Choosing between IBGP and EBGP
Both IBGP and EBGP can be used for the purpose of peering between regions. Without any extensive policies between regions, and if a default route will be sent from each IBGP router into each region, you would have to use the registered AS number for the IBGP core. You need the registered AS number in the Internet connection because the core would send a default route into the regions. Also, the default route should originate only from the core. Because the core is all-knowing, it should maintain the Internet connection. In addition, you cannot peer with ISP via BGP if you do not have a registered AS number.
Redistribution of IBGP into IGP can create difficulties. If you decide to send BGP attributes, IBGP might not be a wise choice. For example, in Figure 16-13 you can see the physical connection: All the routers that are running IBGP must be logically full-meshed. Alpha.com is trying to implement policies and does not want to fully mesh BGP, which is against the standard.
Figure 16-13. Partial-Mesh IBGP
Figure 16-13. Partial-Mesh IBGP
Consider the situations that would complicate the issue. Suppose that you want optimal routing and want to redistribute an IBGP-learned route in your local IGP of the region. This method is not advisable because you risk the formation of routing loops. In Figure 16-14, USA.Cal.R1 router is sending routes to Euro.Lon.R1, and USA.Tex.R5 is sending routes to Euro.Fra.R2.
Figure 16-14. IBGP Route Advertisement between the United States and Europe
Figure 16-14. IBGP Route Advertisement between the United States and Europe
Alpha.com wants optimal routing, so Euro.Lon.R1 takes USA.Cal.R1 for subnet 172.16.{0-7}.0/21, and Euro.Fra.R2 takes USA.Tex.R5 for subnet 172.16.{8-25}.0/21. Router failure would cause the router to choose other paths. USA.Tex.R5 and USA.Cal.R1 must advertise different routes and send a shorter summary in case of failure. The configurations of routers would be the following:
router bgp 1
neighbor 172.16.64.2 remote-as USA.Tex.R5 )
neighbor 172.16.64.3 remote-as Euro.Lon.R1)
neighbor 172.16.64.5 remote-as Asia.Kla.R2)
network 172.16.0.0 mask 255.255 network 172.16.0.0 mask 255.255
1 (Loop back address of
1 (Loop back address of
1 (Loop back address of 248.0
240.0 {Shorter prefix}
Similarly, the configuration for USA.Tex.R5 would be sent a different summary than USA.Cal.R1, and would be sent via a less-specific route to Euro.Fra.R2:
router bgp 1
neighbor 172.16.64.1 remote-as 1 USA.Cal.R1 )
neighbor 172.16.64.4 remote-as 1 Euro.Fra.R2)
neighbor 172.16.64. 6 remote-as 1 Asia.Sin.R1)
network 172.16.8.0 mask 255.255.24
network 172.16.0.0 mask 255.255.240.0
(Loop back address of (Loop back address of (Loop back address of
{Shorter prefix}
Note that all the peering addresses on the BGP routers are loopback addresses with a /32 mask, so the backbone uses the same subnet with a /32 on all the routers without wasting address space.
Try to leak loopback addresses to each BGP neighbor with a /32 as an IBGP route. Otherwise, BGP will flap. There are two ways to do this: Either send the /32 route with your existing IGP (which would increase the individual IGP of the region into other regions), or run a separate instance of IGP just to carry the next hop information, as shown in Figure 16-15.
Continue reading here: Figure 1615 Separate IGP to Carry Next Hop Information
Was this article helpful?