[c-nsp] BFD feedback?

[Sam Stickland]

Chris Woodfield wrote:
> BFD is a lifesaver where you have circuits such as metro ethernet  
> links that don't lose link state when something in the middle blocks  
> connectivity. It's less useful across WAN links that depend on end-to- 
> end connectivity to maintain line protocol.
>
> As Arie, said, the hardest part of implementation is the timer  
> tuning, to get the best balance of response time vs. CPU utilization  
> as well as preventing false timeouts, another side effect of setting  
> the timers too low.
>
> That said, I do feel that tying BFD to routing protocol events only  
> is a bit shortsighted - why not have an option to just change line  
> protocol to down in a case of BFD timeout failure, and let the  
> routing protocols react the that naturally?
>   
Surely this wouldn't work on multi-access networks (e.g. ethernet ;) )? 
BFD forms adjacencies according to the routing protocol neighbors, not 
according to the physcial links. Consider the followinng topology:

R1 ----- SW1 ---- R2
                 |
                R3

R1, R2 and R3 share a broadcast domain via SW1 and each one has formed 
an OSPF adjacency with each other (full mesh). This means that there is 
a full mesh of BFD neighbors also. If the link from SW1 to R3 goes down 
BFD will detect this and take down the appropiate OSPF neighbors. If BFD 
shut the interface down instead it would also sever the communication 
between R1 and R2, which isn't want you want.

Sam