can be processed by ASIC in lieu of CPU (≈ 50ms detection)
informs client process that peer is down
client process is responsible for discovering peer: first negotiation, then keepalive
unicast MAC, point-to-point
requires CEF, IP routing
client process is required to start session (e.g., IGP, BGP)
NX-OS: disable IP IDS, because src = dst
modes:
Async:
control only
continuous exchange
Async demand:
control only
on peer liveness check
Echo:
control + echo
looping packets back through peer
hardware
Echo packets have dst = src to verify TCP/IP stack
; BFD Echo enabled by default
(config-if)# no bfd echo
; not to overload CPU
(config-if)# no ip redirect
; ms: interval for generating Control and Echo packets
; rx_ms: min interval between ingress Control and Echo packets
; num: 3 default, number of sequentially lost packets to indicate that peer is down
(config-if)# bfd interval <ms> min-rx <rx_ms> [multiplier <N>]
; 1000ms default, interval for Control packets, Echo uses interface timers
(config)# bfd slow-timers <ms>