Data Center Architecture and Networking Fundamentals
This domain tests how Juniper data center fabrics are built: spine-leaf topology, the routed IP underlay, the VXLAN/EVPN overlay, and multi-chassis redundancy.
Underlay and overlay
A Juniper data center design separates the physical IP fabric (underlay) from the virtual network (overlay). The underlay is typically a spine-leaf Clos with routed links and BGP; each leaf peers with every spine, and ECMP spreads traffic across equal-cost paths. The overlay uses VXLAN to carry Layer 2 frames in UDP, with EVPN as the control plane for MAC/IP learning and VTEP discovery. This avoids spanning tree and lets workloads move without re-addressing.
Core rules
- Spine switches provide high-capacity transit; leaves connect servers and perform VXLAN encapsulation.
- Use unique AS numbers per leaf in eBGP underlay, or iBGP with route reflectors.
- Anycast gateway (IRB) lets the same default gateway IP/MAC exist on every leaf.
- MC-LAG provides dual-homing for servers or switches without STP.
- VXLAN VNI maps to a VLAN/bridge domain; EVPN route types 2 (MAC/IP), 3 (IMET), and 5 (IP prefix) matter.
Worked example
Server A on leaf1 sends to Server B on leaf2 in the same VXLAN. Leaf1 learns A's MAC on its local access interface, advertises an EVPN type-2 route with the VNI and next hop (leaf1 VTEP), and encapsulates the frame in VXLAN. The underlay routes the outer packet to leaf2's VTEP. Leaf2 decapsulates, looks up the VNI and MAC, and forwards to B. Return traffic uses leaf2's advertisement. ECMP hashing uses the outer UDP source port to spread flows.
Traps
Do not confuse the underlay BGP session with the EVPN overlay session. Do not stretch VLANs across the fabric with STP; the fabric is routed. Remember that VXLAN adds 50 bytes of overhead, so MTU must be increased. A common error is assuming EVPN requires a multicast underlay; ingress replication is also valid.