Hi,
I can share what we did, note that this datasource generates a lot of data
1. We put a write optimized DSO between datasource and cube, mainly because we are not keeping data in PSA and our requirement is to only load a part of the data in the cube according to complex logic
2. Agree, but see point 1.
3. We are using pseudo delta for INPER and also parallel extractions (different infopackages) by company and employee type.
Hope it helps