and scale 1, 5 and 10 and look whether data inside tables can be compressed with ZDDs, report results for different scales,
2. Implement joins from TPC-DS queries (without GROUP BY - queries 16 and 96 are two such examples) and report on memory requirements and runtimes for different scales.

The library to use: https://github.com/kunisura/tdzdd (maybe some other - there are libraries for parallel ZDD operations).