一個虛擬化的雲端平台可能同時有多個虛擬機器同時對外提供雲端服務，然而基於實體網路環境的限制，這些虛擬化的雲端服務會共享網路頻寬，當某特定服務突然面臨極高網路負載流量，就會影響平台上其它服務之品質。針對這個問題，我們可以使用動態頻寬分配的技術來限制各個應用程式與網路服務的流量。然而，一個雲端服務可能會使用兩台以上的虛擬機所構成的虛擬叢集來提供服務，因此動態頻寬分配需要做到虛擬叢集層級的頻寬使用管理。本論文主要提出一個虛擬叢集層級的網路QoS管理技術。這個技術是利用一邏輯集中的QoS控制器來動態分配頻寬，再透過每個節點上的網路流量限制元件 (Rate Limiter) 控制節點網路流量，讓管理者以虛擬叢集為單位監控叢集內的跨實體機虛擬機器網路連線，簡化監控的程序。當叢集內的跨實體機連線超過預設的臨界點時，系統可自動依據需求動態加入頻寬管制機制，使平均流量低的網路連線的頻寬保證有較高的優先權。我們將這個技術實作在以OpenStack為基礎的雲端平台SAMEVED-Stack上，並提供幾種不同的頻寬分配機制，最後透過模擬實際串流資料進行實驗來觀察這個技術的效益。我們發現，本研究所提出的網路服務品質動態頻寬分配機制在高網路負載的情境下，針對影音服務等較ㄧ般的使用情形，可有效降低近三分之一的封包遺失率，因此可避免特定的巨量資料串流(如DDoS網路實驗虛擬叢集)搶占頻寬而影響整個系統提供的網路服務。;A virtualized cloud platform may host several cloud services/applications simultaneously, each of which may use a virtual cluster consisting of several virtual machines (VMs) with a virtual network. Due to the constraint of the physical network environment, some sets of the virtual cluster may share network bandwidth. Thus, the quality of the cloud services, which share the same physical networking resources, can be affected when one of the cloud services produces high traffic load. To this end, we propose a mechanism that can dynamically allocate network bandwidth to virtual clusters in this paper. In the proposed mechanism, a logically centralized QoS controller is responsible for allocating network bandwidth to virtual clusters dynamically. The QoS controller uses a monitor to continuously gather network usage of each VM, and calculates how bandwidth is consumed in the granule of virtual clusters. A rate limiter is located at each physical machine to trigger bandwidth re-distribution. As soon as a VM produces a network flow that exceeds a predefined bandwidth threshold, the rate limiter then notifies the QoS controller to do bandwidth re-distribution. We have implemented the proposed mechanism on an OpenStack-based cloud platform, SAMEVED-Stack, which is developed to support on-line network security experiments. We also implemented several network bandwidth re-distribution strategies on the proposed mechanism. Through the experiments of emulating real network flows in the platform, we find that the proposed approach can effectively reduce nearly one third of pack loss rate when running a video streaming service in a high network traffic scenario.
