[c-nsp] Sub-second Fast Reconvergence for OSPF?

[Asbjorn Hojmark - Lists]

>>> Also, if HSRP and GLBP could be 'BFD-enabled' ('BFD
>>> clients'?), that would be extremely useful. Static routes
>>> would also be neat.

>> I second the motion for BFD enabled HSRP.

> Explain exactly what you are thinking here and I will see 
> about submitting a feature request.

I have a couple of boxes doing GLBP for 150+ VLANs. To get fast
GLBP failover, I have them configured with 'timers msec 200 msec
750'. While that does work, it also puts bit of a tax on the CPU
to do 750 hellos per second. So, if BFD could be used as a more
scalable way of detecting GLBP neighbor loss, that would be
great.

Also, as you know GLBP hellos are sent as IPmc, which means that
with fast hellos every client in the VLAN will see 10 hellos
every second. If BFD could be used for detecting GLBP neighbor
loss, the clients would be relieved of that traffic (since BFD is
unicast).


>> I also wonder if there could also be a feature to augment
>> tracking to be based on BFD neighbor as well: 'standby track
>> bfd neighbor x.x.x.x'

> Now this makes sense.  If an upstream adjacency for a tracked 
> interface is deemed unreachable, then I would want the router
> to lower its priority so that it stops attracting traffic and
> lets the standby handle it.

IMO to make a lot of sense, it would have to be object tracking,
so 'glbp X weight track Y decrement Z', where Y is a BFD object
(or 'standby X track Y Z', in the case of HSRP).

-A