BFD Scale:
| Capability |
CRS-1 LCs |
CRS-3 LCs |
| Max. number of sessions per LC (includes single and multipath) |
1250 |
1250 |
| Minimum timer interval for multipath sessions |
100ms |
50ms |
| Minimum timer interval for single path sessions |
33ms |
33ms |
| Minimum BFD multiplier |
3 |
3 |
| PPS rate per LC for multipath sessions |
4000 |
7000 |
|
PPS rate per LC for singlepath sessions
|
7000 |
7000 |
| Max. number of sessions per system |
2000 |
2000 |
Please note the minimum timer intervals are tested values. Although the configuration allows to configure a minimum interval of 15ms it is not recommended to deploy it because it may cause too many false positives.
BFD and QoS:
Locally originated BFD Control packets (in Asynchronous or Echo Mode) are generated by LC CPU and hence placed directly in the high-priority queue and transmitted with IPv4 Precedence set to 6 regardless if a QoS policy is present on the interface. Any QOS policy applied to the interface will be ignored with the packets being placed in the interface's default high priority queue. This behavior is different from other Locally Originated Control Packets, which are originated from RP and hence they are subject to the service policy configured on the egress interface.
From IOS XR 3.8 release onwards the ingress ASIC is able to identify transit BFD Echo packets and will mark them as 'vital' and place them into the high-priority queue without the need for a specific QOS policy / class-map statement in the ingress direction. The 'vital' bit setting ensures that under congestion, the packets are not dropped on either the ingress or egress Line Card.
If an ingress QOS policy is present on the interface on which BFD echo packets are received, then the BFD Echo packets are marked as vital packets and all QOS actions of the matching class except for tail drop and WRED are performed on the packets. The BFD Echo packets will still be placed in high priority queue, overriding the queue selected by the ingress QoS policy. In the egress direction, the BFD Echo packets are treated like other vital packets (LOCP) and are sent to the high priority queue of the interface.
- If no QoS policy is configured, the vital packet will take high priority queue
- If QoS policy is configured on the interface, and the policy has NO priority class, then vital packet will take high priority queue
- If QoS policy is configured on the interface, and the policy has priority class, then the vital packet will take the queue as specified in the class it matches.
BFD on bundles:
On CRS BFD over Bundle (BoB) and BFD over logical Bundle (BLB) is supported. BLB was introduced in IOS-XR release 4.2.3.
While BoB has a BFD session on each bundle member BLB treats a bundle interface with all its members as one pipe. BLB is a multipath (MP) singlehop session.
If BLB is running on a bundle there is only one BFD session running. This implies that only one bundle member is being monitored by BFD, at any given time. This creates following limitations.
- A failure of bundle members, which BFD is not running on is not detected.
- A failure of a bundle member, which BFD is running on will cause BFD to declare a session failure on the bundle, even if there are sufficient numbers of other bundle members available and functional.
However, BoB does not provide true L3 check and is not supported on subinterfaces.
But it is possible to run BoB and BLB in parallel on the same bundle interface. This provides the faster bundle convergence from BoB and the true L3 check from BLB.
A configuration command is available, which allows to select which coexistence mode is used for BoB and BLB coexistence.
bfd bundle coexistence bob-blb [inherited|logical]
The command has to be configured in global mode.
When the “inherited” coexistence mode is configured then a BLB will always create a virtual session and never a BFD session with real packets.
When the option "logical" is used BLB will always create a real session even when BoB is on. There is one exception if the main bundle interface has an IPv4 address. In this case the session is inherited when BoB is on.
Please note a MP session requires the the following configuration under bfd:
multipath include location R/S/CPU0
Please note: BFD multipath packets are not sent to the explicit HP queue. The vital bit is not set. In case of a multipath session it is recommended to configure a corresponding policy-map to avoid drops of BFD packets in the case of congestion.
