Convergence and High Availability · Lesson 01 of 01
Failing over faster than the timers
Default BGP convergence is measured in minutes. BFD, next-hop tracking and PIC bring it down to sub-second when the design needs it.
Written lesson
Why default is slow
The default hold timer is 180 seconds — BGP will wait that long before declaring a silent neighbour dead. Even with tuned keepalives, detecting a failure through BGP alone is slow. The faster path is to detect the failure elsewhere and tell BGP immediately: link-down events, IGP next-hop withdrawal, or a dedicated liveness protocol. BGP timers are the last resort, not the first line of defence.
Written lesson
BFD, next-hop tracking, PIC
BFD (Bidirectional Forwarding Detection) is a lightweight hello protocol that detects a path failure in milliseconds and signals BGP to tear the session down at once, bypassing the hold timer. BGP next-hop tracking watches the RIB and reacts the instant the IGP withdraws the route to a next hop, rather than waiting. PIC (Prefix Independent Convergence) pre-installs a backup path in the FIB so failover does not depend on reconverging every prefix one by one — the whole table switches to the backup in one operation.
Configuration
BFD for a BGP neighbour
interface GigabitEthernet0/0
bfd interval 300 min_rx 300 multiplier 3
!
router bgp 65001
neighbor 10.12.0.2 fall-over bfdGraceful Restart is the opposite goal
GR and NSR keep forwarding alive while the control plane restarts — they hide a restart rather than speed a failover. Use fast-failover tools for real failures and GR/NSR for planned or software-level control-plane events.