Virtualization and Virtual Data Management for Field Data Sensing

High-data-rate sensing and processing in the field are important for emerging applications in intelligent transportation, surveillance, and environmental management. For these systems, virtualization and virtual data management bring the benefits of easy deployment, flexible control, and strong data containment. Despite limited resources in the field, we show that optimized virtualization support can realize nearnative performance for data-intensive sensing workloads. To further reduce storage and network I/O costs, we present new virtual storage deduplication mechanisms that identify duplicates in continuous sensing workloads at low CPU and memory uses. Finally, we support efficient encryption of sensor-captured and derived data in a virtual disk through storage versioning and deduplication. Our Xen-based system prototype is deployed on an Intel Atom-equipped device (below 7 W). We experimented with realistic field workloads and data streams for vehicle traffic analysis, face detection, and motion capture. Measurements show that the virtualization incurs no more than 8% CPU overhead. Additional data deduplication and encryption lead to a total of below 18% CPU cost and 0.1 W power use. Our I/O saving mechanisms reduce Flash I/O writes by up to 33% and save VM installation/upgrade network I/O by 3 to 10 times. This work was supported in part by the National Science Foundation grants CNS-1217372, CNS-1239423, and CCF-1255729, and by a Google Research Award.