Bi-directional Forwarding Detection (bfd) minutes
Minutes taker: Tom Nadeau

Wednesday, August 3 9:00 AM

Dave: Introduction. Read Note Well.

Short agenda: Dave will talk about changes, applicability.
Presentation by Suping.

Status of the drafts. LC base specs. Technically, last call is finished.
Released an intermediate draft containing only editorial comments. No comments
(0) during LC. Assuming that most folks are happy with single-hop and MH specs.
Will send out call for implementation experience.

MH spec. minor tweak.

MIB: may need another spin. Will go to LC after quick review to determine if
any changes needed. Tom, is this okay? Yes.

MPLS spec. WG LC start next week.

Generic spec. version 00 is out. version -01 is imminent. Potentially to LC
after draft settles down or 3 week comment period.

Milestones: Status to be updated. Expect last meeting in Vancouver.

Discuss specification changes details: Base spec., SH, MH spec. Major change
is a port change used to discern MH or SH cases.

Comments on new version of MIB? Dave asks for comments. No comments.

New author set on MPLS draft. Basic changes to draft are to send original
port, but replies will use the MH port on replies. Dave asks room for
comments. No comments.

Describes diffs between generic draft 00 and 01.

Dave: Alex Z. is here. Can we get a MIB Doc review for MIB?

Tom: can we do the LC and MIB doctor review at the same time?

Alex: I will send an email to MIB doctor pool and get one assigned.

Dave: back to diffs...

Dave: I would like comments on multiple topologies or adj multiplexed over one
interface. Must have proper packet demux before session demux. Is this a good
change?

PackeS: Yes, this is a good change.

BFD should not be used to carry any other application-specific information.

Lowest level of control plane should be attached to the BFD session. Others
should be keyed to the events (control dependencies) from this FSM.

Last slide is on static routes. Must be configured on both end points to
work. Recommending BFD is bootstrapped the same way its done for the static
case. Transition to/from states processing should be the same as non-static
case.

BFD should be run inside tunnels for max to operate at most basic level.

Dave: should this draft be informational?

Alex: could be okay for standards track.

Rob (Laurel): you said in a hierarchy that BFD should interact with the lowest
level. The lowest level should bootstrap too. So from a bootstrapping AND a
listening situation BFD should interact with the lowest level? Clarify?

Dave: You can bootstrap the BFD session using the best method. The lowest
point of the control plane should interact with the BFD session.

Rob: for iBGP, ignore BFD. don't set up BFD sessions to loopbacks. Let
OSPF/ISIS do it. Its not aware that BFD is running.

Dave: There is nothing subtle there. You can do whatever you think makes
sense.

PackS: We dont have any documents yet that describe the static route case nor
do these show how to run this in iBGP.

Dave: I think we do. Lets think about this...we do have this in the documents.
You generally don't want to run this for iBGP peers; you run it on IGP. In MH
or eBGP case we do explain why this is needed. External BGP reacts to the
downstate of the directly connected link or if. We describe the l3 state and
tear down the session quickly. Maybe we need stronger text? With static
routes we don't do bootstrapping, so when its configured, you can have a BGP
configuration too. Local decision.

PackS: What I was looking for is guidance of what to do in these cases, to have
consistent implementations.

Dave: Whether or not you deactivate the route is a local decision, but will
most often result in not transmitting data.

PackS: I still don't get it.

Dave: the sentence I would like to see states that you can stop xmitting data.

PackS: Another eBGP question. Issues with graceful restart?

Dave: It could: When receiving a BFD failure notification, in eBGP you would
tear down the state and if the router was enabled for Graceful Restart, it
would do that. We are only detecting failure of data plane.

Tom: in MPLS draft I think that we recommended that you do the graceful restart
procedures if sessions are lost.

Dave: We can suggest some language in draft. In the eBGP case the assumptions
of graceful restart might want to be avoided.

Alia: I like the discussion of the lowest level of the control plane attachment
to BFD. What I want is the text to state that the BFD session down notification
should take the same action for the other applications.

Dave: We state that in the SH draft, but in the BFD Generic draft we try not to
use "BFD failure?" == "L2 down" Its whether or not the text would be
confusing.

Alia: It seems that if you don't have the text that way you could have some odd
corner cases. Something to give the philosophy/guidance.

Pecket: Just to clarify that my initial proposal for this text was to
explicitly take explicit action when the even comes in. I have no objections if
some objections would like to remove it from the routing when the if goes down;
might be an implementation case.

Dave: I think we are going to put down the routing case. I see your point.

Q: Where are the drafts?

Dave: -01 is coming out ASAP.

Dave: Is there any decent to make the gen draft a WG doc?

room: none.

Dave: anyone to present idea? I will present slides on BFD init with
BGP/static routes...

Dave: Has anyone read the draft.

Packet: Before we have this clarifying discussion. I am a bit sympathetic to
the first bullet. From the perspective of the operator.

Tom: Do not make a WG document; clarify the existing drafts.

Richard Spencer: Should we put all applications in this draft. We are asking
for a WG document. Are we going to end up with multiple drafts. If there is
something to be standardized to implement interop. we need to agree what. In
the static case, we need to configure the src/dst points manually. I think if
you want to have a sep. draft that is a WG draft,then we need to say what is
missing from current drafts. If there is anything missing, then we should
incorporate it into existing drafts.

Dave: I concur. I do not think we need to extend BGP, and to clarify what to do
in the base documents. So the BGP extensions are unnecessary to bring up a BFD
session, and so the extentions as prescribed are not applicable to get the BFD
session up. No issue to be solved here. WRT finishing generic applicability --
I strongly prefer that we do not have a draft for every application. Instead
have a generic application document that includes these.

Dave to the Room: are there any WG members that want to accept this as a WG
draft? None.

Tom: Ask coauthors to clarify text in generic draft.

Packet: Maybe some appendices to explicitly clarify popular cases.

Dave: I know we have used the "appendix out" in the past, but I think we will
just put it inline into the document.

Dave: Comments/questions? None. Attempt is to have a short LC after documents
have been updated.