Network Virtualization
Network Virtualization
This domain tests overlay design: how tenants get isolated L2/L3 topologies over a shared IP fabric, and how control planes scale. Expect EVPN/VXLAN, VNI, route-target, VTEP, and MTU questions.
Core rules
- An overlay carries tenant traffic in tunnels such as VXLAN, Geneve, or MPLS-over-UDP/GRE across an underlay IP fabric. The underlay routes VTEP loopbacks; the overlay provides tenant semantics.
- VXLAN uses a 24-bit VNI, about 16 million segments, and a UDP header. The outer Ethernet/IP/UDP/VXLAN overhead is 50 bytes, so underlay MTU must accommodate it or fragmentation occurs.
- EVPN is the control plane: MP-BGP advertises MAC/IP (Type 2), inclusive multicast (Type 3), and IP prefix (Type 5) routes. Route targets control import and export and therefore tenant isolation.
- Contrail uses EVPN/VXLAN with vRouter; MPLS over GRE/UDP is also supported. Anycast gateway gives the same distributed default gateway on every leaf.
Worked example
Two tenants both use 10.0.0.0/24. Assign VNI 10010 and route target 10010:10010 to tenant A; assign VNI 10020 and route target 10020:10020 to tenant B. Reasoning: identical prefixes are fine because VRFs and route targets keep them separate; Type 5 routes carry the prefixes inside each VRF. If both used the same VNI and route target, EVPN would leak routes and MACs between tenants.
Traps
- Confusing VLAN ID with VNI. They are different namespaces; mapping is local.
- Sizing underlay MTU for 1500-byte tenant frames without tunnel overhead.
- Assuming EVPN is data-plane encapsulation. EVPN is the BGP control plane; VXLAN is the data plane.