基礎架構服務(Infrastructure-as-a-Service)是雲端服務的重要類型之一，它使用逐漸成熟的虛擬化技術(Virtualization Technology)，將傳統的硬體資源以軟體的方式虛擬化呈現與整合多個虛擬機器(Virtual Machine) ，來運行服務或提供使用者使用，以此提高實體機器(Physical Machine)資源的使用效率與彈性。除了虛擬機器的虛擬化技術外，網路虛擬化是近年來用來強化IaaS的技術，可提供使用者設定自己所需的網路環境，並且建立自己所需的虛擬叢集(Virtual Cluster)或是虛擬機房(Virtual Datacenter)。然而，利用人工進行虛擬機房的建置與管理會造成效率的問題，也不適合管理大量動態產生與關閉的虛擬機房。因此我們研發虛擬機房的管理軟體SAMEVEDStackV2，來解決虛擬叢集與虛擬機房的資源配置與管理問題。SAMEVEDStackV2是由開放原始碼專案OpenStack所延伸發展的系統，它利用我們發展的虛擬機房監控機制來擷取虛擬叢集的資源使用量，做為系統資源配置的參考。此外，我們利用虛擬機房上的應用程式的特性去設計新的排程演算法，使得虛擬叢集配置時可以考慮實體機器負載，讓實體網路的負載達到最小化可能。我們透過多種實驗案例的測試與調整，發現我們的佈署策略在多組虛擬叢集的運算情況下可讓實體網路的負載大幅降低，但對於CPU使用量卻只有少量的影響。;Infrastructure-as-a-Service (IaaS) is an important type of cloud service. It uses server virtualization technology to provide users the software-based virtual machines as the computing power over the physical computing resources in a datacenter, such that the physical computing resources can achieve better utilization rate and manageability. The network virtualization technology is another key technology of IaaS, which can create virtual network environments on top of the physical network environment. With server virtualization and network virtualization, the IaaS users are able to create user-defined virtual clusters/datacenter. The problem is that, manual resources allocation and management for virtual clusters/datacenters is not practical because setting up virtual clusters/datacenters is time-consuming. In addition, the virtual cluster/datacenter placement problem is challenging because we need to consider the constraints of physical resources as well as the relationship among the virtual resources. To solve these problems, we have developed an IaaS management system, SAMEVEDStackV2, based on an open-source software package, OpenStack. The SAMEVEDStackV2 supports a high-level API for virtual clusters/datacenters management, and uses a monitoring/profiling mechanism for tracking and classifying virtual clusters/datacenters. The scheduler uses the information of virtual clusters/datacenters to reduce unnecessary network bandwidth consumption while placing virtual cluster/datacenters on physical computing resources. Our preliminary experimental results show that, the proposed placement strategy can reduce 37% of the physical network bandwidth consumption when comparing with the default OpenStack scheduler.
