Re: draft-mmm-bfd-on-lags: IPv4 _and_ IPv6 ?

[Marc Binderberger <[e-mail]>]

Hello Manav, Pablo et al.,

I think we will work on an update with a better wording. What we have now can be misunderstood. I agree with Manav's comment that what we really want to enforce is:

when you run a micro session for a particular address family on one member link then you must run micro sessions for this address family on all member links (of the LAG).


> Do you have some scenarios in mind?


I have a (big) customer in mind who was just asking for this :-)
The reason is that BFD micro sessions do cover layer-3 aspects as well and the customer thinks we should cover both, IPv4 and IPv6 code path. Sounds reasonable.


Regards, Marc



On 2012-08-16, at 17:12 , Bhatia, Manav (Manav) wrote:

> I believe there is nothing in the draft that precludes somebody from running separate IPv4 and IPv6 micro BFD sessions on the same member link. I could be missing something but i just cant think of why somebody would want to do that. Do you have some scenarios in mind?
>  
> Cheers, Manav
> 
> From: Pablo Frank [mailto:[e-mail]] 
> Sent: Thursday, August 16, 2012 8:02 PM
> To: Bhatia, Manav (Manav)
> Cc: Marc Binderberger; [e-mail]; [e-mail]
> Subject: Re: draft-mmm-bfd-on-lags: IPv4 _and_ IPv6 ?
> 
> In my original comments on this topic, I had suggested that one should be able to run separate IPv4 and IPv6 micro-BFD sessions on the same member link.  They would be separate sessions for all intents and purposes.  Because of the scale problem, I also suggested that you should run one of the address families at a more sedate rate but I was still thinking it would be a set of (separate) micro-BFD sessions.  
> 
> It seems inconsistent to me to suggest that one address family should use micro-BFD and the other address family should use regular BFD.  If that's acceptable, then why did we use IP encapsulation in the first place for micro-BFD?  
> 
> regards,
> Pablo
> 
> On Wed, Aug 15, 2012 at 11:17 PM, Bhatia, Manav (Manav) <[e-mail]> wrote:
> >
> > for draft-mmm-bfd-on-lags-05 I would like to bring a question
> > into the discussion that I have heard now several times
> > (thanks Pablo!) since the latest draft is out: why is the
> > draft so strict in enforcing the micro sessions to be either
> > IPv4 exclusive-or IPv6 ?  Why not allowing both?
> >
> >
> > 2.1.  Micro BFD session address family
> >
> >    Only one address family MUST be used for all micro BFD sessions
> >    running on all LAG member links.  I.e. all member link micro BFD
> >    sessions of a particular LAG MUST either use IPv4 or IPv6.
> 
> The above statement was written to preclude the possibility of someone using IPv4 for one member link and IPv6 for the other member link within the same LAG. Clearly, what we're saying here is that either use IPv4 or IPv6 for all micro sessions - don't do a mix and match.
> 
> The text as it appears seems to suggest that we're discouraging running both IPv4 and IPv6 for all micro sessions. The reason is scale, as already answered by Wim. If an IP interface has both IPv4 and IPv6, folks could use one address family (IPv4 or IPv6) for all micro BFD sessions and the other address family for a more sedate BFD session over that IP interface.
> 
> > As we want to cover the layer-3 aspects of the LAG and member
> > links as well, having a liveness test for both IPv4 and IPv6
> > seems reasonable. The draft states
> >
> >
> > 5.  Detecting a member link failure
> >
> >    When a micro BFD session goes down then this member link MUST be
> >    taken out of the LAG L2 load balance table.
> >
> >
> > One could rewrite this statement that when _any_ micro BFD
> > session of a member link goes down that the link MUST be
> > taken out of the load balance table. Plus the comment that in
> > case the LAG load balance table is address-family aware that
> > an implementation MAY remove the link from the load-balance
> > table only that matches the failing micro session's address
> > family (not sure if such an implementation of LAG load
> > balance tables exist at all? Otherwise we could skip this 2nd part).
> 
> I don't think there are such address family aware load balancers existing today. However, I don't see a harm in adding such a text.
> 
> Cheers, Manav
> 

--
Marc Binderberger           <[e-mail]>