在實體與虛擬環境下應用資源使用之比較

Abstract

隨著雲端科技以及虛擬化技術的興起，愈來愈多的企業試圖將內部的應用程式轉移到雲端環境之中。然而，虛擬化技術讓應用程式在執行上會產生額外的資源消耗(overhead)，而這些應用程式又擁有各自的效能目標(performance goal)。因此，在這篇論文中，我們希望能夠回答出下列問題：「在給定工作量與效能目標的情況下，讓一個應用程式在虛擬環境上運作需要分配多少資源？」我們側重在找出特定應用程式在實體與虛擬化環境中資源消耗的差異與關係，以利於在移轉應用程式至虛擬環境時能夠得到妥善的配置。 在為了能夠得知應用程式在虛擬環境中所消耗的資源，我們提出了一個模型的雛形來預測。在給定應用程式的效能目標以及工作量下，輸入應用程式在實體環境中資源消耗的參數，進而預測出該應用程式在虛擬環境中資源消耗的量。 此外，我們以串流服務做為主要研究的應用程式。我們嘗試藉著實驗找出客戶端數量、影片位元傳輸速率以及資源消耗間的關係。

As the development of cloud computing advances, increasing number of enterprises start to move their existing physical machine hosted applications to virtualized cloud environments. However, virtualization technology will cause extra overhead while running application in virtual environment. Thus, knowing about resource consumption of the application in virtualization environment becomes an important issue. Each application has own performance goal, which will be defined in service level agreement (SLA). According resource consumption and SLA of the application, we’d like to answer the following question: “How much resources are necessary to be allocated to the application in the virtualized environment given workload subject to its target performance goals?” In this thesis, we focus on finding out the relationship of resource consumption of the specific class of application running in physical and virtual environment. Hence, the application can be allocated appropriate resource while moving from physical into cloud. In order to know the resource consumption (i.e. CPU usage) of application running in virtual environment, we propose a miniature model to predict. By inputting the parameter matrix of workload, resource consumption in physical machine and performance goal, the model can predict the resource requirement of application in virtualization environment subject to performance goal. In this paper, we choose streaming service as our research application. We try to discover the relationship of number of sessions, bit rate of video and resource consumption. Through the experiment we find out a simple relationship between them.