在軟體定義網路下最小化監控網路延遲之成本

Abstract

在次世代的網路中，軟體定義網路勢必是一大趨勢。在軟體定義網路的概念裡，我們可以使網路程式化，依據我們對網路的要求進一步的將想法透過程式化來控制網路。而網路監控是一件很重要的事情，眾多的網路服務已經存在於我們的生活周遭，因此當網路發生異常時，迅速的找到原因並解決它是必須的，因此，隨時監控網路的狀況是一項基本且重要的工作。擁有好的監控資料，我們才能即時的解決問題。過去在傳統網路中所使用的網路測量工具如：sFlow、NetFlow，若要使用這些工具則交換器本身必須要具備能夠支援此服務的能力，且過去方法對於網路造成的額外負擔也相對的大，如此一來便大大提升了網路監測的成本。而目前在 SDN 最被廣泛使用的協定：OpenFlow，我們可以利用 OpenFlow 來做網路監測。儘管利用 OpenFlow 來實施網路監測，OpenFlow 依然會對 Controller 以及 Switch 造成額外的負擔，而這些負擔同時也會影響 Controller 跟 Switch 的效能表現，並且進一步的影響整體網路的狀況。而本篇論文提出了動態調整監控網路的頻率，以及有效地找出監控目標的時機，在降低監控網路成本的同時，也可以得到精準的測量資料。

In the field of next-generation networking, software-defined networking (SDN) is on the verge of becoming popular. The main concept of SDN is that networks can be rendered programmable. Network monitoring is an essential technique that can be employed to obtain large amounts of information from a network; for example, information regarding throughput, delay, and packet loss. Tools such as sFlow and NetFlow can be employed to monitor traditional networks; however, network devices require the capability to support such protocols. Furthermore, traditional network monitoring methods usually engender considerable overhead on the network. Although SDN and OpenFlow enable more efficient network monitoring, employing OpenFlow to monitor the network still results in additional overheads to the controller and switches, which affect their performance. This example is considered a trade-off. This thesis proposes a method of network monitoring based on adaptive polling that employs several control messages in OpenFlow to monitor the network efficiently, thereby achieving a highly accurate result with minimal costs.