Streaming piecewise linear approximation for efficient data management in edge computing.

In our digitalization era, where large and continuous data streams are produced by an ever increasing number of sensors, data retrieval and storage from edge devices is hampered when data volumes exceed the communication bandwidth of cyber-physical systems. Piecewise Linear Approximation (PLA), which trades space against precision by representing some portion of data by segments, could reduce the volume of transmitted and stored data and thus be beneficial to a wide range of edge/fog system architectures, saving communication bandwidth and addressing the aforementioned drawback. Porting a well-established tool such as PLA into the streaming paradigm is nonetheless challenging, and attention has to be payed to balance achievable compression, delays and imprecision. We analyze such challenges and propose different solutions to meet them. Our main contribution is a set of streaming PLA techniques that allow compression of the input data stream on the fly, tolerating a bounded maximum error. Through an experimental study based on real data, we demonstrate the superiority of our techniques in all sought aspects over preceding methods.