File System Support for Adjustable Resolution Applications in Sensor Networks

Flash memory is often the technology of choice for sensor networks because of its cost-effectiveness and attractive energy properties. In storage- constrained sensor network applications, the monitored data is typically stored in multi-resolution fashion. This allows reclamation of some storage space when needed by reducing the quality of stored data by eliminating some of the precision. Existing sensor network file systems are optimized for sequential logging of the data. However, flash memories have a number of unique properties that require careful consideration in file system design. In this paper, we show that in applications where adjustable resolution occurs, sequential logging file-systems result in an inefficient implementation of adjustable resolution. We propose an alternative implementation of the file system where data components are grouped with each other according to resolution. Thus, reducing resolution is implemented by simply erasing the pages with the excess resolution components. We have implemented the proposed scheme on crossbow MICA2 sensor nodes. In addition, using TOSSIM simulations, we show that as compared to the existing approach, the proposed scheme results in significant savings in read and write operations to the flash (thereby in turn saving energy, and reducing wear). Further, we show that wear leveling can be maintained over time by assigning the most significant data to the most frequently used pages.