Routing Protocol Design
On a VXLAN network, routes must be available for both the underlay network and overlay network. On the underlay network, BGP, OSPF, or IS-IS can be deployed and loopback addresses are used as the VTEP IP addresses. On the overlay network, EVPN must be deployed and spine nodes are configured as RRs. Table1 lists the advantages, disadvantages, and usage scenarios of the routing protocols.
Underlay Route |
Overlay Route |
Advantage and Disadvantage |
Recommended Usage Scenario |
|---|---|---|---|
OSPF |
IBGP-EVPN |
Advantage: easy deployment and fast convergence Disadvantage: limited number of OSPF routing domains |
Small- and medium-sized networks (fewer than 100 neighbors) |
IS-IS |
IBGP-EVPN |
Advantage: easy deployment, small network resource usage, good scalability, and fast convergence Disadvantage: poor customer acceptance |
Medium- and large-sized networks (fewer than 500 neighbors) |
EBGP |
IBGP-EVPN |
Advantage: independent routing domains for each area, controllable fault domain, flexible route control, good scalability Disadvantage: complex configuration |
Medium- and large-sized networks (fewer than 1000 neighbors) |
Underlay Routes
OSPF Routes
Figure 1 and Figure 2 show the spine/leaf two-layer distributed architecture and gateway/spine/leaf three-layer distributed architecture, respectively. On the two networks, all the devices are planned in OSPF area 0 and use the addresses of Layer 3 routing interfaces to establish OSPF neighbor relationships, implementing connectivity of the underlay network. The recommended network type is P2P.
EBGP Routes
Figure 3 shows the hardware distributed VXLAN network using the spine/leaf two-layer architecture. On this network, two spine nodes are planned in the same AS and each group of leaf nodes (in an M-LAG or stack) are planned in an AS. The devices use addresses of Layer 3 routed interfaces to establish EBGP peer relationships, and the underlay network advertises local VTEP IP addresses to ensure underlay route reachability.
In Figure 4, two border leaf nodes are planned in the same AS, two spine nodes are planned in the same AS, and each group of leaf nodes (in an M-LAG or stack) are planned in an AS. The devices use addresses of Layer 3 routed interfaces to establish EBGP peer relationships, and the underlay network advertises local VTEP IP addresses to ensure underlay route reachability.
IS-IS Routes
Figure 5 and Figure 6 show the hardware distributed VXLAN networks using the spine/leaf two-layer architecture and gateway/spine/leaf three-layer architecture. On the two networks, all the devices are planned in Level-2 and use the addresses of Layer 3 routed interfaces to establish IS-IS neighbor relationships. To increase the size of the local fabric, plan more devices in Level-1. To remain the size of the local fabric and interconnect fabrics in multiple cities, plan devices of each fabric in Leavl-1 and devices in the core area connecting the fabrics in Level-2.
Overlay Routes
The overlay network uses the IBGP EVPN routing protocol.
Figure 7 and Figure 8 show the hardware distributed VXLAN networks using the spine/leaf two-layer architecture and gateway/spine/leaf three-layer architecture. On the two networks, all the spine, leaf, and border leaf nodes are planned in the same AS, which differs from the ASs on the underlay network. The spine nodes function as RRs in BGP EVPN, and spine and leaf nodes establish EVPN peer relationships using independent loopback addresses.
When M-LAG is used, it is recommended that the loopback address of the DFS group be the same as the loopback address of the EVPN peer, but different from the VTEP IP address.
A static route is recommended between a gateway and an egress router. To transmit a large number of specific routes between them, a dynamic routing protocol can be used. When the underlay network uses OSPF, OSPF is recommended and an independent process is planned. BGP is the secondary choice to simplify configuration. When the underlay network uses EBGP, OSPF is recommended and BGP is the secondary choice (same AS number as that in EBGP) to reduce configuration complexity.
