BFD for AI-ML Data Centers
Hardware Assisted Inline BFD Overview
The Bidirectional Forwarding Detection (BFD) protocol is a simple hello mechanism that detects failures in a network. BFD enables sub-second detection and convergence, preserving the continuity of AI-ML workloads. Hardware-assisted inline BFD sessions run on the ASIC firmware. The Routing Engine creates BFD sessions and passes them to the ASIC firmware for processing. The device uses existing paths to forward any BFD events that need to be processed by protocol processes.
With hardware-assisted inline BFD, the firmware handles most of the BFD protocol processing. The ASIC firmware processes the packets more quickly than the software, so hardware-assisted inline BFD is faster than regular inline BFD. We support this feature for single-hop and multihop IPv4 and IPv6 BFD sessions.
We support hardware-assisted inline BFD sessions for both underlay and overlay. For example, you can run BFD sessions between EVPN overlay BGP peers.
To enable hardware-assisted inline BFD, use the set routing-options ppm
inline-processing-enable CLI command.
Use Feature Explorer to confirm platform and release support for specific features.
Benefits of Hardware-Assisted Inline BFD
-
BFD enables faster detection and recovery from link failures.
-
Hardware-assisted inline BFD reduces latency and decreases the load on the Routing Engine.
-
BFD timers are adaptive. You can adjust them to be more or less aggressive.
How to Adjust BFD Timers
The BFD failure detection timers are adaptive and can be adjusted to be faster or slower. The lower the BFD failure detection timer value, the faster the failure detection and vice versa.
-
To specify the BFD timers, include the
detection-timestatement:bfd-liveness-detection { detection-time { threshold milliseconds; } }Specify the threshold value. This is the maximum time interval for detecting a BFD neighbor. If the transmit interval is greater than this value, the device triggers a trap. The range is 1 through 255,000 milliseconds.
-
A
holddown-intervalvalue sets the minimum time that the BFD session must remain up before sending a state change notification.To specify the hold-down interval, include the
holddown-intervalstatement:bfd-liveness-detection { holddown-interval milliseconds; }You can configure a number in the range from 0 through 255,000 milliseconds, and the default is 0. If the BFD session goes down and then comes back up during the hold-down interval, the timer is restarted.
-
The
minimum-intervalvalue indicates the time interval for transmitting and receiving data.This value represents the minimum interval at which the local routing device transmits BFD packets, as well as the minimum interval in which the routing device expects to receive a reply from a neighbor with which it has established a BFD session.
To specify the minimum transmit and receive intervals for failure detection, include the
minimum-intervalstatement:bfd-liveness-detection { minimum-interval milliseconds; }You can configure a number in the range from 1 through 255,000 milliseconds.
Note:BFD is an intensive protocol that consumes system resources. Specifying a BFD minimum interval below 1 second on devices that do not support hardware assisted inline BFD can cause BFD flapping.
-
The
minimum-receive-intervalstatement specifies the minimum interval in which the local routing device expects to receive a reply from a neighbor with which it has established a BFD session:bfd-liveness-detection { minimum-receive-interval milliseconds; }You can configure a number in the range from 1 through 255,000 milliseconds.