透過您的圖書館登入
IP:3.22.248.208
  • 學位論文

資料中心中軟體定義網路的快速錯誤回復機制

Fast Failover Mechanism for SDN-enabled Data Centers

指導教授 : 黃仁竑
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


資料中心網路隨著服務及用戶的增加而日趨龐大而複雜,僅憑傳統的網路技術將衍伸出許多議題,而軟體定義網路(SDN)之概念的提出,可望為此帶來新的可能,如快速錯誤偵測與回復、網路虛擬化、高吞吐率、及負載平衡等。在網路的管理中,許多策略會依據拓樸來決定,因此即時監控拓樸的狀態相當重要一環。此外,為了確保網路使用者的服務品質,快速修復網路中的鏈結錯誤以維持服務的運行亦是備受重視的項目。因此本論文將以保障用戶之服務的可靠性為核心,設計適用於大型資料中心網路之快速錯誤回復機制。 本論文考量資料層以及控制層,提出一個適用於支援多租戶之軟體定義網路資料中心的快速錯誤回復機制,實作自動拓撲發現機制來獲得網路拓撲,及提出自動拓撲偵測機制以即時監控網路狀態,並於網路鏈結錯誤發生時,觸發錯誤回復機制。在資料層的錯誤回復方面,我們分別使用Protection及Restoration策略實作,並基於拓撲中的完全二分圖來設定備用路徑。由於考量in-band網路,因此提出基於Restoration策略的控制層錯誤回復機制,根據可用網路介面的權重來重建control channel。此外,我們亦提出支援多租戶的封包隔離機制。在效能分析的部分,我們使用實體環境及模擬環境中進行測試,根據實驗結果顯示我們提出的錯誤回復機制能使錯誤在100毫秒內被排除,且使用Protection策略進行資料層的錯誤回復可獲得較佳的實驗結果。此外,我們亦進行可擴性評估,包括計算Flow Entry數以及演算法時間複雜度。

並列摘要


In this thesis, we present mechanisms for fast failover of both control and data channels in SDN-enabled multi-tenant data centers. An auto-topology discovery and auto-topology monitoring mechanisms are aim to obtain the initial topology and keep tracing the state of network components. The recovery mechanism is launched when a link failure occurs. For the data channel recovery, both restoration and protection mechanisms are proposed. The SDN controller computes backup paths based on complete bipartite graphs of the network topology and configures flow entries of SDN switches according to both mechanisms, respectively. For the control channel recovery, we assume that the SDN network adopts in-band control. As a consequence, restoration mechanism is adopted which re-builds the control path for switches losing control channel based on a weighting function. In addition, a novel traffic isolation mechanism is proposed for multi-tenancy. We evaluate the proposed mechanisms via a real network as well as Mininet simulations. Our numerical results show that, for data channel recovery, the average recovery times for both restoration and protection mechanisms are less than 100ms and the protection mechanism provides a faster recovery time. We also address the scalability issues, in terms of the size of flow table and the time complexity, of the proposed mechanisms.

參考文獻


[29] T. K. Chiueh, C. C. Tu, Y. C. Wang, P. W. Wang, K. W. Li, and Y. M. Huang, “Peregrine: An All-Layer-2 Container Computer Network,” in Proc. ICC, 2012. pp. 686-693.
[3] M. Al-Fares, A. Loukissas, and A. Vahdat, “A Scalable, Commodity, Data Center Network Architecture," in Proc. ACM SIGCOMM, 2008, pp. 63-74.
[8] F. Pakzad, M. Portmann, W. L.Tan, and J. Indulska,”Efficient Topology Discovery in Software Defined Networks,” in Proc. ICSPCS, Gold Coast, QLD, 15-17 Dec. 2014. pp. 1-8.
[10] Sharma, S., Staessens, D., Colle, D., Pickavet, M., & Demeester, P., “OpenFlow: Meeting carrier-grade recovery requirements.”in Proc. ICC, 2013, pp. 656–665.
[12] J.-P. Vasseur, M. Pickavet, and P. Demeester. “Network Recovery: Protection and Restoration of Optical, SONET-SDH, IP, and MPLS,” in Morgan Kaufmann Publishers Inc., 2004.

延伸閱讀