数据中心内部光互联结构及控制理论研究

发布时间：2018-06-23 11:29

本文选题：数据中心光互联 + 波分复用　；参考：《北京邮电大学》2016年博士论文

【摘要】：在过去的几年中,由于流媒体,社交网络,以及云计算等应用的崛起,互联网的流量正在呈幂指数次增长,这就对数据中心提出了更高要求,这些数据中心基于数以万计的服务器通过高性能交换机互联。面向数据中心的应用,如云计算和搜索引擎应用,更是属于数据密集型业务,需要服务器之间高效的互联。这样的互联使网络对于高带宽和低时延的通信提出了更高的要求,更有甚的是,为了降低运营成本,这些数据中心必须满足低能耗的需求。本论文围绕数据中心内部光网络高带宽低能耗的需求,采用光互联的解决方案,重点从网络结构的可扩展性、交换的灵活性、业务粒度的多样性等方面进行数据中心内部光互联结构的设计和研究,主要工作和创新点包括以下几个方面:(1)将波分复用(WDM)技术和数据中心光互联网络相结合,根据数据中心内部光网络的特点以及业务需求,将数据中心目前以太网网络结构和WDM光网络进行了融合,采用簇间(inter-cluster)微机械光开关以及簇内(intra-cluster)微机械光开关级联的方式,来应对数据中心网络规模不断扩大对扩展性的需求。该结构采用集中式统一控制方式,由中央控制器下发配置命令并进行路径的计算和优化。(2)在支持波分复用数据中心内部光网络中,提出了基于复用/解复用器件以及波长选择开关的MUX/DEMUXWSS结构,该交换模块可以在光层上进行业务的汇聚和疏导,将来自于不同源节点具有同一目的地址的业务同时交换到同一目的节点。实现多点到单点、多点到多点的同时通信。该模块支持集中式控制方式。(3)在支持可变带宽收发机的数据中心内部光网络中,提出了基于复用/解复用器件和频谱选择开关的MUX/DEMUXSSS结构,该结构充分利用了灵活频谱网络更细的粒度以及较高的频谱利用率,对由一个源节点产生的多个子载波组成的超级信道(Superchannel)业务进行选择输出,使去往不同目的端口的业务能够同时交换到宿节点。实现点到多点、多点到多点的即时通信。(4)提出了数据中心内部网络可变带宽发射机和固定栅格发射机相结合的发射接收模式,使用了单载波频分复用(SCFDM)技术作为可变带宽发射接收机的支撑技术,并且在控制上采用了流量监测的方法,对不同粒度的业务进行区分和鉴别,以分配不同的收发机进行交换配置,以实现资源的最大利用,降低系统的阻塞率和时延。(5 )提出了混合光电路和光突发交换数据中心内部光互联结构,将服务器通过嵌入其内部的交换接口卡同混合光电路和光突发交换开关直接连接,省去了机柜顶端交换机,将数据中心内部突发业务通过突发汇聚交换来完成。通过对业务时延敏感度进行区分,将不同传输需求的业务打包汇聚到一起,分别经过光电路交换开关和光突发交换开关实现交换路由。
[Abstract]:Over the past few years, Internet traffic has grown exponentially because of the rise of streaming media, social networking, cloud computing, and so on, putting higher demands on data centers. These data centers are based on tens of thousands of servers interconnected through high-performance switches. Data center-oriented applications, such as cloud computing and search engine applications, are more data-intensive, requiring efficient interconnection between servers. Such interconnection makes the network more demanding for high bandwidth and low delay communication. What is more, in order to reduce the operation cost, these data centers must meet the requirements of low energy consumption. This paper focuses on the demand of high bandwidth and low energy consumption of optical network in data center, and adopts the solution of optical interconnection, focusing on the extensibility of network structure and the flexibility of switching. The design and research of data center internal optical interconnection structure are carried out in the aspects of the diversity of service granularity. The main work and innovation include the following aspects: (1) combining wavelength division multiplexing (WDM) technology with data center optical interconnection network, According to the characteristics of the optical network inside the data center and the service requirements, the current Ethernet network structure of the data center and the WDM optical network are fused, and the inter-cluster (inter-cluster) micro-mechanical optical switch and the intra-cluster (intra-cluster) micro-mechanical optical switch are cascaded. To cope with the data center network scale continues to expand the expansion of the demand. The structure adopts centralized unified control mode, and the configuration command is issued by the central controller and the path is calculated and optimized. (2) in the optical network supporting WDM data center, A MUX / DEMUXWSS structure based on multiplexing / demultiplexing device and wavelength selective switch is proposed. The switching module can converge and dredge traffic on the optical layer. Exchange services from different nodes with the same destination address to the same destination node at the same time. Multipoint to single point, multi-point to multi-point simultaneous communication. The module supports centralized control mode. (3) in the data center optical network with variable bandwidth transceiver, a MUX / DEMUXSSS structure based on multiplexing / demultiplexing device and spectrum selection switch is proposed. The structure makes full use of the finer granularity and higher spectrum efficiency of the flexible spectrum network, and outputs the superchannel services composed of multiple subcarriers generated by a single source node. Enables services going to different destination ports to be switched to the host node at the same time. To achieve point-to-multi-point, multi-point to multi-point instant communication. (4) A transmission and reception mode combining variable bandwidth transmitter and fixed grid transmitter in data center internal network is proposed. The single carrier frequency division multiplexing (SCFDM) technique is used as the support technology of the variable bandwidth transmit receiver. The traffic monitoring method is used in the control to distinguish and distinguish the different granularity services. In order to realize the maximum utilization of resources and reduce the blocking rate and delay of the system, the switching configuration of different transceivers is used. (5) A hybrid optical circuit and an optical interconnection structure within the optical burst switching data center are proposed. The server is connected directly with the hybrid optical circuit and the optical burst switch through the switch interface card embedded in the server, which saves the top switch of the cabinet, and completes the burst service in the data center by means of the burst convergent switch. By differentiating the sensitivity of service delay, the services with different transmission requirements are packed and converged together, and the switching route is realized by optical circuit switching switch and optical burst switching switch respectively.
【学位授予单位】：北京邮电大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TN929.1

【相似文献】