Embedded Operating System Design For Wireless Sensor Networks

Recent advances in low-power wireless technologies have enabled us to make use of wireless sensor networks (WSNs), a new class of networked embedded systems. Researchers have envisioned a wide variety of applications, such as environmental monitoring, military surveilance, and infrastructure protection, etc. Infrastructural support for numerous WSN applications in the form of operating systems (OSes) is of great importance. Despite many research efforts in the sensor OS design space, there is a huge gap towards demanding applications. To this end, we survey existing sensor OSes in a systematical manner. This work was published in IEEE Communications Surveys and Tutorials. Based on this survey, we abstract four critical issues in the OS design space. We have also designed and implemented a new sensor OS, SenSpire OS, to address these issues.(1) Flexibility. We propose a flexible programming model for both event-driven programming and multithreaded programming. SenSpire OS's hybrid model advances prior works in three aspects. First, it allows decoupling of the event-driven subsystem and multithreaded subsystem, hence is flexible in expressing and customizing different scheduling policies. Second, it supports event preemption,which is important for achieving system predictability in the event-driven subsystem. Third, it employs a novel differentiated kernel scheduling scheme to exploit stack sharing, hence can reduce the stack memory consumption. This work was accepted by IEEE Transactions on Computers.(2) Real-Time performance. We propose FIT, a real-time scheduling mechanism for sensor nodes. FIT is the first work that performs detailed real-time analysis for sensor nodes. We have also propose an algorithm, MTSP, to optimize the resource consumption, which is important for resource-constrained sensor nodes. This work was published in IEEE/ACM DCOSS 2008 and IEEE Transactions on Parallel and Distributed Systems.(3) Expandability and modularity. We add modular support to SenSpire OS and propose a flexible and efficient module file format, SELF, for dynamically linking and loading. Our design has two major advantages. First, we identify the size and flexibility tradeoff, and propose the SELF format that minimizes the module file size while remaining necessary information for flexible modular programming and inter-module communications. Second, we address the module dependency issue for improved reliability and efficiency. This work was published in IEEE MASS 2009. The modular approach is very effective in reduce energy consumption in network reprogramming. Borrowing the modular idea from SenSpire OS, we have also proposed a new reprogramming approach based on TinyOS, Elon, to reduce the dissemination cost and prolong the reprogramming lifetime. This work was published in ACM SIGMETRICS 2010. We have also proposed a lightweight and density-aware reprogramming protocol for efficiently disseminate the program code. The proposed reprogramming protocol can significantly reduce the dissemination cost. This work was accepted by IEEE Transactions on Mobile Computing.(4) Adaptability. Adaptable communication is important for a WSN which is typically deployed in the wild with dynamically changing environments. We propose DPLC, a dynamic packet length control scheme, to optimize the packet length for improved transmission throughput and energy efficiency. To the best of our knowledge, this is the first work that dynamically optimize the packet length for WSNs. Compared to a typical data collection protocol, DPLC results in 13% reduction in transmission overhead and 41.8% reduction in energy consumption. This work was published in IEEE INFOCOM 2010 and submitted to IEEE Transactions on Computers (under revision).