It helps with having smaller RAM requirement. And I have the group by, and materilized indices, which helps a lot to preserve huge modified datasets.