無線感測網路之動態軟體更新

Abstract

無線感測網路是由許多便宜、微小、自治的裝置組成，他們常用於監測環境況狀，並彼此交換通訊資訊。在將他們佈建之後，仍有可能需要更新上面的程式，例如為了修正程式錯誤、需求修改，或是其他的維護。因為在無線感測網路下軟硬體或環境較為受限，在無線感測網路上的軟體更新成為一項挑戰與議題。 在先前的研究，軟體更新後往往需要將裝置重新啟動。然而，重新啟動的成本所費不貲，因為裝置運行所有蒐集與建立的珍貴資料都會遺失。例如需要時間重新與其他裝置同步，或是需要額外的網路頻寬來重新路由表。 這篇提出了uFlow，他是一個程式範式和他的簡單的原型實作。uFlow可以讓程式在更新後，直接套用新的程式，而不需要重新啟動裝置，因此得以保留資料。uFlow執行時，大約會有88時脈週期的額外負擔，只發生在工作切換時。而原型引擎大約需要420位元組用於執行迴圈與核心功能，而TinyOS-1.x的執行核心則大約需要400位元組。

A wireless sensor network consists a number of small, cheap and autonomous devices communicating with each other for monitoring environment conditions. After the deployment, there is still a need for updating the software in the nodes due to bug fixes, requirement changes, or other maintenance reasons. Software updating in wireless sensor networks has became a challenge because of the constrained hardware resources and environment. In previous works, the node is required to reboot after software update. However, reboot is costly since the previous runtime status is lost. It needs time and bandwidth to synchronize with other nodes or rebuild routing table. The thesis presents uFlow, a programming paradigm and a prototype implementation for wireless sensor networks. uFlow allows application to update without rebooting the node therefore preserving precious runtime states. The execution overhead is around 88 clock cycles, which is slightly larger than TinOS, but fortunately the overhead only occurs when task transitions. The prototype implementation is about 420 bytes for task loop and core helper APIs and 146 bytes for helper APIs used to facilitate the dynamic update, while TinyOS-1.x costs around 400bytes.