架設雲端資料中心的目的就是要提供多項的服務，而這些服務未來會使用越來越多 的網路資源。因此要如何有效的分配在雲端資料中心的網路資源，變成一項重要的問題。 以往最常被使用的 Dijkstra’s 演算法，是使用在普遍的網路拓樸，但是雲端資料中心的 拓樸以及資料流的特性與普遍的網路有別，因此需要為了雲端資料中心設計一個路徑演 算法來解決頻寬的分配。 我們設計出了 pairing 演算法，將以往大部分是從一個節點開始尋找路徑，變成從 兩個節點出發一起尋找。由於雲端資料中心為了能夠讓大量的伺服器之間快速且可靠的 傳輸訊息，因此雲端資料中心的網路拓樸都有很多條路徑而且是被經過設計的。根據這 些設計的規律，從兩個節點一起尋找路徑不但不用擔心在尋找的過程中，這兩個節點可 能永遠不會連結起來而成為一條真正的路徑，而且還可以同時根據兩個節點所看到的網 路狀況，尋找真正適合的路徑。除了路徑演算法的部分，我們還提出了兩種分流的方式， 讓資料流有更多的機會使用更多的頻寬，減少資料流被擋住的可能性。 由模擬結果可知，本論文提出之方法，可以有效的減少資料流被完全阻擋的比例， 而且滿足率超過 80% 但不包括完全滿足的資料流比例也比較多。這是因為我們有效的 分配瓶頸的鏈結。同時也可以看到，當網路越壅塞時，改善的幅度是越來越明顯。因此， 我們有效的利用網路資源，讓比較多的資料流能夠建立連線，也讓比較多的資料流能夠 有更高的滿足率。;The purpose of deploying cloud data center is to provide different kinds of services. However, these services would require more and more network resources in the future. Therefore, the efficiency of allocating network resources in cloud data center has become an important issue. Dijkstra’s algorithm is one of the most used routing algorithm for general networks. But the topology and traffic pattern in cloud data center are different from the general networks, so it’s essential to design a routing algorithm for cloud data center to provide a solution for allocating the network resources. We designed the pairing algorithm, which searches a path from two nodes instead of one. In order to transfer data among massive servers with guaranteed speed and reliability, the network topology is especially designed with multiple path. It’s able to search a path according to the pattern of the topology from two nodes and don’t need to concern that these two nodes will not be able to meet to provide the path. This makes pairing algorithm allowed to discover a path from two aspects simultaneously as well. Besides routing algorithm, we also proposed two method of splitting flows, which allows the flows to use more network resources and reduces the chance of a flow to be blocked. According to the simulation result, the pairing algorithm and split with allocation constraint is able to reduce the ratio of complete blocking effectively. Also, the ratio of flows satisfied over 80% without fully satisfied is higher. Therefore, we allocate the network resource more efficiently, not only more flows can be established, but also more flows achieve over 80% of satisfaction rate.
