分散式軟件定義網絡控制平台的負載自適應控制器容錯機制

Abstract

軟件定義網絡是一個興起的技術,它通過將網路中控制平面和傳輸 平面分離開來,實現一個集中的具有全局觀的的控制平面架構。通過 實現具有全局觀的集中化的控制平面,軟件定義網絡使得網絡能夠利 用運行在控制平面上的應用程序進行編程式的管理,這不但使得網絡 的管理變得更加容易,同時也是的網絡創新的週期變得更短。然而, 當今的數據中心的規模越來越大,依靠單一控制器進行管理變得不可 能。因此,能夠提供更好的可靠性和可擴展性的在邏輯上集中的,但 是物理上分佈的多控制器平面變得非常必要。 為了能夠保持多控制器平面在其中某個控制器停止運行的時候能夠持續正常工作,現有的方法通過實一致性領導選擇機制來保證每個交換器在同一時刻有且僅有一個控制器對其進行主控制,與此同時將每個控制器的本地狀態保持備份,以保證交換器變換控制器之後正常運行。本篇論文說明了現有方法的不足,同時提出了一個在軟件定義網絡中的負載自適應的多控制器容錯機制。控制器和對應的交換機的負載信息被收集用於決定控制器停止運行時交換機的重新連接。狀態分割的概念被使用保證在控制器恢復時的狀態同步。一個機制的原型也被實作,并得到令人滿意的實驗結果。

Software-Defined Networking(SDN) is an emerging technique that decouples network into centralized control plane and distributed data plane. By achieving a logically centralized control plane with global network view, SDN allows network to be managed by applications that run on the centralized control plane, which enables easier management and faster innovation. However, many of today’s data center network consists of huge number of device which can’t be controlled by a single controller instance. Therefore, a logically centralized, but physically distributed SDN control platform that provides better scalability and reliability is necessary. In order to keep control platform working during controller failure, most existing solutions enable a consensus based leader election that is performed for each switch to ensure at most one controller instance is in charge for each switch, meanwhile local state is backed up among controllers. This thesis presents the insufficient of existing solutions and proposes a load adaptive controller fail-over mechanism for distributed control platform in SDN. Load measurement of controller and its switches are collected to determine the process of switch reconnection when controller failure happens. Concept of state partition is used to ensure state consistency during controller fail-over. A prototype of our mechanism is also implemented, and the result of experiment shows desired results.