Resource Sharing in Continuous Sliding-Window Aggregates

Abstract

We consider the problem of resource sharing when processing large numbers of continuous queries. We specifically address sliding-window aggregates over data streams, an important class of continuous operators for which sharing has not been addressed. We present a suite of sharing techniques that cover a wide range of possible scenarios: different classes of aggregation functions (algebraic, distributive, holistic), different window types (time-based, tuple-based, suffix, historical), and different input models (single stream, multiple substreams). We provide precise theoretical performance guarantees for our techniques, and show their practical effectiveness through a thorough experimental study.