HoppingTimer: A Near-optimal Framework for Basic Estimation of Data Streams in Hopping Windows

In high-speed data streams, recent items are often much more significant than outdated ones. Therefore, basic estimation of data streams in hopping windows is an important topic. Basic estimation tasks include cardinality estimation and membership query. There are three classic algorithms for basic tasks in fixed windows. The design goal of this paper is to devise a generic and near-optimal framework to adapt them to hopping windows. In this paper, we propose the HoppingTimer, a generic and near-optimal framework which can adapt fixed-window algorithms to time-based and count-based hopping windows for basic tasks. The key idea of HoppingTimer is to use hopping timestamps and local cleaning to clean outdated items. We apply HoppingTimer to three algorithms for basic tasks. Experimental results show that HoppingTimer is near-optimal in hopping windows, and achieves false positive rate about 1000 times lower than the state-of-the-art when using metrics of sliding-window model.¹