A Study of Hybrid Network Load Balancing Architecture in Cloud Environment

近年來，雲端運算已經成為一個非常熱門的話題。雲端運算以網路為基礎，將網路資料中心(Internet Data Center, IDC)的運算資源(電腦和其他設備、 軟體和資訊等資源)虛擬化，再依照需求動態分配使用者。因此，企業資訊人員與IDC業者，在規劃與建置雲端運算服務時，除了成本考量因素外，資源與服務虛擬化之效能，更是重要考量的因素。 由於雲端運算服務的低維護成本和易取得的優點，讓越來越多的使用者將 Web 服務移轉到雲端環境。為了提升Web服務的高可用性，系統管理者大多會結合Web Cluster和硬體式網路負載平衡器；然而，硬體式的網路負載平衡器除了價位高且不易擴展外，同時也必須與Web Cluster環境放置在一起，地域性也受到限制。 本研究透過文獻探討並以Open Source作為基礎，提出一個可在雲端環境中動態配置的虛擬化網路負載器系統架構，再以系統雛形方式建立實做環境，進一步以實驗方式收集結果與分析其效能，驗證本研究提出的架構可行性。 為考量網路負載平衡器擴展性的需求，本研究利用免費的LVS （Linux Virtual System）來建置軟體式的網路負載平衡器，並建置在雲端環境上以虛擬機器方式提供全域性且高可用性的負載平衡系統服務。我們設計出一種兩層式負載平衡架構(Cloud Hybrid Load Balancer, CHLB)，可運用兩種或以上不同的負載平衡方法來處理使用者的需求，同時也可滿足不同雲端服務間的負載平衡需求。 在本研究中，為了評估與驗證CHLB之績效，我們同時發展虛擬化效能的評估方法，並針對虛擬化應用系統效能，影響實體與虛擬主機效能比較，以及不同虛擬化平台軟體的效能比較進行實驗分析;結果說明不同應用系統虛擬化後效能與實體上的效能高度相似，而在不同的虛擬平台上效能也差異很小。本研究並以上述評估方法針對所提出的雲端複合式負載平衡器，對網站系統的服務效能和單層網路負載平衡架構效能進行實驗比較;主要針對網站服務的回應時間和處理需求數進行實驗測量。由於實驗資源限制，我們模擬兩個雲端平台，架構兩個網路資源不同的情境，並選擇了輪替（Round Robin）和加權最少連接(Weight Least Connection) 兩種常用的負載平衡方法。利用這兩種方法對單層和兩層的負載平衡架構進行效能比較。實驗結果顯示，本研究提出的雲端複合式負載平衡器（CHLB）在固定時間內能處理較多47.96%的需求數，而處理相同需求數時所需的回應時間較少33.08%，驗證出本研究的架構能協助提供更佳的服務效能。 本研究結果顯示CHLB在雲端環境中確是一個可行且低價的全區域性網路負載平衡器。未來可進一步探討如何提升雲端複合式負載平衡器效能，和比較其他不同的負載平衡方法影響。同時為了有效運用運算資源，未來也將研究如何結合虛擬平台的資源管理，進一步動態配置雲端複合式負載平衡架構中所需的虛擬主機資源。Cloud Computing has become a very hot topics for the past few years. Cloud computing is based on the Internet that dynamically allocates resources to the consumers with virtualization of Internet Data Center (IDC) computing resources such as computation computers, software information and other peripherals. Therefore Information Technology (IT) staffs and IDC companies should consider the resource and the performance of service virtualization besides the cost factor while planning and establishing the cloud services. With the lower cost and convenience of cloud computing services, users have increasingly put their Web resources and information in the cloud environment. In order to enhance high availability of the important web services, the system administrators usually combine Web Clusters with hardware-based network load balancers. However, the hardware-based network load balancers are expensive and difficult to expand. There is a regional limitation which the hardware-based network load balancer must be placed together with a Web Cluster environment. This research adopts the method of literature review. We propose a multiple-layer virtualized network load balancing architecture which is based on Open Source and can be dynamically deployed in cloud environments. Next we applied the system prototyping method and we implemented the testing environment. Further, we made experiments and collected the results to analyze the performance of our proposed framework and to verify the feasibility of the architecture. This study applies the free Linux Virtual System (LVS) for implementation of the software-based network load balancer for the scalability of load balancers and put the system as a virtual machine in the cloud environment to provide a global and high available load balancing service. We propose a kind of two-level load balancing framework (Cloud Hybrid Load Balancer, CHLB), the framework can be applied with two or more load balancing methods for handling with requests from users. It can fit the load balancing requirements between different cloud services. To evaluate and verify the performance of CHLB, we develop the evaluation method of virtualization performance. And to focus on the virtualized application performance, we make experiments and analysis about the performance emphasis between physical machines and virtual machines, and the performance comparison of different hypervisors in this research. According to the testing results, the performance results of different workloads after virtualization are highly similar. There are very small differences between the performances of different hypervisors. We adopt the above evaluation method and focus on the proposed CHLB and to test the performance of the web services in CHLB and the one with single-level network load balancer. We focus on the experiments and the measurement of the response time and the finished requests. Because the limitation of the experiment resources, we simulate two cloud environments with different network resources and we select two common load balancing methods, which are Round Robin and Weight Least Connection. Single load balancer compared with two-level load balancer by these two methods. The experimental results show that the proposed CHLB in this study can finish more 47.96% requests within the same time period and the response time taken by processing the same requests is more less 33.08% than the one of the single load balancer. These results verify that the framework of this study can help provide better service performance. Although this research results show that CHLB is feasible and low-cost global network load balancer in cloud environments. In the future, we can further explore how to improve the effectiveness of cloud hybrid load balancer and comparison with other methods on different load balancing methods. In order to efficiently use computing resources, in the future we will also study how to combine the resource management of the hypervisor. Further it can be dynamically deployed the required virtual machines of CHLB