From “flood and learn” to “tell and learn”
Lessons 1 and 2 showed VXLAN as a data plane: how frames are wrapped and carried. Flood-and-learn discovers MACs by flooding. EVPN adds a control plane: leaves tell each other which MACs, IPs and subnets live behind them, using BGP. When the applicable MAC/IP route has been advertised and imported, a leaf can use its forwarding table rather than flooding for an unknown location.
🗣️ Flood and learn
“Who has this MAC?” goes to everyone. The owner’s reply teaches the others. Works, but every unknown destination causes a flood, and nothing stops a mistake from spreading.
📣 EVPN (tell and learn)
When Leaf 1 is configured to advertise its locally learned MAC, it announces “MAC 00:AA…:01 is behind me (10.0.0.1)” in a BGP message. Other leaves can learn the location before they need to forward traffic to that host.
| Flood and learn | EVPN | |
|---|---|---|
| How MACs are found | Flooding + data-plane learning | BGP route advertisements |
| Unknown-unicast flooding | Common | Rare (most MACs known in advance) |
| ARP broadcasts | Flooded to all VTEPs | May be answered locally when ARP suppression is enabled and a valid MAC/IP binding exists; otherwise may be flooded |
| Routing between subnets | Needs an external gateway | Distributed anycast gateway and symmetric IRB; Type 2 host routes and optional Type 5 IP prefixes support different routing cases |
| Host moves | Wait for aging or a new flood | MAC mobility sequence number |
| Multi-vendor standard | Basic VXLAN only | RFC 7432, RFC 8365, RFC 9136 |
MP-BGP: BGP that carries more than IPv4
Plain BGP carries IPv4 routes. Multiprotocol BGP (MP-BGP) adds address families so one session can carry other kinds of information. EVPN is the L2VPN EVPN address family: AFI 25 (L2VPN), SAFI 70 (EVPN).
1. The underlay gets VTEPs reachable
First, an IGP or eBGP underlay makes every leaf loopback (10.0.0.1 … 10.0.0.4) reachable. This is the transport for VXLAN packets and for the BGP sessions.
2. The overlay BGP session
Leaves peer with the spine (route reflector) between loopbacks. The EVPN address family is activated on those sessions, and extended communities must be sent, because route targets travel as extended communities.
3. Next hop = VTEP loopback
Every EVPN route says “reach this MAC via next hop 10.0.0.x”. That next hop is the VTEP IP that becomes the outer destination IP of the VXLAN packet. Spines must not change it.
Two common session designs
Pick oneRoute Distinguisher (RD) and Route Target (RT)
These two names confuse almost everyone. Remember them with one sentence each:
🏷️ RD = “make it unique”
An RD is glued in front of a route so that identical-looking routes from different places stay separate entries in BGP. It is a label for uniqueness, not for filtering. Format: IP:number or ASN:number.
🎯 RT = “who wants it”
An RT is a tag (an extended community) that says which tenants/VNIs a route belongs to. A device exports routes with its RT and imports only routes whose RT it is configured to accept. It is a filter.
RT import filter
Change the tagsLeaf 1 advertises a MAC in VNI 10100 and attaches an export RT. Each other leaf imports the route only if the RT matches one of its import RTs.
Why unique RDs matter
Leaf 1 and Leaf 3 both advertise the same MACExample: a MAC reachable behind two leaves (dual-homed host, or an anycast gateway MAC). What does the route reflector keep?
Auto-derived RD and RT (NX-OS style)
Type your valuesLive walkthrough: sessions, Type 3, Type 2, ARP suppression, traffic
One complete EVPN lifecycle, seven steps. Purple chips are BGP messages. Press Next step, and watch the tables under the diagram fill in.
