Section 1: Pure L3 Data Center

In this section, we will design a spine-leaf IPv6 fabric with unnumbered BGP, ECMP, BFD.

Discuss the protocols and topology of the underlying network to build a .

1. How the Data Center has evolved, and how to make it scale: Discuss Pure Layer 3 Underlay and how BFD is used for sub-second link failover to meet high availability SLAs
   • IPv6 is used for easy management of IP addresses. Using Neighbor Discovery Protocol and Link-Local Addresses, we can achieve Route Advertisement without any configuration.
   • Spine and Leaf Topology is used to avoid the overhead of Layer 2 networking.
   • Border Gateway Protocol (BGP) is used for load balancing and advertising routes to all servers, routers, and switches. In terms of scalability, BGP is currently handling ~700,000 routes of the public internet. BGP can use link-local addresses to exchange information about IP addresses configured in a data center.
   • BIG TCP: Extends TCP's window-scaling and congestion control, improving performance and throughput as packets enter the networking stack.
   • BBRv3 is used for
   • Stateless Address Autoconfiguration is used for
2. Case Study - Nokia Kubernetes Service

📄️ Spine-Leaf Underlay (IPv6 Link-Local)

Modern data centers use a spine-leaf topology for predictable, scalable L3 networking. Every leaf switch connects to each spine, ensuring at most one hop between any server. This fabric is purely Layer-3: no spanning tree, no VLANs across the fabric – just IP routing.

📄️ BGP Unnumbered and Fast Failover (BFD)

Continuing our pure L3 design, we use BGP between all spine-leaf pairs without assigning any IP addresses to interfaces. This is accomplished with BGP Unnumbered, where neighbors are identified by interface name and IPv6 link-local address.