Rethinking Data Management for Storage-centric Sensor Networks.

TLDR

Data management in wireless sensor networks has been a major research focus, with many systems treating sensor data as a continuous stream processed as it flows from sensors to users. The authors contend that advances in flash memory and embedded platforms require rethinking sensor network architecture toward storage‑centric nodes. They propose equipping nodes with high‑capacity, energy‑efficient flash storage and redesigning data management to exploit local storage and processing, and present StonesDB, which uses energy‑efficient flash, multi‑resolution storage and aging, query processing, and intelligent caching. StonesDB demonstrates the feasibility of the storage‑centric architecture by efficiently leveraging flash, providing multi‑resolution storage and aging, supporting query processing, and employing intelligent caching.

Abstract

Data management in wireless sensor networks has been an area of significant research in recent years. Many existing sensor data management systems view sensor data as a continuous stream that is sensed, filtered, processed, and aggregated as it “flows” from sensors to users. We argue that technology trends in flash memories and embedded platforms call for re-thinking this architecture. We articulate a vision of a storage-centric sensor network where sensor nodes will be equipped with high-capacity and energy-efficient local flash storage. We argue that the data management infrastructure will need substantial redesign to fully exploit the presence of local storage and processing capability in order to reduce expensive communication. We then describe how StonesDB enables this vision through a number of innovative features including energyefficient use of flash memory, multi-resolution storage and aging, query processing, and intelligent caching.