Prediction-based geometric monitoring over distributed data streams

摘要: Many modern streaming applications, such as online analysis of financial, network, sensor and other forms data are inherently distributed in nature. An important query type that is the focal point application scenarios regards actuation queries, where proper action dictated based on a trigger condition placed upon current value monitored function receives. Recent work studies problem (non-linear) sophisticated tracking manner. The main concept behind geometric monitoring approach proposed there, for each site to perform over an appropriate subset input domain. In work, we examine whether mechanism can become more efficient, terms number communicated messages, by extending framework utilize prediction models. We initially describe local estimators (predictors) useful applications consider which have already been shown particularly past work. then demonstrate feasibility incorporating predictors show prediction-based fact generalizes original framework. propose large variety different models threshold complex functions. Our extensive experimentation with real sets, functions parameter settings indicates our approaches provide significant communication savings ranging between two times up three orders magnitude, compared transmission cost