I already improved a lot the performances of our ETL (x20 !) but I still want to know where I can improve performances. I seems that tables stats and column indexing slow down a bit writings so I want to decrease dataSkippingNumIndexedCols to match the small columns subset on which filters are actually applied, and columns with low distinct values. I wonder if it's worth doing that because it requires development work and columns order is imposed.
The documentation says it is worth doing that if your columns contain textual information for example because indexing is inefficient.