基於虛擬網路通訊強度之雲端虛擬叢集排程策略

雲端運算近年來的發展越趨成熟，除了帶給人們生活上更多的便利，也改變了傳統應用程式的架構和提供軟體服務的模式。然而，在不為一般人所知的背後，這些雲端計算資源的管理與分配正是驅動軟體服務架構改變，並讓人們生活更加便利。也因此，雲端環境中的資源分配問題是值得研究的議題，尤其針對由許多虛擬機器與虛擬網路所構成的虛擬叢集 (Virtual Cluster)資源分配問題更是具有挑戰性。現今的雲端平台資源排程器大多將這些屬於相同虛擬叢集的虛擬機器視為單獨的虛擬機器而無法察覺他們之間的關聯，因此在資源使用上可能比較沒有效率。此外，虛擬叢集的資源管理機制需要去解決一個重要的問題，也就是當平台上的虛擬叢集網路流量過大時，平台所提供之服務的品質會嚴重被影響。因此，本實驗開發了新一代的虛擬叢集排程器─ Insight Scheduler來改善虛擬叢集使用過多網路頻寬時導致服務品質降低的情形並增加資源分配的效率。Insight Scheduler透過資源監控與分析找出虛擬叢集的資源使用特性，再利用這些資訊對虛擬叢集的網路架構作優化，最終達成降平台整體網路流量以及更佳的資源分配效率等兩個目標。我們透過多個實驗並輔以數據、圖表，證實我們的虛擬叢集排程機制可優化大網路流量之虛擬叢集資源配置以達到降低網路流量達85%，與增加資源配置效率16%。

關鍵字

雲端計算；虛擬叢集；排程機制

並列摘要

The cloud computing paradigm becomes increasingly popular as more and more applications and services, running on various cloud data centers, are provided based on the paradigm. The use of cloud computing technology provides more flexibility and convenience to people’s daily lives, even though the cloud users do not know how the enabling technologies work in a datacenter. One of the key enabling technologies is virtualization, which enables the use of virtual machines with virtual networks running on physical machines. With server virtualization and network virtualization, users are able to create user-defined virtual clusters to host their applications. In most cases, some VMs of a virtual cluster may frequently access each other frequently. As a result, many virtual clusters are communication-intensive in practice. The problem of virtual cluster placement is considered more complicated than VM placement since the communication-intensiveness of VMs has to be considered. In addition, existing VM placement (or scheduling) mechanism cannot handle this problem well. In this study, we have to proposed a new virtual cluster placement strategy, namely the Insight Scheduler. The proposed scheduler not only lowers the bandwidth consumption caused by the virtual clusters, but also achieves efficient resource provisioning. The proposed scheduler relies on a monitor to collect resource usage of a virtual cluster, as well as network consumption of each VM. Then it uses a profiler to classify the types of VMs. Finally the scheduler uses the processed information to place VMs on physical machines. Based on our experimental results, the Insight Scheduler can reduce the physical network load by 85%, and increase application efficiency by 16%.

並列關鍵字

Cloud computing ； OpenStack ； Virtual Cluster Placement ； Scheduling

參考文獻

[3] I. Foster, Yong Zhao, I. Raicu, and Shiyong Lu, “Cloud Computing and Grid Computing 360-Degree Compared,” Grid Comput. Environ. Workshop 2008 GCE 08, pp. 1–10, Nov. 2008.

[4] P. Mell and T. Grance, “The NIST definition of cloud computing,” National Institute of Standards and Technology, Gaithersburg,Maryland, Special Publication 800-145, 2011.

[5] R. P. Goldberg, “Survey of virtual machine research,” Computer, vol. 7, no. 9, pp. 34–45, 1974.

[8] Al Muller, Seburn Wilson, Don Happe, Gary J. Humphrey, and Ralph Troupe, Virtualization with VMware ESX Server. Syngress, 2005.

[15] B. A. A. Nunes, M. Mendonca, X.-N. Nguyen, K. Obraczka, and T. Turletti, “A Survey of Software-Defined Networking: Past, Present, and Future of Programmable Networks,” IEEE Commun. Surv. Tutor. Rev.

國際替代計量

基於虛擬網路通訊強度之雲端虛擬叢集排程策略

未授權

主題瀏覽