I85PREP
⌘ K
JNCIP-DC study materials
JNCIP-DC study resources

JNCIP-DC: Data Center Architecture and Deployment

JNCIP-DC · Work through your exam, one topic at a time.

All guide sections

Data Center Architecture and Deployment

The Data Center Architecture and Deployment domain tests how you translate a requirement into a validated Juniper IP fabric: spine-leaf topology, underlay routing choices, overlay placement, port sizing, and Day-0/Day-1 bring-up tooling.

Underlay design rules

  • Every leaf connects to every spine (or a documented consistent subset) so the fabric is non-blocking and every uplink is ECMP-eligible; size uplinks from rack oversubscription targets.
  • Use /31 (IPv4) or /127 (IPv6) point-to-point links and a unique /32 loopback per device; advertise loopbacks into the underlay.
  • eBGP underlay is the Juniper-validated default: unique private ASN per leaf, shared ASN on spines (for example 65000), multipath multiple-as, BFD for sub-second failure detection.
  • Keep underlay and overlay separate. The overlay is typically iBGP with the spines as route reflectors; some designs use eBGP for both layers.

Overlay placement

  • VXLAN tunnels terminate on loopbacks, so loopback reachability is a hard prerequisite for any tenant traffic.
  • Pick MC-LAG for dual-homed hosts on non-EVPN Layer 2 designs; pick ESI-LAG when the fabric runs EVPN multihoming.

Worked example

Requirement: 2 spines, 4 leaves, dual-homed servers, L3 exit at a border leaf. Assign leaf ASNs 65001-65004 and spine ASN 65000. Each leaf peers eBGP with both spines and advertises its /32 loopback, giving two ECMP next hops per destination. The overlay runs iBGP with both spines as route reflectors and the leaves as clients. The border leaf originates a default route into tenant VRFs with an EVPN Type-5 route so every leaf has an exit without a full internet table.

Traps

  • Giving all leaves the same ASN in an eBGP overlay: BGP loop prevention stops re-advertisement of EVPN routes.
  • Forgetting the loopback advertisement: BGP sessions to spines are UP but VXLAN tunnels never form.
  • Underlay MTU below roughly 1600 bytes, which silently drops full-size tenant frames after encapsulation.
  • Mixing MC-LAG and ESI-LAG for the same server pair, producing duplicate forwarding.
  • Sizing uplinks without accounting for failure: N+1 spine capacity must still carry the load.