【摘要】：SDN (Software Defined network, SDN)網(wǎng)絡(luò)是近些年新興的網(wǎng)絡(luò)架構(gòu),它實現(xiàn)了網(wǎng)絡(luò)控制和數(shù)據(jù)轉(zhuǎn)發(fā)功能的分離,分成控制平面和轉(zhuǎn)發(fā)平面,這兩個平面構(gòu)成了開放可編程的網(wǎng)絡(luò)架構(gòu)體系�？刂破矫嫱ㄟ^OpenFlow南向接口獲取網(wǎng)絡(luò)拓撲和狀態(tài)信息,制定相關(guān)路由策略實現(xiàn)對轉(zhuǎn)發(fā)平面的控制管理,同時抽象底層網(wǎng)絡(luò)用于上層應(yīng)用的數(shù)據(jù)獲取。SDN網(wǎng)絡(luò)由于控制器的瓶頸,注定了單個SDN網(wǎng)絡(luò)的規(guī)模不會太大。數(shù)據(jù)顯示,SDN控制器最多可以支持1000個交換機和250000個主機,因此需要一種擴展性更強的網(wǎng)絡(luò)架構(gòu)模式,即控制器水平交互方式�？刂破魉浇换シ绞叫枰獙DN網(wǎng)絡(luò)劃分為多個子SDN網(wǎng)絡(luò),在本文中又稱為SDN區(qū)域,通過每個區(qū)域的控制器交互完成基本的域間路由。為了保證運行在分布式計算節(jié)點的控制器的數(shù)據(jù)同步和共享,本文采用分布式協(xié)調(diào)模塊ZooKeeper完成控制器的狀態(tài)信息同步和域間路徑信息一致性保證,實現(xiàn)域間路由。通過基于跳數(shù)的最短路徑完成域間路由有著明顯的問題,難以滿足復(fù)雜的流量特性,造成網(wǎng)絡(luò)網(wǎng)絡(luò)部分資源利用率不均衡,影響服務(wù)質(zhì)量。通過業(yè)務(wù)量矩陣估計可以實現(xiàn)網(wǎng)絡(luò)地負載均衡和路由算法地優(yōu)化,因此業(yè)界針對業(yè)務(wù)量矩陣估計也有多種實現(xiàn),本文比較SDN網(wǎng)絡(luò)業(yè)務(wù)量矩陣的非直接測量方式和直接測量方式的優(yōu)缺點,并結(jié)合業(yè)界一些實現(xiàn)提出了基于直接測量方式的一種啟發(fā)式算法,相對精確地完成業(yè)務(wù)量矩陣測量。本文最終通過所有區(qū)域控制器同時維護統(tǒng)一的全局拓撲,并結(jié)合業(yè)務(wù)量矩陣測量和區(qū)域內(nèi)虛連接的帶寬測量,虛擬化更為細粒度的虛擬全局拓撲,由源端控制器完成全局的最短路徑計算,并通過流路徑共享使其他控制器協(xié)作完成域間路徑的路由,下發(fā)相應(yīng)流表項。本文第一章介紹了控制器水平架構(gòu)下的相關(guān)技術(shù),第二章結(jié)合ZooKeeper實現(xiàn)基本的域間路由,第三章介紹了業(yè)務(wù)量矩陣測量的幾種方式,第四章則實現(xiàn)區(qū)域間不完備信息的路由,第五章搭建仿真平臺,完成相關(guān)數(shù)據(jù)測量和對比,最后則針對全文總結(jié)并計劃下一步工作方向。
[Abstract]:SDN (Software Defined network, SDN) network is a new network architecture in recent years. It realizes the separation of network control and data forwarding, which is divided into control plane and forwarding plane. These two planes constitute an open programmable network architecture system. The control plane acquires the topology and state information of the network through the southern interface of OpenFlow, and formulates the related routing strategy to realize the control management of the forwarding plane. At the same time, it abstracts the data acquisition for the upper application of the underlying network. SDN network is due to the bottleneck of the controller. Doomed to a single SDN network size will not be too large. Data show that SDN controller can support up to 1000 switches and 250000 hosts, so it needs a more scalable network architecture mode, that is, controller horizontal interaction mode. The horizontal interaction mode of controller needs to divide the SDN network into several sub-SDN networks. In this paper, it is also called the SDN region, and the basic inter-domain routing is accomplished by the controller interaction in each region. In order to ensure the data synchronization and sharing of the controller running in the distributed computing node, the distributed coordination module ZooKeeper is used to complete the synchronization of the controller state information and the consistency of the inter-domain path information, so as to realize inter-domain routing. The inter-domain routing based on the shortest path based on the number of hops has obvious problems, which is difficult to meet the complex traffic characteristics, resulting in uneven utilization of some resources in the network and affecting the quality of service. Network load balancing and routing algorithm optimization can be realized by traffic matrix estimation. In this paper, the advantages and disadvantages of indirect measurement and direct measurement of SDN network traffic matrix are compared, and a heuristic algorithm based on direct measurement method is proposed to measure traffic matrix accurately. In this paper, the unified global topology is maintained at the same time by all the regional controllers, and combined with the traffic matrix measurement and the bandwidth measurement of the virtual connection within the region, the virtual global topology with finer granularity is virtualized. The global shortest path is calculated by the source controller, and the other controllers are coordinated to complete the routing of the inter-domain path through the flow path sharing, and the corresponding flow table items are sent down. The first chapter introduces the related technologies under the controller level architecture, the second chapter realizes the basic inter-domain routing with ZooKeeper, the third chapter introduces several ways to measure the traffic matrix, and the fourth chapter realizes the routing of incomplete information between regions. In the fifth chapter, the simulation platform is built to measure and compare the relevant data. Finally, the author summarizes the whole paper and plans the next work direction.
【學(xué)位授予單位】：電子科技大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2016
【分類號】：TN915.05

【參考文獻】

相關(guān)碩士學(xué)位論文 前1條

1 鄭偉;業(yè)務(wù)量矩陣估算算法的設(shè)計與實現(xiàn)[D];天津大學(xué);2005年

，

本文編號：2203115