| 摘要: | 在目前網路的環境中，不論是一般上網(Internet)、多媒體服務(Media)、企業內部網路 (Corporate)...等等，皆會透過設立許多可提供服務內容的存取伺服器來達到地域性的備援與 負載平衡(Load Balancing)，但礙於各伺服器的負載狀況是彼此獨立的，當端對端品質產生異 常時，傳統網路架構無法容易的在當下判斷是否為服務內容提供者的伺服器異常，又或是在 傳輸電路路徑當中有壅塞等之類的事件發生，導致問題排除的時間不容易掌控。
對於如何有效解決既有網路中的品質判斷問題，軟體定義網路(SDN, Software Defined Network)的可程式化管理方式可以解決此類問題。有別於傳統網路的分散架構，SDN 將網路 劃分成控制層(Control Plane)與資料層(Data Plane)，透過程式化的方式對控制器(Controller)進 行控制層的運作，將各個獨立的網路設備，集中管理設備的路由規則表(Rule tables)，而交換 器(Switch)僅需負責資料層的封包傳輸。
本研究將使用 SDN 可程式化作為實驗的方向，透過控制器監控著端對端的封包回應狀況， 統一回傳給資料蒐集中心(Collector)進行所有控制器之間的數據共識決策與判斷，將達成共識 的高負載，透過控制器修改 Open vSwitch 的路由表，把流量轉移至負載較輕的路徑，並經由 實驗三個測試環境，分別為傳統方式(Traditional Base)、最長延遲路由調整(Longest Delay Reroute)以及本研究提出的共識決策(Consensus Base)，搭配 KVM 模擬器進行以上三種環境進 行效能分析。而經由實驗結果顯示，本研究提出的共識決策能夠合理的判斷端對端的負載狀 況，將高負載的路徑從 Open vSwitch 中修改成低負載的路徑，使整體端對端的平均回應時間 (Avg Time)有效降低，讓低負載的伺服器之資源有效地利用。;In the current Internet environment, all the geographic redundancy and load balancing, including Internet, media, corporate...etc, can be achieved by establishing servers that are able to provide service content. However, in view of the independence of server loading status, it is difficult to instantly determine whether it is a server malfunction from the content provider, or it is a circuit congestion occurs in the transmitting route, which result in the time uncertainty of trouble shooting under the traditional Internet structure.
To efficiently address the problems that happen when determining the quality of existing network, the programmable management of SDN(Software Defined Network) is the solution. Different from the traditional decentralized network, the network of SDN consist of Control Plane and Data Plane, operating the Controller in the Control Plane by programming it. SDN also centralizes the management of all rule tables from each independent network devices — Switch, which is only responsible for the package transmission.
The study takes the programmable SDN as the direction of the experiment. It monitors the end- to-end packet response status through the Controller, and send the data back to the central data collector to process the consensus decision-making within all controllers. Then, the result of the process will be sent to the controllers, further allowing them to modify the routing table of Open vSwitch and to transfer the flow from the determined heavy loading route to the one with lower loading. The performance analysis is conducted with using KVM simulator under three different testing environments, including Traditional Base, Longest Delay Reroute, and Consensus Base suggested from the study.
The result of the experiment shows that the Consensus base suggested by the study can correctly determine the loading status, modifying the route in Open vSwitch from heavy loading path to lowest loading path. It significantly reduced the overall responding time of the packet, and further utilized the resources from the low loading servers efficiently. |
