The need for BGP within an AS typically occurs when the multiple routing policies exist, or when transit connectivity is provided between autonomous systems. In Figure 1-3, AS65200 provides transit connectivity to AS65100 and AS65300. AS65100 connects at R2, and AS65300 connects at R4.
Figure 1-3 AS65200 Provides Transit Connectivity
R2 could form a BGP session directly with R4, but R3 would not know where to route traffic from AS65100 or AS65300 when traffic from either AS reaches R3, as shown in Figure 1-4, because R3 would not have the appropriate route forwarding information for the destination traffic.
Figure 1-4 Transit Devices Need Full Routing Table
Advertising the full BGP table into an IGP is not a viable solution for the following reasons:
Scalability: The Internet at the time of this writing has 600,000+ IPv4 networks and continues to increase in size. IGPs cannot scale to that level of routes.
Custom Routing: Link state protocols and distance vector routing protocols use metric as the primary method for route selection. IGP protocols always use this routing pattern for path selection. BGP uses multiple steps to identify the best path and allows for BGP path attributes to manipulate the path for a specific prefix (NLRI). The path could be longer, which would normally be deemed suboptimal from an IGP protocol’s perspective.
Path Attributes: All the BGP path attributes cannot be maintained within IGP protocols. Only BGP is capable of maintaining the path attribute as the prefix is advertised from one edge of the AS to the other edge.
IBGP Full Mesh Requirement
It was explained earlier in this chapter how BGP uses the AS_PATH as a loop detection and prevention mechanism because the ASN is prepended when advertising to an EBGP neighbor. IBGP peers do not prepend their ASN to the AS_PATH, because the NLRIs would fail the validity check and would not install the prefix into the IP routing table.
No other method exists to detect loops with IBGP sessions, and RFC 4271 prohibits the advertisement of a NLRI received from an IBGP peer to another IBGP peer. RFC 4271 states that all BGP routers within a single AS must be fully meshed to provide a complete loop-free routing table and prevent traffic blackholing.
In Figure 1-5, R1, R2, and R3 are all within AS65100. R1 has an IBGP session with R2, and R2 has an IBGP session with R3. R1 advertises the 10.1.1.0/24 prefix to R2, which is processed and inserted into R2’s BGP table. R2 does not advertise the 10.1.1.0/24 NLRI to R3 because it received the prefix from an IBGP peer. To resolve this issue, R1 must form a multihop IBGP session so that R3 can receive the 10.1.1.0/24 prefix directly from R1. R1 connects to R3’s 10.1.23.3 IP address, and R3 connects to R1’s 10.1.12.1 IP address. R1 and R3 need a static route to the remote peering link, or R2 must advertise the 10.1.12.0/24 and 10.1.23.0/24 network into BGP.
Figure 1-5 IBGP Prefix Advertisement Behavior
Peering via Loopback Addresses
BGP sessions are sourced by the outbound interface toward the BGP peers IP address by default. Imagine three routers connected via a full mesh. In the event of a link failure on the R1-R3 link, R3’s BGP session with R1 times out and terminates. R3 loses connectivity to R1’s networks even though R1 and R3 could communicate through R2 (multihop path). The loss of connectivity occurs because IBGP does not advertise routes learned from another IBGP peer as in the previous section.
Two solutions exist to overcome the link failure:
Add a second link between all routers (3 links will become 6 links) and establish two BGP sessions between each router.
Configure an IGP protocol on the routers’ transit links, advertise loopback interfaces into the IGP, and then configure the BGP neighbors to establish a session to the remote router’s loopback address.
Of the two methods, the second is more efficient and preferable.
The loopback interface is virtual and always stays up. In the event of link failure, the session remains intact while the IGP finds another path to the loopback address and, in essence, turns a single-hop IBGP session into a multihop IBGP session.
Updating the BGP configuration to set the destination of the BGP session to the remote router’s loopback IP address is not enough. The source IP address of the BGP packets will still reflect the IP address of the outbound interface. When a BGP packet is received, the router correlates the source IP address of the packet to the BGP neighbor table. If the BGP packet source does not match an entry in the neighbor table, the packet cannot be associated to a neighbor and is discarded.
The source of BGP packets can be set statically to an interface’s primary IP address with the BGP session configuration command neighbor ip-address update-source interface-type interface-number on IOS nodes. IOS XR and NX-OS devices use the command update-source interface-type interface-number under the neighbor session within the BGP router configuration.
Figure 1-6 illustrates the concept of peering using loopback addresses after the 10.1.13.0/24 network link fails. R1 and R3 still maintain BGP session connectivity while routes learned from OSPF allow BGP communication traffic between the loopbacks via R2. R1 can still forward packets to R3 via R2 because R1 performs a recursive lookup to identify R2 as the next-hop address.
Figure 1-6 Link Failure with IBGP Sessions on Loopback Interfaces