Аннотации:
© 2013 IEEE. Cloud and Fog computing have established a convenient and widely adopted approach for computation offloading, where raw data generated by edge devices in the Internet of Things (IoT) context is collected and processed remotely. This vertical offloading pattern, however, typically does not take into account increasingly pressing time constraints of the emerging IoT scenarios, in which numerous data sources, including human agents (i.e., Social IoT), continuously generate large amounts of data to be processed in a timely manner. Big data solutions could be applied in this respect, provided that networking issues and limitations related to connectivity of edge devices are properly addressed. Although edge devices are traditionally considered to be resource-constrained, main limitations refer to energy, networking, and memory capacities, whereas their ever-growing processing capabilities are already sufficient to be effectively involved in actual (big data) processing. In this context, the role of human agents is no longer limited to passive data generation, but can also include their voluntary involvement in relatively complex computations. This way, users can share their personal computational resources (i.e., mobile phones) to support collaborative data processing, thereby turning the existing IoT into a global cyber-physical-social system (CPSS). To this extent, this paper proposes a novel IoT/CPSS data processing pattern based on the stream processing technology, aiming to distribute the workload among a cluster of edge devices, involving mobile nodes shared by contributors on a voluntary basis, and paving the way for cluster computing at the edge. Experiments on an intelligent surveillance system deployed on an edge device cluster demonstrate the feasibility of the proposed approach, illustrating how its distributed in-memory data processing architecture can be effective.
