Cloud-Grade Routing: Control Plane, Convergence, and Segmentation
Core concepts
Cloud-grade routing blends familiar protocols with overlay control planes. Candidates should understand underlay versus overlay: the physical/IP underlay provides reachability, while the overlay (VXLAN/GENEVE, EVPN control plane) carries tenant or segment identity. EVPN is a BGP address family used to distribute MAC and IP reachability, replacing flood-and-learn with a control-plane-driven model. Know route types conceptually - MAC/IP advertisement, inclusive multicast for BUM traffic, and prefix routes for summarization. Understand BGP best-path selection at a high level (weight, local preference, AS-path length, origin, MED, eBGP over iBGP) and why local preference is the usual traffic-engineering lever. Convergence depends on fast failure detection (BFD), graceful restart, and dampening of flapping links. Segmentation uses VRFs and route targets: import/export communities decide which prefixes leak between VRFs.
Traps
Questions conflate data-plane flooding with control-plane learning. Another trap: assuming longest-prefix match beats administrative distance or local preference - it does not; those are evaluated first. Route-target mismatch is a favorite: the VRF imports the wrong target and prefixes are simply absent. Also watch for asymmetric return paths after a link failure and for summarization that blackholes more-specific traffic.
Worked example
A tenant prefix is missing on a remote leaf. Check order: (1) is the route in the local VRF? (2) does the route target match the remote VRF's import policy? (3) is the EVPN route advertised with a valid next hop? Rule of thumb: follow the route target before you follow the cable - most "routing" failures in EVPN fabrics are policy/import mismatches.
Rule of thumb
For any routing question, ask: which table, which policy, which direction?