智能M2M系統的容錯研究

Abstract

對於具有 '部署一次，永遠運行' 概念的物聯網而言，容錯移轉是這類分散式服務導向網路的必備條件。當設備更換時或系統出狀況時，必須利用資源再規畫去達成容錯移轉的機制。系統在運作時，異質性的或多工性的設備之間若不僅是端對端的訊息傳輸時，不管是設備或是訊息的複製都是昂貴且累贅。特別是當設備某種服務故障時，可由另外一個有能力提供相同服務的同級設備接替其服務，而不一定要由相同設備取代。利用長帶來記錄一連串複製的服務訊息，每一個同級設備都保存一致的長帶記錄。結合常用於失敗偵測的心跳協定，系統由異常回復的機制可以藉由操控分析分散狀態的長帶來達成。使用Arduino mega 2560相容設備所做的實驗結果顯示，我們已經能夠使小型網路系統故障復原，較大的網路實驗則正在進行中。未來研究方向包括確認網路的可擴展性，網路磁碟分割 處理以及解決同步故障的問題。

Failover for service-oriented distributed networks is a prerequisite to enabling Internet-of-Things (IoT) in the sense of “deploy-once, run forever.” Resource reconfiguration is required to achieve failover mechanisms upon replacement of devices or failure of services. It can be particularly challenging when services in applications have more than end-to-end transmissions between devices that are heterogeneous or versatile, for which duplications can be costly and redundant. Specifically, a device with a failed service shall be taken over by another service peer, instead of a device counterpart to recover application as a whole. Strip is introduced to store a list of duplicated services, and, each service peer maintains a consistent view of strips. In combination with the heartbeat protocol which was widely applied for failure detection, recovery from failure can be achieved by manipulating strips in a distributed manner. Experiments using Arduino Mega 2560 compatible devices show that our approach is capable of failover in small networks, whereas experiments in larger networks are underway. Future research directions include addressing the scalability issue, network partitions and tackling simultaneous failures.