Pull-based modeling and algorithms for real-Time provision of high-frequency sensor data from sensor observation services

Abstract

The widely used pull-based method for high-frequency sensor data acquisition from Sensor Observation Services (SOS) is not efficient in real-Time applications; therefore, further attention must be paid to real-Time mechanisms in the provision process if sensor webs are to achieve their full potential. To address this problem, we created a data provision problem model, and compare the recursive algorithm Kalman Filter (KF) and our two proposed self-Adaptive linear algorithms Harvestor Additive Increase and Multiplicative Decrease (H-AIMD) and Harvestor Multiplicative Increase and Additive Decrease (H-MIAD) with the commonly used Static Policy, which requests data with an unchanged time interval. We also developed a comprehensive performance evaluation method that considers the real-Time capacity and resource waste to compare the performance of the four data provision algorithms. Experiments with real sensor data show that the Static Policy needs accurate priori parameters, Kalman Filter is most suitable for the data provision of sensors with long-Term stable time intervals, and H-AIMD is the steadiest with better efficiency and less delayed number of data while with a higher resource waste than the others for data streams with much fluctuations in time intervals. The proposed model and algorithms are useful as a basic reference for real-Time applications by pull-based stream data acquisition.