魏月华，陈晓，张征
（中兴通讯股份有限公司，中国 深圳 518057）
摘要：受计算规模的驱动，数据中心物理拓扑从接入-汇聚-核心三级网络架构演进到基于Clos的Spine-and-Leaf架构。计算资源的基本单位经历了物理服务器、虚拟机、容器化3个阶段。数据中心底层（underlay）连接协议逐步从以二层协议为主演进到以IP路由协议为主。但传统路由协议存在可扩展性、拓扑可见性、自动化部署能力等诸多问题。结合链路状态和距离矢量的胖树路由协议，解决了超大规模数据中心部署的痛点问题，有望逐渐成为超大规模数据中心底层网络的主流技术。
关键词：Spine-and-leaf；路由；数据中心
Data Center Infrastructure and Underlay Protocol Evolution
WEI Yuehua, CHEN Xiao, ZHANG Zheng
(ZTE Corporation, Shenzhen 518057, China)
Abstract: Driven by the scale of computing, the physical topology of the data center has evolved from an access-aggregation-core three-level network architecture to a Clos-based Spine-and-Leaf architecture. The basic unit of computing resources has gone through three stages: physical server, virtual machine, and containerization. The underlay connection protocol of the data center has gradually evolved from layer 2 protocol to IP routing protocol. However, traditional routing protocols have many problems, such as scalability, topology visibility, and automated provision capabilities. The fat-tree routing protocol, which combines link state and distance vector, solves the pain points of ultra-large-scale data center deployment, and is expected to gradually become the mainstream technology for ultra-large-scale data center underlay networks.
Keywords: Spine-and-Leaf; routing; data center
.png)
.png)
.png)
