Το έργο με τίτλο Prediction-based geometric monitoring over distributed data streams από τον/τους δημιουργό/ούς Giatrákos, Níkos, Deligiannakis Antonios, Garofalakis Minos, Sharfman Izchak, Schuster Assaf διατίθεται με την άδεια Creative Commons Αναφορά Δημιουργού 4.0 Διεθνές
Βιβλιογραφική Αναφορά
N. Giatrakos, A. Deligiannakis, M. Garofalakis, I. Sharfman and A. Schuster, "Prediction-based geometric monitoring over distributed data streams," in 2012 ACM SIGMOD International Conference on Management of Data, pp. 265-276. doi: 10.1145/2213836.2213867
https://doi.org/10.1145/2213836.2213867
Many modern streaming applications, such as online analysis of fi-nancial, network, sensor and other forms of data are inherently distributedin nature. An important query type that is the focal point insuch application scenarios regards actuation queries, where properaction is dictated based on a trigger condition placed upon the currentvalue that a monitored function receives. Recent work [18,20, 21] studies the problem of (non-linear) sophisticated functiontracking in a distributed manner. The main concept behind the geometricmonitoring approach proposed there, is for each distributedsite to perform the function monitoring over an appropriate subsetof the input domain. In the current work, we examine whetherthe distributed monitoring mechanism can become more efficient,in terms of the number of communicated messages, by extendingthe geometric monitoring framework to utilize prediction models.We initially describe a number of local estimators (predictors) thatare useful for the applications that we consider and which have alreadybeen shown particularly useful in past work. We then demonstratethe feasibility of incorporating predictors in the geometricmonitoring framework and show that prediction-based geometricmonitoring in fact generalizes the original geometric monitoringframework. We propose a large variety of different predictionbasedmonitoring models for the distributed threshold monitoringof complex functions. Our extensive experimentation with a varietyof real data sets, functions and parameter settings indicates that ourapproaches can provide significant communication savings rangingbetween two times and up to three orders of magnitude, comparedto the transmission cost of the original monitoring framework.