Evaluating holistic aggregators efficiently for very large datasets