ClusterSheddy : Load Shedding Using Moving Clusters over Spatio-temporal Data Streams

Moving object environments are characterized by large numbers of objects continuously sending location updates. At times, data arrival rates may spike up, causing the load on the system to exceed its capacity. This may result in increased output latencies, potentially leading to invalid or obsolete answers. Dropping data randomly, the most frequently used approach in the literature for load shedding, may adversely affect the accuracy of the results. We thus propose a load shedding technique customized for spatio-temporal stream data. In our model, spatiotemporal properties, such as location, time, direction and speed over time, serve as critical factors in the load shedding decision. The main idea is to abstract similarly moving objects into moving clusters which serve as summaries of their members’ movement. Based on resource restrictions, members within clusters may be selectively discarded, while their locations are being approximated by their respective moving clusters. Our experimental study illustrates the performance gains achieved by our load-shedding framework and the tradeoff between the amount of data shed and the result accuracy.