標題: 基於OpenvSwitch平台的錐狀光資料中心 網路虛擬化定址設計
Design of Network Addressing for virtualized OPMDC based on OpenvSwitch Platform
作者: 蔡孟橋
TSAI, MENG-CHIAO
田伯隆
Tien, Po-Lung
電信工程研究所
關鍵字: 網路虛擬化;光資料中心網路;軟體定義網路;Network Virtualization;Optical Data Center Network;Software Defined Network
公開日期: 2015
摘要: 網路虛擬化 (Network Virtualization)將網路予以抽象,打破實體網路不可切割的障礙。讓每個使用者(tenant) 能夠擁有自己的網路,而達到更好、更安全的方式來使用網路資源。目前已經有多種虛擬化的方法如VXLAN、NVGRE等等,並且大部分的雲端作業系統也提供網路虛擬化的功能。 本篇論文所研究的資料中心網路架構於[3]所提出,錐狀光資料中心(Optical Pyramid Data Center Network architecture,OPMDC) 是一種以三層的光交換機 (Optical Switch) 組成的資料中心架構。因為OPMDC並沒有使用電界面網路層的設備(如: 路由器),網路虛擬化方法VXLAN和NVGRE所提供的Layer 2-in-Layer 3封裝機制無法在OPMDC上發揮它在第三層網路的優點,反而造成傳輸的負擔和額外的設備成本。並且VXLAN和NVGRE在執行使用者地址學習時,所發出的大量多播學習訊息也會降低OPMDC的光路使用效率。除了NVGRE和VXLAN之外,雲端作業系統也提供他們自己的虛擬化機制,但是雲端系統的管理網路會浪費太多OPMDC的光路資源。在本篇論文中我們提出了對虛擬機制的改良,在伺服器(host server)內以虛擬交換機(OpenvSwitch)來改良並且實作虛網路擬化的地址機制。 我們提出的第一個模型稱為對應模型(Mapping Module)來改善原先VXLAN和NVGRE的定址機制,達到減少傳輸上的負擔;並且這個模型也會管理封包內外網路地指的對應。我們提出的第二個模型是溝通模型(Communication module),此模型和pod控制機(pod controller)互相溝通來取得用戶資訊管理或是執行pod控制機給的指令,並且再把新的用戶資訊統整給對應模型。此模型主要是改善多播或是管理網路對OPMDC照成的光路浪費,使得網路虛擬化在OPMDC上有效率的運作。為了展示所提出模型的效果,此論文將會用兩個例子來解釋這個模型運作流程和系統的優點,並且此網路虛擬化系統的封包傳輸負擔和VXLAN比起來少了27.5%,和NVGRE比起來少了14.7%。另外,這個系統的控制機制使用頻外控制使得光路規劃更有效率。
Network virtualization builds logical virtual networks that are abstracted from the underlying network hardware. Such abstraction not only improves resource efficiently, but also allows different tenants to efficiently access data center network resources on a secured and independent manner. The prevailing network virtualization solutions for data center are VXLAN and NVGRE. And most Cloud OSs also provide network virtualization in their system. Our study is based on the Optical Pyramid Data Center Network architecture (OPMDC) [3], which is built on three types of WSS-based optical switching nodes in three tiers. Because OPMDC doesn’t have layer 3 electrical network devices (e.g., router), the L2-L3 encapsulation mechanism of VLXAN and NVGRE can’t bring its layer 3 benefits to OPMDC but cause more packet overhead and more cost. And the multicast tenant address learning traffic of VXLAN and NVGRE will lead to low efficiency in optical path utilization. Aside from VXLAN and NVGRE, most cloud OSs also provide network virtualization. But most cloud OSs require a management network, which costs certain amount of optical channels and is inefficient for OPMDC. In this thesis, we design two network virtualization modules in host server based on OpenvSwitch. The first module we designed is called Mapping module, which improves the encapsulation and addressing mechanism to reduce the packet overhead. This module also handles the mapping of virtual IP address to physical IP address. The second module is called Communication module. This module mainly communicate with pod controller to update tenant information, and communication module will send updated tenant information to Mapping module. Communication module improves low optical path efficiency that cause by multicast address learning or by cloud OS management network. These two module makes virtualization more efficient in OPMDC. To demonstrate the functionality of our module, this thesis will take two different scenario to explain how this module work and its advantages. In this network virtualization system, we reduce the packet header overhead by 27.5% of packet compare to VXLAN, and 14.7% compare to NVGRE. And this system also improves the inefficient wavelength assignment by using out-of-band control.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT070260219
http://hdl.handle.net/11536/127132
Appears in Collections:Thesis