>_nexthopreachable

Extending VLANs with EVPN VXLAN

Ali Bahadır Coşkun7 min read

In the previous lab I stretched one VLAN across two switches with static VXLAN. At the end I said that in real networks people usually replace the static flood list with EVPN. In this second lab we do exactly that. We also add one small twist the same bridge domain uses a different VLAN tag on each switch. It may look weird at first but this can happen during a migration or after a merger. It is also a good way to show that in EVPN the VLAN tag is only local to the switch.

Overview

This lab uses three Arista cEOS switches in a routed triangle and five Linux hosts. The loopbacks, routed links, OSPF, and IBGP sessions for IPv4 unicast are already configured. There are two bridge domains:

  • red is stretched across all three switches, but with a different local VLAN tag on each: 201 on s1, 202 on s2, and 203 on s3
  • blue exists only on s1 and s2, both using VLAN 101

Our job is to build the VXLAN overlay (red as VNI 1000, blue as VNI 1001), enable the EVPN address family on the existing IBGP sessions, and configure one MAC-VRF for each bridge domain. EVPN is the control plane for VXLAN. In simple words, the switches use BGP to tell each other which VNI they belong to and which MAC addresses they learned. Because of that, we do not need to build everything by hand. In the previous lab I typed the flood vtep list myself. Here EVPN builds that list for us. It does this with type-3 routes, also called IMET routes (Inclusive Multicast Ethernet Tag). A type-3 route says, "I am in this VNI, and this is my VTEP address." Other switches use this information when they need to send broadcast, unknown-unicast, or multicast traffic. This copying of traffic to remote VTEPs is called ingress replication. The sending VTEP makes one copy per remote VTEP in the list. Because blue exists on only two switches, the ingress replication list for VNI 1001 is smaller than the list for VNI 1000.

When everything works, hr1, hr2, and hr3 will be in one stretched layer-2 domain, and hb1 and hb2 will be in another one, even if the red hosts sit behind three different VLAN tags.

Step 1: Verify the Underlay and the Existing IBGP

On each switch:

text
show ip ospf neighborshow ip routeshow bgp ipv4 unicast summaryFrom s1: ping 10.0.0.2 source 10.0.0.1From s1: ping 10.0.0.3 source 10.0.0.1

What you want to see:

  • OSPF neighbors in FULL state
  • both remote loopbacks in the routing table
  • the IPv4 unicast IBGP sessions already up
  • loopback-to-loopback pings working

Step 2: Configure the VXLAN Interfaces

Each switch maps its local red VLAN tag to the shared VNI 1000. The VLAN number matters only on the local link between the switch and the host. Across the fabric, the segment is identified by the VNI.

On s1:

text
configure terminalinterface Vxlan1   vxlan source-interface Loopback0   vxlan vlan 201 vni 1000   vxlan vlan 101 vni 1001end

On s2:

text
configure terminalinterface Vxlan1   vxlan source-interface Loopback0   vxlan vlan 202 vni 1000   vxlan vlan 101 vni 1001end

On s3:

text
configure terminalinterface Vxlan1   vxlan source-interface Loopback0   vxlan vlan 203 vni 1000end

Configuration details:

CommandPurpose
interface Vxlan1Creates the VXLAN interface
vxlan source-interface Loopback0Uses the loopback as the local VTEP source
vxlan vlan <local-vlan> vni 1000Maps the local red VLAN to the shared red VNI
vxlan vlan 101 vni 1001Maps the blue VLAN to the blue VNI (s1 and s2 only)

Compared to the static VXLAN lab, there is no flood vtep command. We are not telling the switch by hand where to send BUM traffic. EVPN will handle that part.

Step 3: Enable the EVPN Address Family

The IBGP sessions between the loopbacks already exist. To turn on EVPN, we only need to activate the neighbors under the EVPN address family.

On s1:

text
configure terminalrouter bgp 65000   address-family evpn      neighbor 10.0.0.2 activate      neighbor 10.0.0.3 activateend

Do the same on s2 and s3, and activate the other two loopbacks on each switch.

Step 4: Configure the MAC-VRFs

On Arista EOS, the EVPN bridge-domain policy is under router bgp inside a vlan <local-vlan-id> block. The local VLAN IDs are different, but the route targets must match.

On s1:

text
configure terminalrouter bgp 65000   vlan 201      rd 10.0.0.1:100      route-target import 65000:100      route-target export 65000:100      redistribute learned   vlan 101      rd 10.0.0.1:101      route-target import 65000:101      route-target export 65000:101      redistribute learnedend

On s2, use the same pattern with vlan 202 and rd 10.0.0.2:100 / 10.0.0.2:101. On s3, configure only the red MAC-VRF with vlan 203 and rd 10.0.0.3:100.

Configuration details:

CommandPurpose
rd 10.0.0.x:100Gives each MAC-VRF instance a unique route distinguisher
route-target import/export 65000:100Puts all red instances into one EVPN service
route-target import/export 65000:101Puts the two blue instances together
redistribute learnedAdvertises locally learned MAC addresses into EVPN

The three red instances are configured under vlan 201, vlan 202, and vlan 203, which are three different numbers on three different switches. The shared service is built by the matching VNI and route targets. The EVPN updates carry the VNI. The VLAN tag never leaves the local switch.

Step 5: Verify the EVPN Control Plane

On any switch:

text
show bgp evpn summaryshow bgp evpn route-type imetshow vxlan vnishow vxlan flood vtep

These commands answer these checks:

  • show bgp evpn summary: confirms the EVPN sessions are up
  • show bgp evpn route-type imet: confirms type-3 (IMET) routes exist for red from all three switches, and for blue only from s1 and s2
  • show vxlan vni: confirms the local VLAN-to-VNI mappings
  • show vxlan flood vtep: confirms the flood lists were built by EVPN and not typed by hand

EVPN IMET routes on s1:

In the static VXLAN lab, the flood list was something I typed by hand. In this lab, it is built from the IMET routes for each VLAN. On s1, VLAN 201 floods to both remote VTEPs, while VLAN 101 floods only to s2:

Step 6: Verify Host Connectivity and MAC Learning

From the host shell:

bash
netlab connect hr1 ping -c 3 hr2netlab connect hr1 ping -c 3 hr3netlab connect hb1 ping -c 3 hb2

When we ping from hr1 to hr3, the frame enters s1 on VLAN 201 and leaves s3 on VLAN 203. But for the hosts, it still looks like one flat segment:

After some traffic flows, we can also see the MAC side of EVPN:

text
show bgp evpn route-type mac-ipshow mac address-table dynamic
  • show bgp evpn route-type mac-ip: shows the type-2 routes that carry MAC reachability
  • show mac address-table dynamic: shows where the switch installed local and remote MACs

On s1, the local hosts are on Ethernet3 and Ethernet4, while the remote MACs point to Vxlan1 in both VLANs:

The same check on s3 shows only red MACs. The blue MACs never go there because s3 imports neither the route target nor the VNI.

Key Takeaways

  • VLAN tags are locally significant. The bridge domain is identified by the VNI and the route-target policy, so the fabric can still work even if the switches use different VLAN numbers.
  • The flood list is now part of routing. EVPN type-3 routes build one ingress replication list per VNI, which is why s3 stays out of the blue flood domain automatically.
  • Per-instance RDs with shared RTs use the same import/export idea as L3VPN, but here it is used for MAC addresses instead of IP prefixes.