Re: other comments for single-hop bfd [Re: I-D ACTION:draft-ietf-bfd-v4v6-1hop-01.txt]

[Dave Katz <[e-mail]>]

On Feb 24, 2005, at 12:06 AM, Pekka Savola wrote:

> On Wed, 23 Feb 2005, Dave Katz wrote:
>> It does raise the question of whether we want to continue to add text 
>> into this document for other protocols, or whether it would be worth 
>> having additional documents to cover them so that this one can be put 
>> to bed.  I would prefer to avoid further significant technical 
>> changes to this doc and to have additional stuff be put into a 
>> separate one, in order to advance this stuff without too much more 
>> tinkering.  This should probably be discussed in Minneapolis.
>
> Right.  Obviously, I'm a proponent of putting it here (well, I asked 
> if I should make it a separate draft back in July or so, but then it 
> was felt ok to include it here).

I guess this was overtaken by my slothfulness.  The question at hand 
now should be whether this document should be stalled for however long 
is necessary to incorporate it.  Assuming that it would require another 
IETF meeting to approve moving the document along the standards track, 
then it would seem to me that we should separate it out (assuming that 
this version is otherwise acceptable, modulo editorial issues.)  
Perhaps the WG chairs can step in and outline the pros and cons here.  
My apologies for letting this stuff slip by.

>
>>> .. in fact it would preclude the operation of the IGP in the 
>>> particular BFD-enabled _interface_.  The text seems to be implying 
>>> that having BFD would be an IGP domain-wide toggle like IS-IS 
>>> authentication.  But in reality, it is an interface toggle, so this 
>>> argument is irrelevant.
>>
>> It is not irrelevant on multipoint media.  Perhaps it would more 
>> correctly read "...preclude the *correct* operation of the IGP..."  
>> It's not that it makes the IGP completely fail, but it can create 
>> large-scale black holes if the topology is particularly unlucky.  I 
>> could add the extra word if you feel strongly about it.
>
> No, this clarification would also be misleading.  It would prevent the 
> correct operation of the IGP *on that particular interface*.  So, you 
> must enable BFD on all the systems you have adjacencies to on a 
> multipoint link.

Unfortunately, incorrect operation of the IGP on a single interface has 
network-wide implications (you can partition the network, for example.) 
  Network-wide correctness is predicated on node-by-node (and 
interface-by-interface) correctness.

>
> I'm not sure how common those multipoint links are (we certainly have 
> only one with more than 1 neighbor with which we would like to run 
> BFD; the most common are probably Ethernet IX's but folks don't run 
> IGP there), so unless our setup differs radically from what's out 
> there, this shouldn't be a problem.

It seems likely that someone will have a setup that differs from yours, 
either now or sometime in the future.

--Dave