BGP at Doors of Autonomous Systems
A highly scalable, multi-tenant Data Center fabric utilizing BGP EVPN and VXLAN over an OSPF underlay, built with FRRouting.
Overview
Multi-node GNS3 topologies using FRR to simulate realistic routing behavior with automated configuration.
The Challenge
Traditional Layer 2 networks suffer from spanning-tree blocking and broadcast storms, severely limiting scalability in modern data centers. The challenge was to design and deploy a modern Spine-Leaf architecture that extends Layer 2 broadcast domains across a routed Layer 3 fabric (Overlay over Underlay), mimicking the infrastructure used by major cloud providers.
The Solution
GNS3 topologies with FRR for routing simulation, static and dynamic multicast tests, automated configuration and startup with shell scripts.
Architecture & Topology
Technical Deep Dive
BGP EVPN Control Plane
Replaced traditional flood-and-learn MAC address discovery with a BGP Ethernet VPN (EVPN) control plane. MAC addresses and IP bindings are distributed as BGP NLRI, dramatically reducing broadcast traffic.
VXLAN Overlay
Implemented Virtual eXtensible LAN (VXLAN) encapsulation to tunnel Layer 2 Ethernet frames over UDP/IP. This allowed for 16 million unique VNIs, solving the strict 4,094 limit of traditional 802.1Q VLANs.
OSPF Underlay & ECMP
Deployed an OSPFv2 underlay to ensure loop-free, sub-second convergence for VTEP reachability. Leveraged Equal-Cost Multi-Path (ECMP) routing across the spine layer to actively utilize all available bandwidth.
Route Reflection
Designed an iBGP Route Reflector architecture in the Spine layer to solve the full-mesh BGP scaling problem. This drastically reduced the number of TCP peering sessions the Leaf switches had to maintain.