LE-Stream : a latency and energy-aware framework for data stream processing in the Internet of Things

Abstract

Data processing in IoT is challenging due to its dynamic and heterogeneous nature, and the massive amount of generated data. Sensor data suffers from uncertainty and inconsistency issues, that can affect its accuracy. Several IoT applications are time sensitive, requiring fast data processing. Finally, as IoT devices are often battery powered, processing tasks must be performed in an energy-efficient way. Therefore, there are challenges in data stream processing concerning three dimensions: accuracy, latency and energy. We propose LE-STREAM, a framework to support the data stream processing for IoT systems which jointly addresses these dimensions. It leverages edge computing to bring the data processing closer to the data sources, thus minimizing latency. A novel collaborative adaptive sampling combined with a two-step data prediction model reduce the energy consumption of devices without compromising data accuracy. An active node selection schema improves the workload distribution among devices, also tackling the energy dimension by promoting a graceful degradation of devices resources.