近年來，雲端計算相關技術發展迅速，可更方便地取得計算資源進行應用，計算資源具備高可靠性與高可用性是雲端服務能穩定執行重要的指標。容錯技術是實現高可用性的重要技術之一，透過冗餘設計支持虛擬機器狀態備份，提供機器發生意外故障能快速恢復的能力，讓執行中的服務不被中斷。本研究致力於延續開發中央大學平行與分散式計算實驗室團隊在過去所開發及發展之虛擬機器容錯系統－NCU MFTVM。我們提出了一項投機式資料同步機制，透過追蹤系統記憶體狀態寫入狀態，並制定熱點寫入的頁面判斷規則，從而減少虛擬機器暫停期間需要同步的記憶體狀態資料量、降低虛擬機器暫停時間，增進系統效能。根據實驗結果顯示，改進後的方法能縮短0.35~16.88%的虛擬機器暫停時間。;In recent years, the rapid development of cloud computing related technology, more convenient access to computing resources for application, computing resources with high reliability and high availability is an important indicator of the stability of the implementation of cloud services. Fault tolerance is one of the most important technologies to achieve high availability. It supports virtual machine state backup through redundancy design and provides the ability to quickly recover from machine accidents, so that running services are not interrupted. This study is dedicated to the continuation of the development of the NCU MFTVM, a virtual machine fault-tolerant system developed by the Parallel and Distributed Computing Laboratory team at National Central University (NCU) in the past. We propose a speculative data synchronization mechanism that reduces the amount of RAM state data that needs to be synchronized during the VM stop time, reduces the VM stop time, and improves the system performance by tracking the memory state write state and formulating the page judgment rules for hotspot writes. According to the experimental results, the improved method can shorten the VM stop time by 0.35-16.88%.
