在軟體定義網路上針對線路故障與擁塞的快速切換機制

Abstract

在軟體定義網路(SDN)架構上，交換器(switch)上的控制層集中至外部的控制器(controller)，因此控制器可以收集與維護網路情況(如：網路拓樸、交換器狀態、link狀態等)。控制器透過設定轉發表(flow table和group table)，控制每個資料流(flow)的路由決策。針對線路故障與壅塞問題，本論文設計兩個快速切換機制去維護SDN網路效能。第一個機制針對線路故障。控制器會定期針對每個用戶之間的資料流計算一條主要路徑與備用路徑，並設定到flow table與group table上。當線路故障時，交換器就能將受影響的資料流快速切換至備用路徑上。實驗結果顯示，透過此快速切換機制可以達到低延遲、0%的封包遺失率，並維持正常吞吐量。另外，第二個機制針對節點擁塞，控制器會定期偵測交換器上的每個通訊埠(port)。當交換器上的通訊埠發生擁塞時，控制器通知上游交換器切換一些資料流至備用路徑以消除壅塞情況。此機制上，我們設計四個參數(pt, threshold, Tdetect, α)讓網管人員對壅塞偵測進行調整。實驗結果中，當pt越小，threshold越小，T¬detect越短，α越大會產生較大的系統吞吐量。根據上述的結論，當控制器執行多個機制(快速切換機制針對線路故障與壅塞，和定期調整路徑)會產生更高的系統吞吐量。

In the SDN architecture, switches are only responsible for forwarding packets according to a set of rules, which are managed by a centralized software-based controller. The controller can gather the network situation via the OpenFlow protocol. It is easy for the controller to control the flow by setting flow table and group table. In this work, in order to handle link failure and link congestion, this work presents a fast failover mechanism and a fast switchover mechanism to improve the performance of SDN. In the fast failover mechanism, the controller predetermines active and backup paths between each couple of hosts, and sets flow table and group table to related switches. When a port on a switch becomes faulty, the affected flows are able to be efficiently switched to the backup path. The simulation results show the round trip time of the fast failover mechanism is significantly lower than that of the fast restoration mechanism when link failure occurs. The packet loss rate of the fast failover mechanism is almost 0%. The throughput remains stable. Therefore, the fast failover mechanism can efficiently handle link failure and provide a stable performance in SDN. In the fast switchover mechanism, the controller periodically detects the ports on the switches. When a port on a switch becomes congested, the controller adaptively decreases the sending rate of the upstream switch by switching some flows to the backup path. This mechanism can decrease the number of packet loss and increase the system throughput. In congestion detection, we design four parameters for network administrators to adopt different network environment. The performance evaluation shows that the higher throughput the system gets, as the small persistence interval is, the lower threshold is, the shorter interval of detection is and the larger proportion of raw drop ratio is. According to above conclusion, the system yields higher throughput when the controller runs multi-mechanism (fast failover mechanism, fast switchover mechanism, and periodically adjust path).