考慮虛擬機器間傳輸量的虛擬機器搬移機制

Abstract

在叢集式伺服器上提供穩定的服務及有效利用資源是很重要的一個課題。以往在伺服器上發生資源負荷過載的處理方法為行程中斷或是行程搬移，這兩個方法皆不夠有彈性及透明化。前者直接中斷行程，讓使用者感受到不便；後者雖可有效解決資源負荷過載的問題，但是必須先克服一些技術上的問題。 隨著虛擬機器的普及化，有人研究利用虛擬機器的搬移機制來解決資源負荷過載且擺脫以往不透明及沒有彈性的問題。但是之前的研究全都是把負載最重的虛擬機器搬移到負載最輕的機器上。 在此論文中，我們觀察出虛擬機器間內部傳輸較佳的特性，並根據此特性設計出一套可以增進叢集式伺服器內虛擬機器整體效能的搬移機制。在資源負荷過載時，此搬移機制可以把傳輸量大的虛擬機器聚在一起，籍以求得較佳的效能。

Using resource effectively and providing reliable services are both significant in the cluster. The previous researches to solve system overload are process suspend and process migration, but they are all not flexible and transparent. The former make user be aware of termination of service. And the latter can solve system overload successfully but it need some technique supports. Some researches solve system overload by VM migration, it can successfully solve system overload and get more transparent and flexible. But the main idea of previous researches focus on that migrate the VM with highest load to lowest load destination. In this thesis, we know that the performance of network traffic via Inter-Virtual-Machine Communication is better than via network. According to this feature, we present a Inter-Virtual-Machine Communication aware migration policy. When system overload occurs, the policy will union the VMs that have high traffic to each other in the same physical machine to get better performance.