From CT3

Cisco IOS implementation of Border Gateway Protocol (BGP) has always tried to detect external BGP neighbor loss through signaled by the loss of a directly connected interface. The fast external failover (enabled by default) works only for directly connected EBGP neighbors; IOS relied on BGP keepalives to detect loss of IBGP neighbor or multihop EBGP neighbor.

IOS releases 12.0S, 12.2S, 12.3T and 12.4 introduced BGP support for fast peering deactivation. When the fall-over is configured on a BGP neighbor, the router tracks the availability of the neighbor’s IP address in the IP routing table and drops the BGP session as soon as the BGP neighbor’s IP address is no longer reachable.

In networks using summary routes or default routing, you have to configure a route map which matches potential BGP next-hops to prevent the router from using a default or a summary route as the potential path toward BGP neighbors.

Read more in the Designing Fast Converging BGP Networks IP corner article ...

Fast Session Deactivation is too aggressive in OSPF or IS-IS networks

The IP corner article mentions that this feature might be too aggressive in environments with mixed routing protocols due to a small interval between the loss of primary IP route and installation of backup IP route. However, BGP sessions might be lost in pure link-state IGP environments if a directly connected interface of a BGP router fails.

The following topology will be used to illustrate this undesired behavior:

Figure 1: Test network topology

Read more ...