在軟體定義網路下雲端資料中心之應用程式感知路由機制

Abstract

隨著在雲端資料中心運行的應用程式越來越流行，在目前的網路中，不同類型的應用程式的網路需求不同，若是平等地對待所有類型的應用程式不是一個合適的方法。因此，在網路資源有限的雲端資料中心中，要滿足不同類型的應用程式的網路需求，應用程式感知路由方法是一種被看好的方法。軟體定義網路(Software-Defined Networking, SDN)提供了一個使我們能夠藉由一個外在可編程的網路控制層來實現應用程式感知路由的機會 。有了SDN，我們將可以基於每個應用程式的網路需求來提供最佳化路由。然而，大多數現有的最佳化路由研究只考慮某些特定的應用程式，如影像串流應用程式(類型2)。在應用程式感知路由的領域中最具有代表性的相關研究是CORouting。它考慮了所有種類的應用程式並且將這些應用程式根據它們的網路需求分成三個主要的類型，但是對即時應用程式(類型1)，CORouting只考慮了減少路徑的節點數，而對其它的應用程式類型只做負載平衡。若是網路中類型1的應用程式太多，CORouting的作法可能會導致網路壅塞。在本論文中，我們提出了一個應用程式感知路由機制(App-RS)。它可以滿足在軟體定義網路下雲端資料中心所有類型的應用程式的網路需求。對於即時應用程式，我們所提的App-RS，考慮端點對端點的延遲(end-to-end delay)和連結負載(link load)。對於串流應用程式，我們考慮延遲變化量(delay variation)和連結負載(link load)。對於其他應用程式，我們只考慮連結負載(link load)。我們的方法可以提升三個類型應用程式的效能表現。實驗結果證明，對於三個類型應用程式的平均吞吐量，我們所提的App-RS比CORouting分別高出9.86%,4.53%和4.07%。對於類型1的應用程式的平均端點對端點延遲，我們所提的App-RS比CORouting少25.47%，對於類型2的應用程式的平均延遲變化量，我們所提的App-RS比CORouting少49.65%。因此，我們所提的App-RS比CORouting更能夠滿足在軟體定義網路下雲端資料中心所有類型應用程式對網路的需求。

As running applications in cloud datacenters is getting more and more popular, treating different classes of applications equally is not a proper approach in the current network since the network requirements of different classes of applications are not the same. Application-aware routing is a promising way to meet the network requirements of different classes of applications in cloud datacenters with limited network resources. Software-Defined Networking (SDN) presents an opportunity to realize application-aware routing by offering an external and programmable network control plane. With SDN, it is possible to provide optimal routing based on the network requirements of each application class. However, most of existing researches only consider some specific applications such as video streaming applications (class 2). Content Oriented Routing (CORouting), the most representative work in the study of application-aware routing, considers all application types and categorizes these applications into three main classes according to their network requirements, but it only minimizes the number of hops in paths for real time applications (class 1) and only balance loads for the other applications. CORouting may have the network congestion problem if there are too many class 1 applications. In this paper, we propose an Application-aware Routing Scheme (App-RS) to meet network requirements of all classes of applications in SDN-based cloud datacenters. The proposed App-RS takes into account the end-to-end delay and link load for real time applications (class 1), delay variation and link load for streaming applications (class 2), and link load for miscellaneous applications (class 3) to improve the performance of all classes of applications. Simulation results have shown that the average bandwidth ratio of class 1, class 2, and class 3 applications of App-RS are 9.86%, 4.53%, and 4.07% better than that of CORouting, respectively. The average end-to-end delay of class 1 applications of App-RS is 25.47% lower than that of CORouting. The average delay variation of class 2 applications of App-RS is 49.65% lower than that of CORouting. Therefore, the proposed App-RS can better meet the network requirements of all classes of applications in SDN-based cloud datacenters than CORouting.