虛擬化技術能提升雲端運算伺服器的運作效能。而即時轉移(Live Migration)技術主要能使伺服器(Source)發生預設性/災難性狀況時暫時關機，並且把目前伺服器狀態移到其他伺服器(Destination)上，整個過程能讓使用者不會因為目前提供服務的伺服器停止服務而中斷服務。 評估即時轉移的效能指標為停機時間(Downtime)與總轉移時間(Total Migration Time)。其中使用者會感受到使用中的服務被中斷的狀況，為停機時間(Downtime)發生的時候；目前的技術使用預複製轉移(Pre-copy Migration)雖可減少停機時間，但仍會因為多次重覆轉移相同頁面，造成多次無意義的網路轉移，使得總轉移時間變長。 不同於預複製轉移(Pre-copy Migration)，在本論文中提出利用記憶體預測機制與頁面相對修改率的方法，在各疊代（Round）轉移前先判斷哪些頁面連續修改的機率較低，適合於疊代轉移階段(Interactive Pre-copy)轉移；哪些頁面連續修改的機率較高，適合於暫停-複製階段(Stop-and-copy Phase)轉移。依據記憶體頁面的特性來選擇轉移的時機，藉此減少不必要的網路轉移，進而減少總轉移時間。 Virtualization enhances the performance of cloud servers but live migration technique can turn off the source server temporarily when assumptive or disastrous situations occur and migrate the status of the current server to other destinations. During the process, live migration allows the customers to experience seamless services even without sensing the interruption of services. Two major performance metrics of live migration are downtime and total migration time. When the customers feel that the service is terminated, it means the occurrence of the downtime. Many current techniques use pre-copy migration to reduce the downtime, but to migrate the same pages repeatedly may result in meaningless retransmissions and prolong the total migration time. Different from pre-copy migration, this paper proposes to use memory prediction mechanism and related dirty rate to find out the pages with the lower dirty rate to be migrated in the interactive pre-copy phase, and the pages with the higher dirty rate to be migrated in the stop-and-copy phase. The simulation result proves that by choosing the timing for migration according to the features of the memory pages, our mechanism decreases not only unnecessary transmissions, but also the total migration time.
