How to screen a data stream: Quality-driven load shedding in sensor data streams

As most data stream sources exhibit bursty data rates, data stream management systems must recurrently cope with load spikes that exceed the average workload to a considerable degree. To guarantee low-latency processing results, load has to be shed from the stream, when data rates overstress system resources. There exist numerous load shedding strategies to delete excess data. However, the consequent data loss leads to incomplete and/or inaccurate results during the ongoing stream processing. In this paper, we present a novel quality-driven load shedding approach that screens the data stream to find and discard data items of minor quality. The data quality of stream processing results is maximized under the adverse condition of data overload. After an introduction to data quality management in data streams, we define three data quality-driven load shedding algorithms, which minimize the approximation error of aggregations and maximize the completeness of join processing results, respectively. Finally, we demonstrate their superiority over existing load shedding techniques at real-life weather data.