BGP configuration considerations: - subdivide single "public" AS (100) into multiple internal sub-AS (BGP confederation) - spine layer: all switches reside in one AS (65000) - leaf layer: every switch resides in its own AS (65001, 65002, 65003, 65004) - direct static peering between leaf and spine nodes (interfaces of interconnection network, no loopbacks) - allow load balancing across multiple BGP paths with equal costs (=> ECMP) - maximum path count equals number of used spine switches - only one AS at spine layer => no BGP multipath relaxing necessary - every BGP router is used as next hop in its advertised/forwarded routing messages ip subnet for l3-switch interconnection networks: 10.20.30.0/27 ------------------------------------------------- 10.20.30.0/30 veos-leaf-1 <> veos-spine-1 .1 veos-spine-1 Ethernet1 .2 veos-leaf-1 Ethernet1 10.20.30.4/30 veos-leaf-1 <> veos-spine-2 .5 veos-spine-2 Ethernet1 .6 veos-leaf-1 Ethernet2 10.20.30.8/30 veos-leaf-2 <> veos-spine-1 .9 veos-spine-1 Ethernet2 .10 veos-leaf-2 Ethernet1 10.20.30.12/30 veos-leaf-2 <> veos-spine-2 .13 veos-spine-2 Ethernet2 .14 veos-leaf-2 Ethernet2 10.20.30.16/30 veos-leaf-3 <> veos-spine-1 .17 veos-spine-1 Ethernet3 .18 veos-leaf-3 Ethernet1 10.20.30.20/30 veos-leaf-3 <> veos-spine-2 .21 veos-spine-2 Ethernet3 .22 veos-leaf-3 Ethernet2 10.20.30.24/30 veos-leaf-4 <> veos-spine-1 .25 veos-spine-1 Ethernet4 .26 veos-leaf-4 Ethernet1 10.20.30.28/30 veos-leaf-4 <> veos-spine-2 .29 veos-spine-2 Ethernet4 .30 veos-leaf-4 Ethernet2 l3-switch networks/ip subnets: (advertised using BGP) ------------------------------------------------- veos-leaf-1 172.23.24.0/24 veos-leaf-1 Ethernet3 172.23.24.1 lxc-iperf-1 172.23.24.10 veos-leaf-2 172.23.25.0/24 veos-leaf-2 Ethernet3 172.23.25.1 lxc-iperf-2 172.23.25.10 veos-leaf-3 172.23.26.0/24 veos-leaf-3 Ethernet3 172.23.26.1 lxc-iperf-3 172.23.26.10 veos-leaf-4 172.23.27.0/24 veos-leaf-4 Ethernet3 172.23.27.1 lxc-iperf-4 172.23.27.10