Interview questions¶
A self-test bank, organised by topic and roughly easy → hard. Answers are collapsed — try to answer out loud first, then click to check. If you can explain the ⭐ ones cleanly, you're in good shape for a data-center design interview.
Fundamentals¶
What problem does VXLAN solve that VLANs cannot?
VLANs are limited to 4094 segments (12-bit tag) and rely on spanning tree, which blocks redundant links. VXLAN gives ~16M segments (24-bit VNI) and runs L2 over a routed L3 underlay, so all links are active (ECMP) and L2 segments can stretch across a whole fabric.
⭐ In one sentence each, what do VXLAN and EVPN do?
VXLAN is the data plane — it encapsulates L2 frames in UDP/IP to carry them across an L3 network. EVPN is the control plane — a BGP address family that advertises where each MAC/IP lives, so tunnels are built from knowledge instead of flooding.
What is a VTEP?
VXLAN Tunnel EndPoint — the device (a leaf in spine-leaf) that encapsulates tenant frames into VXLAN and decapsulates them at the far end. It has tenant VLANs on one side and a loopback (the tunnel source) on the other.
Underlay¶
What is the only job of the underlay?
To provide reachability between VTEP loopbacks, with equal-cost multipath. Nothing tenant-specific lives in the underlay.
Which protocols can run the underlay, and what's the trade-off?
OSPF, IS-IS, or eBGP. OSPF/IS-IS are simple IGPs, quick to stand up. eBGP scales better and lets one protocol (BGP) do both underlay and overlay, at the cost of more per-link AS/peer config. (These labs compare all of them.)
Why are /31s used on fabric links?
Point-to-point links only need two addresses; a /31 uses exactly two and wastes none, versus a /30 that wastes two per link.
⭐ Why must the underlay use jumbo MTU?
VXLAN adds ~50 bytes of headers. A 1500-byte tenant frame becomes ~1550 on the wire; a standard 1500 MTU would drop or fragment it. Fabric links run 9000/9216 so the encapsulated packet fits.
VXLAN data plane¶
Name the headers VXLAN adds and which one the spines route on.
Outer Ethernet, Outer IP, UDP, VXLAN (VNI). The outer IP (source VTEP loopback → dest VTEP loopback) is what the underlay/spines route on — spines never inspect the inner frame.
What is the VXLAN UDP destination port?
4789 (IANA). It signals "this is VXLAN" to the receiving VTEP.
⭐ Why is the outer UDP source port important?
It's not a real port — the VTEP sets it to a hash of the inner flow. This gives the underlay per-flow entropy so ECMP spreads different flows across all spine links. It's where the load-balancing happens.
What is BUM traffic and how is it handled in EVPN-VXLAN?
Broadcast, Unknown-unicast, Multicast. EVPN handles it by default with ingress replication (head-end replication): the ingress VTEP unicasts a copy to each remote VTEP in that VNI. The replication list is built from Type-3 routes. (Alternative: underlay multicast.)
Difference between an L2VNI and an L3VNI?
L2VNI maps to a VLAN and is used for bridging (same subnet across leaves). L3VNI is tied to a VRF and used for routing between subnets / to the outside (Type-5 routes, IRB).
EVPN control plane¶
What does EVPN replace, and with what?
It replaces L2 flood-and-learn with BGP advertising: VTEPs announce their local MACs/IPs to each other instead of discovering them by flooding.
EVPN is carried in which protocol, as what?
MP-BGP, as an address family (AFI 25 L2VPN, SAFI 70 EVPN). In Junos:
family evpn signaling.
⭐ Explain Route Distinguisher vs Route Target.
RD makes each VTEP's routes unique in BGP (so two leaves advertising the same subnet don't collide) — typically different per VTEP. RT is an extended community controlling import/export — same RT = same virtual network, so it's shared across VTEPs in a VNI. RD = uniqueness, RT = membership.
What is ARP suppression and why does it help?
The local leaf answers ARP requests from its EVPN (Type-2) table instead of flooding them across the fabric. It cuts broadcast traffic and speeds up resolution — the ARP never leaves the ingress leaf.
iBGP vs eBGP for the overlay — when would you use each?
iBGP: all VTEPs one AS; simple, but needs a full mesh or route reflectors at scale (spines as RRs). eBGP: each leaf its own AS; scales well and lets one protocol do underlay + overlay, common in large fabrics. More per-peer config.
