Users Guide

BGP 1369
Detecting Loss of Adjacency
Dell EMC Networking optionally drops an adjacency with an external peer
when the routing interface to that peer goes down. This behavior can be
enabled globally or on specific interfaces using the bgp fast-external-fallover
and ip bgp fast-external-fallover commands. BGP accomplishes this behavior
by listening to router events. When BGP gets a routing interface down event,
BGP drops the adjacency with all external peers whose IPv4 address is in one
of the subnets on the failed interface. Both fast-external-fallover and fast-
internal-fallover are enabled by default.
Dell EMC Networking also offers an option to quickly detect loss of
reachability to internal peers, and drop the BGP adjacency when such a loss
occurs. Because internal peers are often not on a local subnet (and an internal
peer can be reached through multiple local interfaces), BGP cannot
determine internal peer reachability based on local link state. Instead, when
this feature is enabled, BGP registers for address resolution changes for each
internal peer's IPv4 address. When a peer's address becomes unreachable (i.e.,
the route table manager deletes the route to the peer and no non-default
route to the peer remains), BGP drops the adjacency to the peer. BGP
considers an internal peer to be unreachable if the only route to the peer is a
default route. This feature can enabled or disabled globally for all internal
peers using bgp fast-internal-fallover. Because internal peers are not
associated with a single interface, there is no interface configuration option.
When fast fallover is disabled for a peer, the adjacency remains in the
ESTABLISHED state until the hold timer expires. When connectivity to the
peer is lost, the BGP Next Hop for routes learned from affected peers
becomes unreachable. This change makes the routes unusable, and BGP
immediately removes them from the routing table. So even without the fast
fallover behavior enabled, the routing table reacts quickly to changes in local
interface state. However, when the adjacency remains in ESTABLISHED
state even though the neighbor is unreachable, BGP cannot send UPDATE
messages to the neighbor. If the link is restored before the dead interval
expires, there is no event to cause BGP to resend the failed UPDATEs.
Because BGP does not periodically refresh routing state, and the loss of
UPDATEs is permanent. To avoid this situation, when an UPDATE message
fails to be sent to any member of an outbound update group, BGP
reschedules the update send process to resend the data. Thus, having a
neighbor in an ESTABLISHED but unreachable state causes duplicate data
to be sent to other members of the update group. With fast fallover enabled,