分散式軟體定義網路控制器之能源感知負載平衡

Abstract

有別於傳統網路，軟體定義網路(SDN)架構將交換器的資料層與控制層分離，如此可以讓網路管理者能夠透過外在的控制器有效管理網路中的資料流。使用單一控制器來管理SDN網路會有可擴展性及單一故障的問題。因此，許多論文開始針對分散式控制器去做探討研究。現有針對分散式控制器探討之論文，如EstiCon，提出動態負載平衡機制。然而，分散式SDN控制器的省電議題卻尚未被探討。本論文的主要貢獻為: (1)提出一個能源感知負載平衡機制(EALB)來平衡控制器之間的負載並且在低負載情況下將一些控制器轉入睡眠狀態以達省電功效。控制器最主要的負載來源為packet-in訊息，當要處理的packet-in訊息過多以致於控制器無法負荷時，它的負載就需要被轉移。即藉由將該控制器下的部分交換器轉移出去給其他控制器以達到負載平衡之目的。(2)此外，我們利用類神經網路預測是否能該將控制器轉進睡眠狀態以達省電功效。模擬結果證明，我們提出之具能源感知負載平衡演算法(EALB)能比不具能源感知負載平衡演算法(LB)省下約11%的能源消耗。對於控制器之間的負載平衡，EALB的負載平衡指標在packe-in 訊息率為300到900(訊息數/秒)時為1.19至1.4，僅些微高於LB (1.16至1.35)。

關鍵詞：能源感知、負載平衡、類神經網路預測、packet-in訊息、軟體定義網路。

Unlike the traditional network, the Software Defined Network (SDN) provides a way to control flows in a network by decoupling the control plane and data plane. Since an SDN network using a single controller to manage flows in the network may lead to SDN scalability and a single point of failure problems, some related work puts an emphasis on the issues of distributed SDN controllers. A related work on distributed SDN controllers or multiple SDN controllers, EstiCon, proposed a dynamic load balancing scheme. However, the issue of energy saving for distributed SDN controllers hasn’t been addressed. The main contributions of this paper are as follows. (1) We propose an Energy-Aware Load Balancing (EALB) algorithm for distributed SDN controllers. The proposed EALB balances the loads of distributed SDN controllers and turns some controllers into sleep mode under light load. The loads of SDN controllers are mainly caused by number of packet-in messages received. When packet-in messages become too many for a controller to handle, its loading needs to be migrated. When a controller becomes overloaded, the proposed EALB migrates some switches handled by the overloaded controller to other controllers in order to balance the loads. (2) In addition, we use a neural network based predictor to check whether it can turn a controller into sleep mode for energy saving. Simulation results show that the proposed EALB can save 11% of energy consumption compared to LB (EALB without energy saving). For load balancing of distributed SDN controllers, the load balancing metric (LBM) of the proposed EALB is from 1.19 to 1.4 under packet-in message rates from 300 to 900 (messages/second), which is only slightly higher than that (from 1.16 to 1.35) of LB.

Keywords: Energy-aware, load balancing, neural network-based prediction, packet-in message, software defined network.

Keywords: Energy-aware, Load balancing, neural network-based prediction, packet-in message, software defined network.