EVPN Control Plane Up, Data Plane Down Lab
Diagnose size-dependent EVPN/VXLAN loss through eight interactive stages and ten evidence checkpoints.
Synthetic educational scenario. No live device access, customer captures or real incident reproduction. Browser memory only; reset clears learning progress.
Reviewed: 2026-10-16
1. Symptom
CE-A198.51.100.10/24 and CE-B198.51.100.20/24 share VLAN100/VNI10100/BD100 across two leaves.1472-byte ICMPv4 echo payload with DF fails in both directions;32-byte probes pass. This is size-dependent loss, not every ping failing.
CE-A ↔ LeafA/VTEP192.0.2.11 ↔ Underlay ↔ LeafB/VTEP192.0.2.12 ↔ CE-B
2. Known Good Evidence
BGP EVPN is Established; expected MAC routes are imported and installed; VNI10100 is up. These facts establish signaling context. They do not test the complete encapsulated packet envelope.
3. Ask: next check?
Select a check, reveal its observation and explain what it can and cannot establish. Inspect all ten distinct checkpoints before narrowing the fault domain.
4. Reveal evidence
5. Narrow fault domain
Use the complete evidence set to choose a fault domain. Healthy small packets and service mapping must be reconciled with size-correlated drops.
6. Final diagnosis
Choose the cause that accounts for1422/1423-byte boundary behavior and the observed egress MTU drops.
7. Verification
Educational simulation only. First validate every relevant link/path and both directions plus the platform IP-MTU convention. The post-fix1600 profile is one fixture, not a universal recommendation. Reducing service MTU is an alternative with endpoint/application trade-offs.
8. Lessons learned
Completion requires the supported diagnosis and a simulated verification run. Healthy control-plane state does not prove delivery; evidence must connect endpoint resolution, service mapping, actual encapsulation, packet-size boundaries and correlated drops.
- Keep packet-size/DF/direction context with every probe.
- Treat IPv6 ND and anycast as conditional checks, not universal suspects.
- Revalidate the full path and recovery after a controlled change.
Read the separate incident-style case review
Continue learning
Related course: Implementing Data Center Fabric with EVPN and VXLAN On-Demand — Optional external official paid Juniper/Junos-specific course; access and terms per provider. This lab remains vendor-neutral.
Related topic: Network Monitoring & Observability — General evidence-collection background, not an EVPN-specific topic.
Related tool: MTU & MSS Calculator — Packet-sizing support; it does not prove the fault in this lab.
Source ledger and modeled facts
S-VXLAN — RFC7348 — VXLAN · §4.3 physical MTU/fragmentation; §5 frame format
S-EVPN — RFC8365 — EVPN network virtualization overlay · §5 EVPN with VXLAN; control-plane route interpretation
S-ARP — RFC7432 — EVPN · §10 ARP/ND; MAC-only versus MAC/IP routes
S-IP — RFC791 — IPv4 · §3.1 IHL, total length and DF
S-ICMP — RFC792 — ICMP · Echo/Echo Reply format; fragmentation-needed message
S-UDP — RFC768 — UDP · UDP header format
S-FIXTURE: all captures/counters are synthetic educational observations. S-MODEL: arithmetic and teaching gates are local model rules, not universal thresholds.
Does an Established EVPN session prove data-plane delivery?
No. Check service forwarding and the actual encapsulated packet path, including MTU.