Spatio-temporal Histograms

Abstract

Abstract. This paper presents a framework for building and continuously maintaining spatio-temporal histograms (ST-Histograms, for short). ST-Histograms are used for selectivity estimation of continuous pipelined query operators. Unlike traditional histograms that examine and/or sample all incoming data tuples, ST-Histograms are built by monitoring the actual selectivities of the outstanding continuous queries. ST-Histograms have three main features: (1) The ST-Histograms are built with (almost) no overhead to the system. We use only feedback (i.e., the actual selectivity) from the existing continuous queries. (2) Rather than wasting system resources in maintaining accurate histograms for the whole spatial space, we only maintain accurate histograms for that part of the space that is relevant to the current existing queries. The rest of the space has less accurate histograms. (3) The ST-Histograms are equipped with a periodicity detection procedure that predicts the future execution of the continuous queries. Hence, the query processing engine can continuously adapt the continuous query pipeline to reflect this prediction. Experimental results based on a real implementation inside a data stream management system show a superior performance of ST-Histograms in terms of providing accurate operator selectivity estimations with no extra overhead.

Aggregate Location Monitoring for Wireless Sensor Networks: A Histogram-Based Approach

“…Thus, it takes care of the limitations of wireless sensor networks that include limited power and communication bandwidth. Experimental results show that the adaptive histogram succeeds in giving highly accurate answers and clearly outperforms a basic histogram and the state-of-theart spatio-temporal histogram [20] by orders of magnitudes, especially, for the cases of skewed data distributions.…”

“…In the skewed mobility pattern environment, the accuracy of our adaptive histogram is at least two orders of magnitude better than a basic histogram and an existing spatio-temporal histogram [20]. Furthermore, the experimental results show that the update time of our adaptive histogram is at least two orders of magnitude better than the basic histogram.…”

“…Baseline histograms. We compare the Adaptive histogram with two baseline histograms, basic histogram described in Section 4.2 (denoted as Basic) and an existing spatiotemporal histogram [20] (denoted as Uniform). We choose the Uniform histogram to compare with the proposed Adaptive histogram because it is the closest work to ours.…”

“…It is important to note that it is always the case that the input to these previous works is user's exact locations, while the input to our query processor, that employs a spatiotemporal histogram, is a set of aggregate locations. The closest spatio-temporal histogram to ours is the one proposed in [20]. However, such histogram works only under the strict assumption of uniform distribution.…”

“…However, such histogram works only under the strict assumption of uniform distribution. We will extensively study the performance gain from applying our adaptive histogram with respect to [20]. Figure 1 depicts the system architecture of our aggregate location monitoring system that consists of two major components, wireless sensor network and aggregate query processor.…”