Minutes for BFD at IETF-96
minutes-96-bfd-1

: Bidirectional Forwarding Detection (bfd)
: IETF 96 - Berlin
: 1 hour session, July 22, 2016; 12:20-13:20
:
Meeting starting.
Note well applies.

: Chairs: WG Status. 10 min

Need more reviewers.

:   S-BFD publication.
:   BFD Multipoint - WGLC? (draft-ietf-bfd-multipoint,
draft-ietf-bfd-multipoint-active-tail) :   Fate of MPLS MIB

Have spoken with Alvaro, MIB work might be removed from the charter.

: Fate of extended authentication mechanisms.

Work on security might be postponed for a while.

:   BFD work in other groups.

Existing work in other WGs is focusing on applicability of BFD, not on changes
to BFD itself.

Reshad: Liaised with LIME WG on relationship to BFD work.

:
: Mahesh Jethanandani:
:   Status on BFD Yang Module (draft-ietf-bfd-yang). 10 min.

Reshad presenting.
[presentation]
Changes to routing model had impact to changes to the BFD model. Model
compilation checked with ConfD, it has ability to verify XPath correctness.

Reshad: Do we need to cover MPLS-TP immediately?
Greg Mirsky: We are not actively working on MPLS-TP model, but that is the idea.
Reshad: Do we need to cover RPCs too? That is an open item. This is also
broader than just BFD WG. Greg: The other use for RPC would be BFD echo.
Reshad: What do you mean? Greg: To trigger BFD echo. It is effectively a
self-addressed ping, and you may want to do that for a specific duration.
Reshad: This seems to be implementation specific. Jeff Haas: Exposing RPCs for
LIME may be useful, depending on what they end up doing. Reshad: There are no
changes to the data model. We need to take that to the list. Reshad: VRF
awareness for BFD is implementation specific. This needs to be aligned with VRF
representation in netmod models. This will be for further discussion. Reshad:
LIME WG - there may be mismatch of the views on how OAM protocols are used. We
can have that discussion after the LIME connectionless model is presented.
Greg: As you are talking about LIME model relationship - the same exists with
MPLS-TP model. Reshad: There are a few issues that ConfD is complaining about,
while pyang is happy.

[discussion]

Jeff Haas: The issues related to schema mount are not specific to BFD WG.

: Optimizing BFD Authentication (To be discussed by saag?)
(draft-ietf-bfd-optimizing-authentication). 10 min.

Jeff Haas presenting on behalf of Mahesh who couldn't be here.
[presentation]

Authentication is expensive from platform resource usage perspective.
No substantive changes since last presentation.
Security ADs suggested to talk to saag people.

Greg Mirsky: I have sent this question to the list before - if we are sending a
single authenticated packet among 3 non-authenticated packets, in HW assisted
systems authenticated packet will be punted to control plane. The
non-authenticated packets will continue to be sent and will continue to reset
the dead timer. Jeff Haas: The exact details of failure of authentication needs
to be detailed. Greg: Indeed, the details of session state machine needs to be
covered for this case.

[discussion]

Greg Mirsky: This is an interesting work, the document is reasonable for
adoption, but before LC we need to solve the issue with the state machine. Jeff
Haas: We will take adoption to the list. Who thinks this is the work for
adoption? A few hands. Jeff Haas: Who thinks this work should not be adopted?
No hands.

:   BFD Stability - review requested? (draft-ashesh-bfd-stability). 2 min.

Jeff Haas presenting as virtual Mahesh.
[presentation]

Jeff Haas: This is a standalone work while it also fits into security work
bucket.

[discussion]

Jeff Haas: Who thinks we should adopt it? A few.
Loa Andersson: What was the question?
Jeff Haas: Adoption show of hands. (Loa adds his support.)
Jeff Haas: who is against? No one.

