資料中心基於軟體定義網路及部分軟體定義網路架構下之多播

Data Center Multicast in SDN and Partial SDN Environment

劉雨鑫(Yu-Hsin Liu)

指導教授 ： 周承復

國立臺灣大學/電機資訊學院/資訊網路與多媒體研究所/碩士(2015年)

https://doi.org/10.6342/NTU.2015.00402

摘要

近幾年來，隨著資料量的規模增多，大數據的時代已經到來。由於資料集過度龐大，傳統的資料分析方式已經不敷使用。Hadoop因此被開發出來對大數據進行分析，其核心方法為MapReduce，即將分析工作分割成許多小部分，並指派給資料中心內眾多個節點來平行運算處理，進而加速整體的運算速度。而在將工作指派給資料中心內眾多節點的時候，最有效率的傳遞資料之方法就是多播。

本篇論文提供一個能在資料中心下部分軟體定義網路環境下可以實行多播的架構。過往的相關研究要不只有研究純軟體定義網路架構下的多播，要不就是只有部分軟體定義網路架構但是不支援多播。本篇論文著重在如何使軟體定義網路架構下的多播可行，並對這種情況下的網路壅塞進行處理。

為了使多播可行，支援軟體定義網路的交換器必須能夠以傳統多播協定與傳統交換器進行溝通。本篇論文使用「獨立組播協議-稀疏模式」做為傳統多播協定，並且以「虛擬區域網」設定做為控制多播樹的方法。當發生交通壅塞時，使用一個貪婪演算法去選擇有最大可能流量的路徑去建構多播樹。如此並不用將全部網路設備更換為支援軟體定義網路的交換器，仍然可以獲得軟體定義網路所提供的好處。

關鍵字

資料中心 ； 多播 ； 軟體定義網路 ； 部分軟體定義網路 ； 壅塞

並列摘要

Big data is becoming more and more popular between researchers and business developments nowadays. With high volume, high velocity, and high variety information, it requires new forms of processing rather than traditional data processing applications.

Hadoop is a widely used framework for with big data, and MapReduce is the programming model for processing large data sets with a parallel, distributed algorithm on a data center. It splits files or jobs into many parts and distributes them to the nodes in the data center to process in a parallel way, allowing data processing to be faster than conventional data processing. One way to efficiently distribute data to nodes in the data center is multicast.

This thesis presents an architecture for data center multicast in a hybrid Software Defined Networking (SDN) environment, and propose an algorithm that dynamically changes the multicast tree when congestion happens.

Prior works had proposed either multicast in a full SDN environment or an architecture for unicast for hybrid SDN environment. This thesis focus on how to enable multicast in a hybrid SDN environment and how to deal with congestion in this environment.

To enable multicast in hybrid SDN environment, SDN switches has to use traditional multicast protocol to interact with traditional switches. In this thesis, it uses PIM-SM as the traditional, and VLAN setting for changing multicast tree. When congestion is detected, a greedy algorithm is used that chooses the path that has the maximum possible bandwidth to build a multicast tree.

並列關鍵字

data center ； multicast ； software defined networking ； SDN ； hybrid SDN ； congestion

參考文獻

[10] Nick McKeown, Tom Anderson, Hari Balakrishnan, Guru Parulkar, Larry Peterson,

Jennifer Rexford, Scott Shenker, and Jonathan Turner. Openflow: enabling innovation

and Scott Shenker. Nox: towards an operating system for networks. ACM

[22] Rob Enns, Martin Bjorklund, and Juergen Schoenwaelder. Netconf configuration

[7] Amrit Iyer, Pranaw Kumar, and Vijay Mann. Avalanche: data center multicast using