Lab 03 — Data Centre Fabric

BGP EVPN VXLAN

Creating a BGP EVPN VXLAN fabric: spine-leaf switching with a modern overlay control plane.

EVPNMP-BGPNXOSOSPFVXLAN

Overview

This lab builds a compact data-centre fabric the way modern campuses and clouds run it: two spines, two leaf pairs, and NX-OS virtual switches hosting the tenants. OSPF forms the underlay so every loopback can reach every other loopback, then MP-BGP EVPN advertises MAC and IP reachability between the leaves.

VXLAN carries the tenant frames across that underlay, which means new segments are created with configuration only — no recabling, no spanning-tree redesign, no maintenance window for the physical fabric.

Lab facts

  • PlatformEVE-NG · NX-OS
  • UnderlayOSPF, /32 loopbacks
  • OverlayMP-BGP EVPN
  • Data planeVXLAN, flood-and-learn
  • Fabric2 spines · 4 leaves

What this lab covers

  • Spine-leaf cabling with OSPF unnumbered underlay links
  • MP-BGP EVPN peering from every leaf to both spines
  • NVE interfaces binding VLANs to VXLAN VNIs
  • Anycast gateway so hosts share one default route
  • Host mobility test: moving a VM between leaves
  • Spine loss drill with traffic reconverging via the survivor

Key configuration

Technician patching a network switch Cabled server rack

Results & takeaways

Hosts on the same segment but different leaves communicated at Layer 2 through the overlay, and the moved VM kept its address and gateway after migrating leaves — EVPN simply re-advertised the MAC from the new location. Losing a spine halved the paths but dropped no tenant traffic.

The main lesson: the underlay deserves as much care as the overlay. One mistyped loopback broke EVPN peering in a way that looked exactly like a VXLAN fault — verifying underlay reachability first would have found it in minutes.

Related labs

All topologies Start your project