分區定址動態繞徑演算法的開發與實踐在軟體定義網路之資料中心

Abstract

在雲端運算中，網路的變異及表現能力成為最主流的議題。軟體 定義網路的開發使得網路管理與繞徑更有效率且可以由使用者定義。 早先的動態繞徑演算法:流量搬遷在模擬軟體NS2 中已經表現出優 異的吞吐量和避免雍塞的能力。然而，流量搬遷實際應用於軟體定義 網路的架構下會出現多餘的延遲每當在修改交換器裡的流量表。因此 在流量搬遷的機制下，我們提出一個分區定址演算法應用於對稱的波 特蘭網路拓樸且應用外卡對應的繞境機制達成網路負載平衡及低耗 的交換器流量表修改。在實驗結果顯示我們的方法達到較高的網路吞 吐量相較於傳統的最短路徑生成且降低了70%的流量表修改相較於 早先的流量搬遷機制。因此，分區定址演算法被證明為實際可以增加 網路表現且降低流量表修改的演算法。

Network variation and performance become major issues in cloud computing. The Software Defined Network enable flow-level management makes routing more effective and user-defined in datacenter networks (DCNs).A prior dynamic routing mechanism named Flow-Migration shows a good result in improving throughput and avoiding congestion in simulation tool NS2. However, Flow-Migration applied in real SDN-based for cloud management often incurs extra overhead from modifying the flow tables in switches. Therefore, on the basis of Flow-Migration mechanism, we propose a novel address-partitioned wildcard routing (APWR) scheme, which leverages the symmetry of a Portland topology and applies wildcard matching to achieve network load balance and low cost for flow-table modification. Experimental results demonstrate that on a Portland 4-8-8 datacenter network, the proposed scheme achieves higher throughput over than traditional SPB does, and reduces 70% cost on flow-table modifications used by previous Flow-Migration mechanism. As a result, our APWR scheme is proven a practical algorithm for improving network performance and reducing flow-migration cost, simultaneously.