Providing Virtual Memory Support for Sensor Networks with Mass Data Processing

With the development of sensor networks and emerging of various sensors, sensor networks are capable of acquiring mass data to achieve much more complex monitoring tasks than ever. For example, image sensor nodes take photos using cameras, and images are collected and processed or stored for further processing. So, mass data processing is required for these sensor networks. However, low-power resource-constrained sensor nodes are normally equipped with kilobytes of RAM which might be not enough for storing large data for processing. In this paper, we propose an optimized virtual memory mechanism for large data processing on low-power sensor nodes. We point out the major overhead of virtual memory for large data processing on sensor nodes and introduce efficient solutions to address these issues. Evaluation shows that the overhead of the proposed virtual memory is reduced to an affordable range. We further compare the energy consumption of data processing programs using virtual memory with other means that process or transmit data. Data processing using virtual memory can be significantly more energy efficient than data processing using rich-resource sensor nodes or transmitting data to powerful gateways for central processing.