Watch EVPN build its tables, step by step
Packets in this step
Every EVPN route type, with a live-style example
EVPN defines several route types, each answering a different question. Pick a type to see what it is for, its fields, and what it looks like in show bgp l2vpn evpn on a leaf. Values use our lab: Host A is 00:AA:AA:AA:AA:01 / 192.168.100.11, VNI 10100.
| Type | Name | Question it answers | Used for |
|---|---|---|---|
| 1 | Ethernet Auto-Discovery | Which ESIs are alive? Who else serves this segment? | Multihoming: fast convergence, aliasing |
| 2 | MAC/IP Advertisement | Where is this MAC (and IP)? | Host reachability, ARP suppression, mobility |
| 3 | Inclusive Multicast Ethernet Tag | Which VTEPs are in this VNI, and how do I flood to them? | BUM flood list |
| 4 | Ethernet Segment | Who shares this Ethernet segment? | Multihoming: DF election |
| 5 | IP Prefix | Where is this subnet or prefix? | Inter-subnet / external routing |
| 6–11 | Specialized multicast membership, synchronization and PMSI/leaf discovery routes | Which receivers need multicast, and how is multicast delivery coordinated? | Selective multicast, membership synchronization, regional transport and explicit receiver tracking (feature-dependent) |
Type 2 deep dive: fields, MAC-only vs MAC+IP, and host mobility
Type 2 is the workhorse. Here is one announcement for Host A, field by field.
| Field | Example | Meaning |
|---|---|---|
| Route Distinguisher | 10.0.0.1:32867 | Keeps the route unique per leaf and VLAN |
| ESI | 0 (single-homed) | Non-zero when the host is multihomed |
| Ethernet tag | 0 | Usually 0 with VLAN-based service |
| MAC address | 00:AA:AA:AA:AA:01 | The host MAC |
| IP address (optional) | 192.168.100.11 | Present in MAC+IP routes; can support enabled ARP suppression with valid bindings |
| Label 1 (L2 VNI) | 10100 | VNI to put in the VXLAN header for bridging |
| Label 2 (L3 VNI) | 50001 | VNI of the tenant VRF, for routing (symmetric IRB) |
| Next hop | 10.0.0.1 | Outer destination IP for VXLAN |
| Route target | 65000:10100 (and 65000:50001) | Which VNI/VRF imports it |
| Encapsulation | VXLAN | Tunnel type extended community |
| Router MAC | 0C:01:01:01:01:01 | Inner destination MAC for routed traffic |
MAC-only route
Sent when the leaf has learned a MAC but no IP yet. Lets others bridge to it if imported and programmed.
MAC+IP route
Sent when the leaf also knows the IP (from ARP/GARP). Can support local proxy ARP (when enabled and the binding is valid) and host routing as configured.
Host mobility with the sequence number
Move Host A between leavesFor a single-homed host moving between leaves, the new leaf normally advertises the MAC with an increased MAC-mobility sequence number. Route preference and withdrawal follow EVPN mobility rules; multihoming and duplicate-MAC handling introduce additional conditions.
Choose the event: which EVPN route type solves it?
These are not trivia questions; they reflect operational symptoms. Select a scenario and predict which control-plane advertisement matters. The answer includes the expected forwarding state.
Configuration across vendors
Every vendor needs the same four ideas: (1) BGP session with the EVPN family on, (2) extended communities sent, (3) an RD and RT per VNI, (4) the VNI mapped to a VLAN or bridge.
Troubleshooting EVPN
Work from the bottom up: underlay → BGP session → routes received → routes imported → forwarding table.
| Layer | Question | Typical check |
|---|---|---|
| 1. Underlay | Can VTEP loopbacks reach each other (with the right MTU)? | ping from loopback to loopback, with DF bit and large size |
| 2. BGP | Is the L2VPN EVPN session established? | show bgp l2vpn evpn summary |
| 3. Received | Did the route arrive with the right next hop and RT? | show bgp l2vpn evpn route-type 2 |
| 4. Imported | Does the VNI import that RT? | compare RT export and import on both leaves |
| 5. Forwarding | Is the MAC or IP in the table with the right VTEP? | show l2route evpn mac all / show mac address-table |
What’s wrong? Pick a symptom
Troubleshooting helperTest what you learned
Ten questions, each with an explanation.
Mastery checklist
Tick each one when you can explain it in your own words.
Cheat sheet
- EVPN = BGP address family 25/70 that distributes MACs, IPs, prefixes
- RD makes routes unique; RT decides who imports
- Type 1 A-D · 2 MAC/IP · 3 IMET · 4 ES · 5 IP prefix
- Next hop = VTEP loopback = VXLAN outer destination
- Higher MAC-mobility sequence wins after a host move
- Send extended communities, or RTs never arrive
What’s next
Separate deep-dive lessons will cover routing with Type 5 and symmetric IRB and multihoming with Types 1 and 4.