隨著雲端計算與虛擬化技術的快速發展，資訊產業得以利用相關技術提升實體機器的利用效率並達成彈性的資源分配；然而在將多個伺服器整合到同一實體機器之時，也產生單一主機硬體故障即會導致多個服務失效的問題。基於虛擬化技術的容錯系統可以在主機硬體發生故障時，保護關鍵服務之虛擬機器運作狀態與其執行的 soft real-time 程式，進一步提升服務的可用性。 本研究基於 QEMU 3.0.0 、 libvirt 5.7.0 與持續同步的架構實作可透過外部管理介面控制的容錯系統，其中的持續同步架構藉由不斷同步主要虛擬機器與備援虛擬機器的狀態、並保證對外輸出的一致性，以達到容錯系統之基本要求。同時本研究以引入壓縮工具降低同步所需之頻寬、感知虛擬機器工作負載並進行參數設定等方式，協助系統管理者提升服務於容錯系統運作之效能。;The IT industries have commonly adopted the concept of cloud computing and virtualization, making resource management more efficient and elastic. However, as more servers are consolidated into one physical server, availability will be threatened by a single physical host′s hardware failure. A virtualization-based fault-tolerant system can protect mission-critical virtual machines running soft real-time applications from such hardware failures, thus improving the services′ availability. Based on QEMU 3.0.0, libvirt 5.7.0, and continuous checkpointing, this study implements a virtualization-based fault-tolerant system with a management interface. Continuous checkpointing keeps replicating internal states of VM on the primary host to backup host to meet the requirements of fault tolerance, and outputs are buffered to ensure consistency. This study also designed and implemented two methods to reduce the performance degradation of guest applications brought by the system; by adjusting the checkpointing parameter automatically and utilizing compression tools to speed up dirty pages transfer on demand, system administrators can set up the system without finding out suitable parameter for every application and have more flexibility to deploy the system.
