針對巨量資料處理所設計的一套了解應用需求的SDN繞徑機制

Abstract

隨著資訊技術日趨發達，欲處理的資料量也日趨增加，單一機器已不能負荷如此巨大的資料量，為了能夠有效率的儲存、處理這些資訊，佈署於雲端網路中的巨量資料應用程式已成為常見的解決方法。為了能夠儲存大量資料與進行平行運算，巨量資料應用通常需要透過網路與各方伺服器機換資訊，網路資源分配恰當與否將大大影響巨量資料應用的計算速度，而由於傳統網路的分散式架構，其難以理解應用程式的需求，使得對此特定應用的網路資源分配能力不足，從而降低巨量資料應用的計算速度，軟體定義網路概念的崛起，使得網路能夠簡單地與巨量資料應用緊密結合，其能夠依據應用程式需求分配網路資源，做到真正了解應用需求，加速巨量資料應用計算。 本論文中，我們以Hadoop做為巨量資料應用的例子、利用其運作時產生的網路流量特性與網路拓樸特徵，設計一套針對巨量資料的了解應用需求的軟體定義網路繞徑機制。

With the development of computer science technology, the amount of information becomes too large to be processed by a single machine. To store and process information efficiently, Big-Data applications deployed in cloud become the common solutions. To be able to store large data and do parallel computing, a Big-Data application usually communicates with servers through a network. The ability to efficiently allocate network resources will greatly affect the speed of computing for Big-Data applications. Because of the distributed architecture of legacy networks, it is hard to understand the needs of applications, and therefore it has little capacity to be application-aware. As a result, it is hard to accelerate the speed of computing. The rise of software-defined networking makes it easy for the network to integrate with a Big-Data application closely. The network can allocate network resources based on application requirements, which is application-aware, to accelerate the speed of Big-Data computing. In this thesis, we use Hadoop as a Big-Data application example, and use the traffic pattern generated while processing data and the characteristics of network topology to design an application-aware software-defined networking routing scheme for Big-Data processing.