:
: Greg Mirsky:
:   Multi-chassis LAG (draft-tanmir-rtgwg-bfd-mc-lag-ip,
draft-tanmir-rtgwg-bfd-mc-lag-mpls)

Greg Mirsky presenting.
[presentation]

[discussion]
Reshad: What do you mean by the reverse direction?
Greg: From the redundancy group.
Reshad: Can you choose the local address on B and C?
Greg: It is a hint to the implementation that when you configure it on RG, you
need to associate it with a specific LAG group. Reshad: Will it use the local
source address? Greg: Yes. Jeff Haas: Are you looking for this work to be
adopted by BFG WG? Greg: It seems to fit better in the BFD WG. Jeff Haas:
Question to MPLS WG chair - is it ok to adopt it here even if it has an MPLS
component? Loa Andersson as MPLS WG chair: I would be comfortable with document
being adopted in BFD, just want to be us copied in adoption and WGLC calls.
John Messenger: Is this compatible with DRNI in IEEE802? Greg: Content ....
John Messenger: LAG standard has been updated recently to cover Distributed
Resilient Networking Interface. This standardizes the dual homed case. Greg: I
haven't looked at the latest version and how it is applicable. From BFD
perspective this is an implementation detail. We are injecting packets on a raw
adjacency and we are bypassing the aggregation. John Messenger: Would suggest
work done in 802.1 to analyze those considerations with 802, it would be
necessary to coordinate to work with them. Send a mail to IETF-IEEE802
coordination group. Jeff Haas: Can we sign you as a reviewer to this draft?
John Messenger: I could, although I might be good at finding spelling errors.
Jeff Haas: Show of hands for adopting this work. Few hands. Jeff Haas: Who
would be against? No hands.

:   Fault Management for EVPN networks (draft-gmsm-bess-evpn-bfd-00). 10 min.

Greg presenting.
[presentation]

Jabber comments from Prasad: I personally think there are merits in the first
method of encapsulation (i.e. use L2 header inside GAL/G-Ach payload) (next
slide) Also one more request - comparing the consistency of the BFD-EVPN with
BFD-VxLAN may be helpful

BFD for EVPN and BFD for VXLAN consistency.
Greg: BFD VXLAN proposal was to use VNI 0. Before Berlin we entertained the
idea of using ACH for BFD for VXLAN.

Reshad: The draft does not explain a scenario when there is no IP header. Also
when LSP ping is necessary. Greg: We will clarify and respond after the
meeting. As ACh is used, there is no source address in BFD encapsulation. There
was a proposal in BESS to use discriminator distribution via BGP. There might
be conflicts if you started BFD with a LSP ping and then send out with a
different discriminator - that may result in race conditions. Reshad: The
comment was as BGP is used already, it could be a logical place to extend it.
Greg: This needs to be correlated to the general practice of extending BFD.
Reshad: That would make it somewhat more similar to S-BFD. Greg: In general
yes. We need to look at the point to point case. Jeff Haas: Where should this
document live - that is an interesting question. Greg: My personal view is
somewhere between BESS and MPLS. Jeff Haas: Encapsulation issues move it closer
to BFD, from MPLS perspective there are no changes. Jeff Haas: The conversation
should cover chairs of BESS, BFD, and MPLS.

Prasad: I have no preference on which WG should work on it.
[discussion]

-----

Connectionless OAM model from LIME. Presented by Zitao (Michael) Wang

Michael presenting.

[presentation]

Same presentation as given in LIME WG session.

[discussion]

Jeff Haas: What assistance are you looking from BFD WG? BFD is one of the
components to be used in the model. Michael: BFD could help with verifying the
reality of BFD use cases and technology specifics.

Reshad: We discussed in LIME meeting yesterday that there are certain construct
like MPLS-TP identifiers and others, those are quite complex, and not certain
whether we can solve it with schema mount. Michael: You could use leafrefs to
address the hierarchy. Jeff Haas: Would you expect to use S-BFD, or regular BFD
as well? Michael: Our goal would be for common generic structure. Jeff Haas:
For connectionless mode where it is on-demand rather than pre-provisioned.
Michael: Both proactive and on-demand can be used. Jeff Haas: Will you be
making a suggestion to BFD YANG DT what RPCs to add? Michael: Yes. Jeff Haas:
Do you think discriminators should be discovered via IGP, or should it be done
via topology model? Michael: Topology model can show the test point and their
relationships. LIME wants just to provide the TPs and their relationships, then
some model instance could use an instance of this model. Michael: Use one model
to point into another.

Jeff Haas: Any other questions? None.

The meeting is concluded.