Large VLAN domains depend on spanning tree. Every topology change triggers recalculation and temporary forwarding loss. EVPN-VXLAN replaces large Layer 2 domains with routed underlays and distributed MAC learning. Regional ISPs use this to connect data centers, IX edge infrastructure, and broadband aggregation networks without stretching spanning tree across sites
VXLAN encapsulates Ethernet frames inside UDP. EVPN distributes MAC and IP reachability using BGP. This removes the traditional flood-and-learn behavior from the fabric. This breaks when route targets or VNIs mismatch, resulting in silent MAC reachability failures between leaf switches.
Traditional Design
+--------+ +--------+
| Switch |-------| Switch |
+--------+ +--------+
\ /
\ /
+-----------+
| Spanning |
| Tree |
+-----------+
MAC tables grow across all switches, spreading broadcast traffic everywhere. This causes instability during loops or link flaps. We see temporary packet loss during convergence events.
How EVPN-VXLAN Changes the Model
EVPN uses a routed spine-leaf fabric.
EVPN-VXLAN Fabric
+---------+
| Spine 1 |
+----+----+
|
-----------------------
| |
+----+----+ +-----+----+
| Leaf 1 | | Leaf 2 |
+----+----+ +-----+----+
| |
Servers IX Edge
Leaf switches terminate VXLAN tunnels. The underlay uses IP routing and ECMP to forward the traffic across all available links. This removes spanning tree blocking. Every uplink forwards traffic simultaneously, increasing available bandwidth.
Control Plane Operation
EVPN distributes endpoint information using BGP.
Example MAC Advertisement
MAC: aa:bb:cc:dd:ee:ff
VNI: 5000
Next Hop: 192.0.2.10
Each leaf advertises local MAC addresses it knows about. Remote leaves install forwarding entries without flooding unknown traffic. This reduces broadcast volume inside the fabric. This results in lower control-plane churn during VM movement or subscriber migration, such as due to a failed link.
ARP Suppression and Broadcast Reduction
Traditional VLANs flood ARP requests everywhere. This is how things are discovered.
Without EVPN
Host A -> Broadcast ARP -> Entire VLAN
With EVPN ARP Suppression
Leaf switch replies locally using EVPN database
The leaf switch answers requests from cached endpoint information. Traffic stays local when possible. This lowers broadcast utilization on edge segments which reduces CPU load on switches handling subscriber aggregation.
Multi-Homing with Ethernet Segments
EVPN supports active-active uplinks.
Example
Customer Router
/ \
Leaf 1 Leaf 2
Both links forward simultaneously. EVPN elects a designated forwarder for BUM traffic. This avoids spanning tree failover delays. Traffic continues across the remaining path during link failure.
Failure Handling
The underlay detects failures using routing protocols and BFD.
Failure Sequence
- Leaf loses uplink
- BFD detects failure
- Route withdrawn from fabric
- ECMP recalculates paths
Convergence occurs within seconds or milliseconds depending on timers. This limits packet loss during maintenance or fiber cuts.
Operational Example for a Regional ISP
A regional ISP operates:
- Broadband aggregation routers
- IX edge infrastructure
- Hosted services
The ISP extends Layer 2 services between two metro sites using EVPN.
Deployment Goals
- Avoid spanning tree across metro transport
- Reduce broadcast traffic
- Enable active-active uplinks
Fabric Design
Metro A Leaf <--VXLAN--> Metro B Leaf
\ /
\ /
Spine Layer
Customer VLANs map to VNIs. BGP EVPN distributes endpoint state between sites. This allows subscriber mobility without extending physical VLAN trunks.
Configuration Example
VXLAN Interface
interface nve1
source-interface loopback0
member vni 5000
BGP EVPN
router bgp 64500
address-family l2vpn evpn
advertise-all-vni
VNIs map Layer 2 segments into the overlay. Route targets control import and export policy between leaves.
Incorrect route targets isolate segments unexpectedly. This shows up when MAC routes appear in BGP but traffic does not forward.
Monitoring and Validation
Track:
- BGP EVPN neighbor state
- MAC route count
- VTEP reachability
- BFD session state
Useful Commands
show bgp l2vpn evpn
show nve peers
show vxlan counters
Rising unknown unicast traffic usually indicates MAC learning issues. High BUM replication often indicates missing ARP suppression.
Metrics Overlay
| Metric | Traditional L2 | EVPN-VXLAN |
|---|---|---|
| Convergence | Seconds | Sub-second |
| Blocked Links | Yes | No |
| Broadcast Scope | Large | Reduced |
| Active-Active Uplinks | Limited | Native |
https://j2sw.com
https://startawisp.info
https://indycolo.net
#packetsdownrange #routethelight
Discover more from Justin Wilson (j2sw)
Subscribe to get the latest posts sent to your email.