Cisco 350-401: VXLAN Underlay and Overlay

For candidates preparing for 350-401 ENCOR v1.2, the published virtualization tasks are to describe LISP and VXLAN concepts, not to implement a production BGP EVPN fabric. The detailed underlay/overlay example in this article is a valid data-center VXLAN EVPN design: its routed underlay provides VTEP and BGP-peer reachability, while BGP EVPN distributes overlay endpoint information. A Cisco SD-Access campus fabric differs: Cisco Catalyst Center manages the solution, LISP serves as the control plane, and VXLAN carries the data plane with TrustSec policy context. Keeping those architectures distinct is central to sound Cisco network engineering and avoids mistaking optional Nexus EVPN labs for mandatory ENCOR v1.2 configuration work.

The distinction is more than terminology. Traditional VLAN designs often mix endpoint reachability with the same Layer 2 failure domain that carries control traffic. VXLAN creates a routed fabric underneath and a virtual network above it. That architectural shift is why VLAN-to-VXLAN transition changes scaling behavior rather than simply increasing the VLAN number space.

The underlay should be boring, fast, and fully reachable

Cisco’s current VXLAN EVPN guidance treats the underlay as the routed network that provides reachability among VTEPs and control-plane endpoints. Common choices include OSPF, IS-IS, or eBGP. The best choice is the one the team can operate consistently; the overlay does not become more reliable because the underlay uses a fashionable protocol. What matters is deterministic convergence, equal-cost paths, clear failure domains, and complete loopback reachability.

Every leaf-spine link should have an explicit MTU strategy. VXLAN adds encapsulation overhead, so an underlay that was sized only for endpoint frames can create fragmentation or drops when the overlay starts carrying larger packets. MTU inconsistency is especially painful because BGP and basic pings may work while application traffic fails only at particular sizes.

ECMP is another foundational property. The fabric should have multiple equal-cost paths and the forwarding hardware should distribute flows predictably. Unequal routing or hidden policy-based routing in the underlay can create one-way reachability that masquerades as an EVPN problem.

Underlay addressing deserves the same discipline as routing. Point-to-point links, loopbacks, and infrastructure prefixes should have a predictable plan that makes failures easy to localize. Designers should decide whether link addressing is numbered or unnumbered, how loopbacks are advertised, and whether infrastructure prefixes are summarized. The goal is not address conservation by itself; it is making the fabric easy to reason about during failure. If every leaf and spine follows the same convention, an engineer can infer intended reachability from the topology. Irregular exceptions turn a simple routed fabric into a collection of special cases that later masquerade as overlay faults.

VTEPs are the seam between the two systems

The VXLAN Tunnel Endpoint is where an endpoint-facing segment becomes an overlay object. A VTEP maps local VLANs or bridge domains into VXLAN Network Identifiers, encapsulates frames or routed traffic, and sends the result across the IP fabric. Troubleshooting therefore needs both local attachment context and remote VTEP reachability. If the local endpoint never reaches the correct VTEP context, the overlay cannot repair that error.

VXLAN separates data-plane encapsulation from the control plane that tells VTEPs where remote endpoints and prefixes live. Its VXLAN fundamentals therefore matter operationally: the VNI identifies a virtual network, but it does not by itself provide endpoint discovery, policy, or routing.

Route reflectors or EVPN peering design can become a hidden scaling boundary. Centralized route reflectors simplify session count but concentrate control-plane dependency. Distributed designs reduce some concentration but can increase configuration complexity. Capacity planning should include the number of EVPN routes, update churn, convergence during maintenance, and what happens when one reflector disappears. A fabric should be tested during a control-plane failure while endpoint traffic is active. If established forwarding survives but new endpoint learning fails, the operational symptom will differ from a total outage, and runbooks should recognize that distinction.

BGP EVPN distributes reachability instead of flooding everything

EVPN gives the fabric a control plane for endpoint and prefix reachability. Rather than depending on data-plane learning across the whole fabric, VTEPs advertise MAC, IP, and prefix information in defined route types. This reduces unnecessary flooding and gives operators a structured way to ask whether the control plane actually knows the remote destination.

That does not eliminate BUM traffic. Broadcast, unknown-unicast, and multicast behavior still needs a replication design. Depending on platform and topology, a fabric may use ingress replication or multicast-assisted replication. The important operational point is to know which model is configured and how to verify it. Teams often troubleshoot missing ARP or neighbor discovery as a host issue when the real problem is replication state.

In VXLAN EVPN, reducing Layer 2 scope improves failure containment, but the control plane becomes more important and must be observable. The design choice is therefore not simply “more overlay” versus “less overlay”; it is where reachability state lives and how quickly operators can prove it is correct.

