減少雲網路流量和主機數量之虛擬機器動態放置技術

Abstract

網際網路的進步，至今已發展至雲端應用。對基礎建設即服務供應商 (Infrastructure-as-a-Service, IaaS)而言，為了提供穩定的服務品質，從早期單一資 料中心(雲)，衍生至兩個以上資料中心。同時，相較實體機器，虛擬機器技術可方 便進行搬遷至異地的特性，更容易滿足使用者想要隨時隨地使用服務的需求。此 外，搭配動態搬遷也使得雲端可達成負載平衡、提升系統資源利用率、以及提高 系統強健度以避免單一資料中心服務失效之問題。然而，為達到上述功能，虛擬 機器間一定會有資料傳遞產生網路流量，若彼此互相處於不同資料中心，因需要 透過 ISP 供應商提供的線路，會有傳輸費用的產生。隨著雲端盛行，對供應商而 言，就會有更多的網路費用成本的支出。因此，在本論文中，我們針對降低雲端 之間的網際網路流量，設計以流量搭配減少主機數量之虛擬機器動態放置策略。 其特色在於透過搬遷虛擬機器的方式，對於虛擬機器之間有流量且分處於不同資 料中心的情形，將其中一虛擬機器動態搬遷至另一端資料中心，使此二虛擬機器 均處在同個資料中心，原二者之間產生的網際網路流量也就轉成內聯網流量，不 會有 ISP 收費產生。此外，替虛擬機器選擇適當實體主機來放置，我們也藉此過 程達成搬空使用率較低之主機目的。

With cloud computing growing rapidly and available to all, cloud provider can build datacenter, or called cloud, through internet, share resources to achieve coherence and economies of scale, similar to a utility over a network. To achieve it, physical resources provided by Infrastructure-as-a-Service (IaaS) of clouds can be partitioned into flexible and scalable virtual computing units, namely Virtual Machines (VMs) through virtualization technologies. Besides, cloud dynamic allocation leverages the ability to do live migration of VMs and cloud provider through this types easily achieve different kinds of purposes, such as commonly loading balance of physical machines, high utilization of servers and high available of services. For these features of cloud maintenance, traffic generated by VMs (inter-VM traffic) could be across internet. When this is happened, the inter-VM traffic which has to be passed through the rent ISP link is called inter-cloud traffic. Moreover, with more inter-cloud traffic, the less available bandwidth, cloud provider may need to rent more bandwidth from ISP or in face of low performance. Hence, the aim of our work has two parts. The first part is to reduce inter-cloud traffic. We can calculate and find out which inter-cloud traffic is worth transited to intra-cloud through VM migration in advance. Especially, we consider VM size (migration cost) of a designate VM and can do live migration of VMs. The second part is to reduce power consuming through VM allocation in intra-cloud after the first part is done.