Silent L2 Failure: BGP Timers Versus BFD
AS 65000 uses a Nokia SR OS PE with eBGP to a provider edge over a Layer 2 metro Ethernet service. The BGP neighbor is 192.0.2.1, peer-as 64500, hold-time 180 seconds, keepalive 60 seconds. No BFD is configured. The provider's L2 service can fail in a way that does not bring down the local interface: the PE interface remains up/up, but frames to the provider are dropped. During a maintenance window this failure occurred at 10:00:00; the backup eBGP path via 198.51.100.1 was installed at 10:02:55. The SLA requires failover within 1 second. The provider supports BFD asynchronous mode with 300 ms minimum interval and multiplier 3. The PE has 4,000 BGP prefixes and enough CPU. Change control asks whether reducing BGP keepalive to 10 and hold-time to 30 is sufficient, or whether BFD is needed.
1. Calculate the current detection time and the detection time with keepalive 10 / hold-time 30. Is timer reduction sufficient for the SLA?
Read the worked answer
The current hold-time is 180 seconds. Keepalive 60 seconds does not detect failure; the session remains up until the hold-time expires, so detection is up to about 180 seconds, consistent with the observed 2m55s. Changing to keepalive 10 and hold-time 30 reduces worst-case detection to about 30 seconds, plus route installation, which is still far above 1 second. It also increases keepalive traffic and does not help when the L2 path silently drops frames. Timers alone are not sufficient.
2. How should BFD be configured and what detection time should be expected?
Read the worked answer
Enable BFD on the BGP neighbor on both ends and configure asynchronous BFD with 300 ms desired minimum TX/RX and multiplier 3. Detection is roughly 3 times 300 ms, or 900 ms, plus scheduling and jitter; it meets 1 second only if both ends honor the values and CPU is not overloaded. In SR OS, enable BFD under the BGP neighbor and configure the BFD session parameters under the router BFD context. Verify with show router bfd session and show router bgp neighbor 192.0.2.1. BFD then tears down BGP when the path fails, allowing the backup to install before the hold-time.
3. What if the provider supports only 1-second BFD with multiplier 3? What should the operator do?
Read the worked answer
That gives roughly 3 seconds of detection, which misses the 1-second SLA. The operator should negotiate 300 ms BFD with the provider, use a transport with faster failure signaling, or formally change the SLA. Setting BGP hold-time to 3 seconds is not a substitute: it still gives about 3 seconds of detection and adds instability. Validate the final design by dropping frames in the L2 service, not just by shutting the local interface.