Anycast gateways improve host mobility because the same default-gateway address and MAC can exist on multiple VTEPs, but they increase the importance of consistent configuration. A single leaf with a different gateway MAC, subnet definition, or VRF association can create intermittent behavior that follows the endpoint’s attachment point. Automation should therefore validate fabric-wide invariants before deployment. A configuration system that merely pushes text without checking consistency can replicate the same defect to every leaf faster than a human could create it manually.

Layer 2 and Layer 3 VNIs solve different reachability problems

A VXLAN fabric may carry bridged segments with Layer 2 VNIs and route between tenant networks using Layer 3 VNIs and distributed anycast gateways. Confusing those roles leads to configuration that appears plausible but sends traffic into the wrong routing context. Operators should be able to map an endpoint to its VLAN or bridge domain, L2 VNI, VRF, L3 VNI, local gateway, and remote next-hop behavior.

Symmetric and asymmetric integrated routing and bridging designs also place different state requirements on VTEPs. A team should not copy configuration fragments without understanding the selected forwarding model. The right verification commands and route expectations depend on that design.

Operational teams should also distinguish endpoint mobility from MAC instability. Legitimate virtual-machine movement, container scheduling, or dual-attached hosts can change the location of a MAC or IP. Repeated movement beyond expected patterns may indicate a loop, duplicate address, or misconfigured multihoming. Baselines for MAC move frequency and EVPN route updates make those conditions visible. Without baselines, engineers may either ignore dangerous churn or overreact to normal workload mobility.

Overlay failures should be diagnosed from control plane to data plane

When two endpoints cannot communicate, first confirm local attachment and VLAN/VNI mapping. Then verify the underlay can reach the remote VTEP loopback. Next inspect EVPN routes for the endpoint or prefix, including next hop and route target. Finally inspect VXLAN forwarding state and packet counters. This sequence avoids spending time on encapsulation when the control plane never installed the destination.

Route-target errors are a classic example. BGP sessions can be established and EVPN routes can exist globally, yet a tenant VRF may not import the route because the route target is wrong. The symptom is “BGP looks healthy” while the tenant remains isolated. That is why verification must be done in the actual VRF and VNI context.

Operational documentation should include a fast underlay/overlay decision tree. For example, if VTEP loopbacks cannot ping with the intended source, stay in the underlay. If loopbacks are reachable but the expected EVPN route is absent, inspect BGP control-plane import and advertisement. If the route exists but traffic fails, inspect VNI state, next-hop resolution, replication, and local attachment. This simple branch prevents a common outage pattern in which several engineers troubleshoot different layers simultaneously and change unrelated configuration. The fabric becomes supportable when everyone uses the same evidence sequence.

Failure domains should be explicit in the design

A routed underlay limits Layer 2 blast radius, but the overlay introduces new shared dependencies: route reflectors, anycast gateway configuration, common policy, fabric automation, and VTEP software. Redundancy should therefore include control-plane architecture, not only extra physical links. Route reflectors need capacity and failure planning; automation systems need change validation; and fabric upgrades should preserve EVPN interoperability.

Organizations operating EVPN services still have to choose the service boundary deliberately. VLAN extension should be justified by an application or mobility requirement rather than becoming the default response to every connectivity request.

Change sequencing matters during upgrades. Leaf and spine software, route-reflector versions, and feature support do not always move together. Maintenance plans should document protocol interoperability, graceful-restart expectations, and the maximum number of devices that can be unavailable without breaking path diversity. A lab or canary fabric is useful for validating new EVPN behavior before a fleet-wide rollout. Because the underlay and overlay have different failure modes, upgrade testing should verify both adjacency/convergence and tenant reachability rather than relying on device health alone.

A healthy fabric is observable at both layers

Useful operational telemetry includes underlay adjacency state, ECMP path changes, VTEP loopback reachability, BGP EVPN session health, EVPN route counts by type, VNI state, MAC movement, ARP or neighbor suppression behavior, and drops related to MTU or encapsulation. Baselines are valuable because route counts and MAC movement often reveal a fabric problem before applications report an outage.

Cisco platforms provide the mechanisms, but the design principle is vendor-independent: make the underlay simple enough that it can be proven healthy quickly, and make the overlay explicit enough that every tenant path can be traced. The Cisco fabric becomes much easier to operate when engineers resist treating VXLAN as one large feature and instead validate transport, control plane, and tenant forwarding in that order.

Leave a Reply

How It Works

img
Step 1. Choose Exam
on ExamLabs
Download IT Exams Questions & Answers
img
Step 2. Open Exam with
Avanset Exam Simulator
Press here to download VCE Exam Simulator that simulates real exam environment
img
Step 3. Study
& Pass
IT Exams Anywhere, Anytime!