If you are looking for a rigorous treatment of storage efficiency in storing data from sensor networks, this work covers exactly that. Storage optimization is an important concern in the design of systems broadly termed the Internet of Things. These systems are characterized by continuously growing data collected periodically from a large number of sensors. The authors propose a new scalable algorithm for storing massive volumes of data from a large number of sensors.

The scheme described takes advantage of a few important properties of sensor data. Sensors usually measure physical parameters such as temperature, pressure, and so on. According to the authors, such data values “are highly correlated in nature within spatial and temporal domains.” Correlation helps remove redundancies, resulting in highly compacted data. We may not require high accuracy in storing sensor data due to relatively high measurement error margins among sensors. Most often, approximate results are sufficient to discover trends or out-of-bounds values. Another important property of data captured from sensors is that the frequency of access decreases as time passes. Data aging suggests the lowering of data quality of stored data over long time periods. The major contribution of this work is a scalable management scheme layer deployable on top of any file system as an add-on, with low overhead for performance. The authors compare their work and results with other popular schemes such as the transforms applied in video compression. They also provide a good survey of existing schemes.

Finally, the authors discuss the optimal storage configuration and organization of data blocks. Experimental results and comparisons with popular coding techniques make the presentation complete. This is an important contribution toward building intelligent storage systems for applications involving the collection and management of data from sensor networks.

The paper is mathematically rigorous in presenting the details of the scheme. It will not be an easy read for those without sufficient background with encoding and transformation schemes.