隨著雲端運算與虛擬化技術的普及，資料中心大量依賴虛擬機器以提升資源彈性與服務可用性，但也因集中化而提高單點失效風險，使虛擬機容錯機制成為確保零停機不可或缺的關鍵。其中，記憶體同步機制是維持主要與備援虛擬機器一致性的核心技術，也是系統效能之瓶頸來源之一。本研究延續中央大學所提出的 NCU MFTVM 容錯系統進行改良，於 Ubuntu 24.04.1 LTS 與 Linux kernel v6.7.0 環境中，透過導入 KVM Dirty Ring 同步機制，取代傳統 KVM Dirty Logging 依賴的 bitmap 架構，並分析其原始實作的效能限制，進一步提出延後清除策略與同步互斥鎖擺放的優化，透過平行化與多執行緒設計，優化同步流程並減少鎖競爭對同步效能的負面影響。實驗結果顯示，優化後的 KVM Dirty Ring 在低到中高的記憶體變動頻率下，可較原系統最多減少約 12.88% 的停機時間，並提升約 5.89% 的整體系統效能；而在高記憶體變動頻率下，也可減少約 7.24% 的停機時間，並維持相當的系統效能。;With the widespread adoption of cloud computing and virtualization technologies, data centers heavily rely on virtual machines to enhance resource flexibility and service availability. However, such centralization also increases the risk of single points of failure, making virtual machine fault-tolerance mechanisms essential for ensuring zero downtime. Among these mechanisms, memory synchronization is a core technology for maintaining consistency between primary and backup virtual machines, and is also a major source of system performance bottlenecks. This study builds upon the NCU MFTVM fault-tolerant system, implementing improvements in an Ubuntu 24.04.1 LTS and Linux kernel v6.7.0 environment. By introducing the KVM Dirty Ring synchronization mechanism to replace the traditional bitmap-based KVM Dirty Logging, this research analyzes the performance limitations of the original implementation. Further, a deferred clearing strategy and optimizations in the placement of synchronization mutex locks are proposed. Through parallelization and multi-threaded design, the synchronization process is optimized and the negative impact of lock contention on synchronization performance is reduced. Experimental results show that the optimized KVM Dirty Ring can reduce downtime by up to 12.88\% and improve overall system performance by approximately 5.89\% under low to moderately high dirty rates, compared to the original system. Even under high memory dirty rates, downtime can still be reduced by about 7.24\%, while maintaining comparable system performance.
