Whatever you can squeeze out of columnar store with indexes, where

1. You store & compress at least hundreds of thousands of rows together (as opposed e. g. to InfluxDb, that compresses not more than 1000 rows, it limits the compression efficiency. Primarily for string columns with dictionaries.)

2. You don’t care about efficiency of point updates and deletes, so you can create indexes as aggressive as you want, and can use compression that doesn’t allow for random access, update and delete.