多網域軟體定義網路之快速控制器故障切換機制

Abstract

軟體定義網路(SDN)是一種將資料層和控制層分離的網路架構，而OpenFlow被認為是最受歡迎的SDN協定。對於軟體定義網路來說，只使用單一控制器可能會有擴展性和單點故障(SPOF)的問題。因此，多軟體定義網路控制器的概念被提出以解決以上問題。現有的多軟體定義網路控制器使用固定的逾時時間來檢查控制器是否故障，但固定的逾時時間很難預設，而且其長度會影響故障偵測的時間。此外，大部分的相關文獻，如Devolved controller和HyperFlow，只選擇任意一台替代控制器去取代故障控制器。它們沒有考慮替代控制器的負載及距離，如此可能會導致替代控制器的過載。因此，我們提出一個多網域軟體定義網路之快速控制器切換機制(FCF-M)去處理SDN控制器故障的問題，我們的機制也解決了錯誤偵測的單點故障的問題。FCF-M由控制器故障偵測及交換器移轉兩個模組組成。在控制器故障偵測的部分，每台控制器檢查它前面控制器的時間戳記。如果某一台控制器故障，它後面的控制器將會接替它檢查它前面的控制器。由於故障控制器並不妨礙它前面的控制器之健康檢查，因此FCF-M能夠防止健康檢查的單點故障。此外，我們也善用調整式的逾時時間去縮減故障偵測時間。至於交換器移轉的部分，當控制器故障發生時，我們需要重新配置在故障控制器下的交換器給替代控制器。我們考慮控制器負載與交換器和控制器之間的距離去選擇替代控制器，以減少延遲。我們所提出的FCF-M在模擬中考慮三種故障案例，並且可以達成網域內的控制器故障切換和網域間的控制器故障切換。此外，FCF-M仍然遵守網域內的Master和Slave控制器之間傳統上的故障切換方式。模擬結果顯示，第三種故障案例中的FCF-M (FCF-M-3)比Devolved controller減少6%的故障切換時間。此外，FCF-M-3比Devolved controller減少6%的平均端對端延遲。

Software defined networking (SDN) is a network architecture which decouples the data plane and control plane, and OpenFlow is considered as the most popular SDN protocol. For SDN networks, using only a single controller may have SDN scalability and a single point of failure (SPOF) problems. Therefore, multiple SDN controller architecture has been proposed to overcome these problems. Failure detection of most of existing multiple SDN controllers uses a fixed timeout delay to check whether a controller is failed; however, it is difficult to set a fixed timeout delay and its length may affect failure detection time. In addition, related works, such as the Devolved controller and HyperFlow, only choose an arbitrary alternative controller to replace a failed controller. They did not consider the load and distance of the alternative controller, which may result in overloading of the alternative controller. Thus, we propose a Fast Controller Failover for Multi-domain SDNs (FCF-M) to handle SDN controller failures and there is no SPOF for healthy check. The proposed FCF-M is composed of controller failure detection and switch reassignment. In the controller failure detection, each controller checks the timestamp of its preceding controller. If a controller is failed, its succeeding controller will check its preceding controller. Because a failed controller does not prevent its preceding controller from healthy check, the proposed FCF-M can prevent an SPOF for healthy check. In addition, we also make use of an adaptive timeout delay to reduce failure detection time. In the switch reassignment, when a controller failure occurs, we need to reassign the switches under the failed controller to an alternative controller. We consider controller load and distance between a switch and a controller to choose an alternative controller for reducing delay. The proposed FCF-M considers three failure cases, and can achieve intra-domain controller failover and inter-domain controller failover. In addition, the proposed FCF-M still conforms the conventional failover mechanism between master and slave controllers. Simulation results show that the proposed FCF-M for failure case 3 (FCF-M-3) can reduce 6% of failover time compared to the Devolved controller. In addition, FCF-M-3 can reduce 6% of average end-to-end delay compared to the Devolved controller.