高效能巨量資料分析系統之關鍵技術研發及其在電信流量管理之應用---子計畫三：高效能巨量資料分析系統之軟體定義網路技術研發及其在電信流量管理之應用

Abstract

隨著在雲端資料中心內運行的巨量資料應用程式越來越多，資料中心內交換器之間 傳送的封包數量也會大幅增加。此現象會導致部分交換器之間的路徑擁塞，因而降低資 料中心整體之效能。為解決上述問題，本子計畫將探討新一代的軟體定義網路 (Software-Defined Networking, SDN)架構。SDN 藉著將交換器資料層與控制層分離的技 術，進一步將雲端資料中心內的網路也虛擬化。透過我們設計的基於需求表之應用感知 SDN 最佳路徑演算法，當有多重路徑可選擇時，可以在路徑頻寬及節點(如交換器)資源 有限下，預先分配及動態調整網路資源，使得雲端資料中心網路在高負載或有故障的元 件下，依然能確保每一個應用程式(含巨量資料應用程式)或用戶，能滿足其所需的網路 服務品質及應用程式之服務水準協議。此外，我們會將SDN 的研究從雲端資料中心網 路延伸至電信核心網路上。我們也將研究何種資料中心之實體網路拓樸，最適合應用 SDN。 本子計畫將利用子計畫五及六所提供的雲端資料中心網路(或電信核心網路)流量事 件處理及資料探勘結果，供SDN 控制器對交換器做存取控制，如更改其流量表，以達 到資料中心網路(或電信核心網路)路徑最佳化，以避免網路壅塞及使用者體驗品質下降 的情況。另外，本子計畫將支援子計畫一、二及四所需之高效能網路存取。本子計畫預 計在三年內完成。在第一年中，我們將廣泛的進行相關文獻的探討，提出適用於雲端資 料中心的應用感知SDN 之路徑最佳化方法，並評估不同的資料中心網路拓樸對SDN 的 適用性。我們將建構資料中心SDN 測試平台，利用Hadoop 讓巨量資料應用程式平行運 作在此平台上，並作效能評估。在第二年中，我們將設計適用於電信核心網路的應用感 知SDN 之路徑最佳化演算法、建構SDN 與電信網路的整合測試平台，以及與其他子計 畫做初步的整合。在第三年中，我們將電信核心網路的SDN 路徑最佳化演算法佈署到 電信網路測試平台，並做效能評估。同時我們也將研究多個控制器之協同運作機制，以 解決SDN 之延展性(scalability)問題。最後，我們將完成與其他子計畫的整合。本子計畫 的預期研究成果將能夠協助雲端資料中心及電信業者設計適用於自身實體網路拓樸的 應用感知SDN 網路，以避免網路瓶頸，從而提高雲端資料中心網路及電信核心網路的 延展性、可靠性及運作效能。

With the increase of big data applications in the cloud data center, the number of packets transmitted between switches greatly increases. It may result in congestion of data paths between certain switches and reduce the overall performance of the cloud data center. To resolve this problem, this subproject will study software defined networking (SDN), which is a form of network virtualization that separates the control plane from the data plane. By the proposed demand table based application-aware SDN path optimization algorithm, when there are multiple paths available, we may pre-allocate and dynamically adjust network resources under the constraints of path bandwidth and nodes (such as switches) resources. In this way, when the data center is overloaded or has failed components, running applications can still meet their service level agreements. In addition, we will extend our SDN study from the cloud data center to the telecommunication core network. We will also study which data center network topology is the best fit for SDN. This subproject will utilize the traffic event processing and data mining results provided by subprojects 5 and 6 for SDN controllers to update the flow tables of related switches so as to achieve path optimization and avoid network congestion and poor user experience in the cloud data center (or telecommunication core network). This subproject will support subprojects 1, 2, and 4 to have efficient network access. This subproject is expected to be completed within three years. In the first year, we will review related work on SDN path optimization. We will propose an efficient application-aware SDN path optimization approach and evaluate SDN feasibility of different network topologies in a cloud data center. We will set up an SDN testbed in a cloud data center and run Hadoop-enabled big data applications to evaluate our approach. In the second year, we will extend our application-aware SDN path optimization approach to a telecommunication core network. Then, we will begin to integrate our subproject with the other subprojects. In the third year, we will deploy our application-aware SDN path optimization algorithm to a telecommunication network testbed and evaluate its performance. In addition, to deal with SDN scalability, we will propose a multiple SDN controllers collaboration mechanism. In summary, the expected results of this subproject can greatly enhance the feasibility of realizing an application-aware SDN-enabled network topology for cloud data centers and telecommunication core networks so as to avoid network congestion and enhance network scalability, reliability, and performance.