修復分散式發布訂閱服務演算法之設計與實作

Abstract

近年來，網路技術應用於遞送災難管理的訊息。隨著人們依賴通訊科技，災難環境中的網路服務品質越來越重要。但是現存多數網路服務原先並非為了災難管理而設計。在不穩定的網路環境，它們的可靠度不足以應付可能發生的服務故障。 本論文提出一種修復機制，可加強現有的分散式訊息遞送軟體的可靠度。修復機制讓廣域網路上故障的服務自動被另外一個服務所替代。另外，本論文也提出分散式演算法解決服務修復過程中的共識問題。藉由修復機制，訊息遞送系統可以容忍並行發生的故障，提供更好的服務品質。

In recent years, network technologies have been applied on message delivery in disaster management. As people relies on these communication technologies, quality of service becomes more and more important in disaster scenario. However, most existing network services are not originally designed for disaster management. Under the environment of unstable network, they do not provide sufficient reliability support to deal with service failure. In this work, we designed recovery mechanisms for an existing distributed message delivery system to improve its reliability. The proposed mechanism makes a failed service be replaced automatically with another one on wide area network. Furthermore, we proposed a distributed algorithm to solve the consensus problem in service recovery. With the recovery mechanism, the message delivery system can tolerate concurrent failure on individual services, and thus provides better quality of service.