Re: a draft about how BFD notifying the state change to applications

[Jeffrey Haas <[e-mail]>]

On Wed, Aug 24, 2005 at 09:57:24PM +0800, Yang Pingan wrote:
> And another purpose of WTR  is that: before BFD notifying forwarding engine,

I think this is one of the key phrases.

BFD need not "notify the forwarding engine".  One's system can be designed
to notify whatever component is appropriate.

The raw "session transitioned up", "session transitioned down" events
can be passed to whatever component you like.  If you wish to damp
behavior of those applications, you could apply your mechanism
and provide an intermediate layer between BFD and your application.

I wonder, based on your example, whether or not a different use of
BFD may be more appropriate.  If you're using BFD in conjuction
with some sort of failover/high availability mechanism, you have
two pieces of information that you are concerned about:

 - If you are verifying connectivity to a given instance of some
   application.
 - You are verifying that forwarding works to this instance.

To give a more concrete example, if you are using BFD for static
routes where the nexthop is a nexthop managed by VRRP, there may
be a forwarding failure to the nexthop.  At some point determined
by VRRP configuration and state, another VRRP host will take over.

If, in this example, you wished to monitor forwarding, BFD to the
VRRP nexthop may be appropriate.  However, the timers probably should
be such that the BFD timeout is higher than the VRRP failover time.

If you wish to monitor a given VRRP host, your BFD session should
be to an address other than the VRRP virtual address.

My apologies if this example misses your point.


-- 
Jeff Haas 
NextHop Technologies