Loading: Experiences with approximating queries in Microsoft’s production big-data clusters