Route types¶
⭐ Which EVPN route type appears first, before any host sends traffic, and why?
Type-3 (Inclusive Multicast / IMET) — advertised when a VTEP has a VNI with an active member. VTEPs use these to discover each other and build the BUM flood list; the VXLAN tunnel forms off the back of it. Type-2 only appears once a host is actually learned.
What does a single Type-2 route enable?
Remote MAC learning (install the MAC pointing at the origin VTEP's tunnel) and the IP portion feeds ARP suppression. On a host move, a new Type-2 reconverges quickly.
When do you need a Type-5 route?
For routing rather than bridging — inter-subnet/inter-VNI routing via an L3VNI, reaching external prefixes, or summarisation. Type-5 carries an IP prefix with no MAC.
Which route types are involved in multihoming?
Type-1 (Ethernet A-D, for fast failover/aliasing) and Type-4 (Ethernet Segment, for Designated Forwarder election).
Design & scale¶
Why spine-leaf instead of traditional three-tier?
Predictable any-to-any latency (every leaf is two hops from any other), horizontal scale (add leaves/spines), and full ECMP — no blocked links.
At scale, why not full-mesh iBGP for the overlay?
N leaves need N(N-1)/2 sessions — it explodes. Use spines as route reflectors (or eBGP) so each leaf peers only with the spines.
⭐ What is an anycast gateway and why use it?
The same gateway IP and MAC configured on every leaf for a subnet. A host always uses its local leaf as the default gateway regardless of which leaf it's on — so routing is optimal and VM mobility doesn't break the gateway.
Symmetric vs asymmetric IRB?
Both do inter-subnet routing. Symmetric uses a shared L3VNI and routes on both ingress and egress leaf (route-bridge / bridge-route symmetrical) — scales better and is the modern default. Asymmetric requires every leaf to have every VNI. Symmetric is generally preferred.
Troubleshooting¶
The BGP-EVPN session is up but bgp.evpn.0 has 0 routes. First checks?
Often expected if no VNI has an active member yet. Check: is a VLAN mapped to a VNI? Is there an up interface in that VLAN? Is the VNI in the extended-vni-list? Do the RTs match between VTEPs?
Hosts can't ping across the fabric but the tunnel is up. Where do you look?
Check Type-2 routes exist for both hosts (show evpn database), MAC learned
remotely (show ethernet-switching table → remote/DR flag), RT import/export
match, correct VLAN↔VNI mapping on both leaves, and MTU on fabric links.
Ping works but large transfers fail. Likely cause?
MTU. Small packets fit; full-size frames + 50 bytes of VXLAN overhead exceed a 1500 underlay MTU. Enable jumbo frames on fabric links.
Juniper-specific (from this lab)¶
⭐ On Junos, you configured a VNI but no Type-3 route appeared. Bug?
No — Junos only originates the Type-3 (IMET) route once the VLAN has an operationally-up member interface. With no access port in the VLAN, there's nothing to flood to, so nothing is advertised. Add an access port and the Type-3 (and tunnel) appear. (Cisco NX-OS advertises as soon as the VNI is defined — a real behavioural difference.)
Which single interface sources the VXLAN tunnel on a Junos leaf?
lo0.0, via set switch-options vtep-source-interface lo0.0. Junos uses one
loopback for router-id, BGP peering, and VTEP source (unlike some platforms
that use a separate VTEP loopback).
What does the DR flag mean in show ethernet-switching table?
Dynamic + Remote — a MAC learned dynamically via a remote VTEP over
the VXLAN tunnel (vtep.xxxxx), as opposed to a locally-learned MAC on a
physical port.
How do you read the type of an EVPN route like 3:10.0.0.21:1::10100::10.0.0.21?
The leading digit is the route type (here 3 = IMET). Then RD
(10.0.0.21:1), then the VNI (10100), then the originator. Reading that
first digit tells you instantly what kind of announcement it is.
How to use this bank¶
- First pass: read a question, answer aloud, then reveal. Mark the ones you fumbled.
- Second pass: only the ⭐ questions — these are the ones interviewers dig into.
- Tie it to the lab: every answer here has a matching
showcommand you ran in lab 01. Seeing it live cements the theory.