IETF 119 LSR Minutes

Chairs:
Acee Lindem (acee.ietf@gmail.com)
Chris Hopps (chopps@chopps.org)
Yingzhen Qu (yingzhen.ietf@gmail.com)

WG Page: https://datatracker.ietf.org/group/lsr/about/
Materials: https://datatracker.ietf.org/meeting/119/session/lsr

Thursday Session II 13:00 - 15:00, March 21, 2024

13:00

Meeting Administrivia and WG Update

Chairs (10 mins)

Acee: No comments. Refer to slides.

13:10

Multi-part TLVs in IS-IS

https://datatracker.ietf.org/doc/draft-ietf-lsr-multi-tlv/
Les Ginsberg (10 mins)

• Les: Document is ready for Last Call
• Acee: Will consider prioritization but want to conclude IS-IS and
  OSPF SR YANG Models.
• Chris: We could possibly run them together since the others are
  YANG..

YANG Model for IS-IS PICS

https://datatracker.ietf.org/doc/draft-qgp-lsr-isis-pics-yang/
https://datatracker.ietf.org/doc/draft-qgp-lsr-isis-pics-srmpls-yang/
Yingzhen Qu (5 mins)

• Yingzhen: Request Adoption.
• Acee: Agree we should adopt draft and learn from experience.

13:25

Intra-domain SAVNET method

https://datatracker.ietf.org/doc/draft-lin-savnet-lsr-intra-domain-method/

Changwang Lin / Yuanxiang Qiu (10 mins)

• Acee: for intra-area why wouldn't you just advertise them all, SPFs
  are cheap.
• Les: I don't think we need any protocol extensions. If you really
  want to limit prefixes you can use tagging or local config
  filtering.

13:35 (13:37)

Signaling Aggregate Header Size Limit via IGP

https://datatracker.ietf.org/doc/draft-liu-lsr-aggregate-header-limit/
Yao Liu (10 mins)

• Tony Li: MSD (Maximum Segment Depth) is a bad name and is not
  extendable to IPv6 Header length. IGP is not a dump truck. This
  information should not be signaled here.
• Yao Liu: Intermediate nodes may need this information signaled in
  IGP.
• Ketan Talaulikar: Problem is more complicated. The problem statement
  needs to be presented in 6MAN. There is work to define extension
  header limits.
  https://datatracker.ietf.org/doc/draft-ietf-6man-eh-limits/
• Yao: Ok. Will discuss on list.
• Yingzhen: OSPF Generalized Transport and IS-IS GenApp is available
  for signaling non-routing information.

13:45 (13:49)

Advertisement of Remote Interface Identifiers for Layer 2 Bundle Members

https://datatracker.ietf.org/doc/draft-glctgp-lsr-l2-bundle-member-remote-id/

Liyan Gong (10 mins)

• Acee: Would be good to add use cases beyond controller L2-bundle
  discovery.
• Liyan: We will update the draft.

13:55 (13:59)

IS-IS and OSPF extensions for TVR (Time-Variant Routing)

https://datatracker.ietf.org/doc/draft-zw-lsr-tvr-extensions/
Sandy Zhang (10 mins)

• Tony Li:(As TVR Chair) This document is not a request of the TVR WG.
• Sandy: Just one proposal for information distribution.
• Yingzhen: Every router in the domain doesn't need to know every
  schedule. What does an overlapped schedule mean?
• Sandy: In case one node or link is used for different use cases, the
  schedules may be overlapped.
• Chris Hopps: (Speaking as Co-chair) The TVR WG doesn't sanction this
  work and so it sort of looks like an end run.
• Sandy: Extension for TVR to IGPs need to be done in LSR. This is
  another way to distribute schedule information.
• Chris: (as wg-member) I don't support distributing this type of
  information in the IGP when YANG can be used.
• Acee: We need to agree on putting this information into the IGPs
  before we jump to encodings and any draft should include what is
  done with the information (e.g., pre-computing SPTs dependent on
  imminent changes).

14:05 (14:12)

IGP Color-Aware Shortcut

https://datatracker.ietf.org/doc/draft-cheng-lsr-igp-shortcut-enhancement/

Mengxiao Chen / Changwang Lin (10 mins)

• Acee: Don't mix short-cuts with IGP CAR. We already admin tags for
  prefixes.
• Peter Psenak: We already have tags for prefixes and tunnel selection
  via tags. And the headend decision is a local decision and doesn't
  need to be specified.
• Gyan Mishra: We have SR policies on the headend to select SR paths
  based on color.
• Liyan Gong: Thanks for discussion on IGP tags. We will discuss on
  the list.We didn't use BGP SR policy because it is for a data center
  network with only OSPF.

14:15 (14:26)

Using Flex-Algo for Segment Routing (SR) based Network Resource Partition (NRP)

https://datatracker.ietf.org/doc/html/draft-zhu-lsr-isis-sr-vtn-flexalgo

Yongqing Zhu/Jie Dong (10 mins)

• Les: We've had constructive conversation on mailing list. No
  agreement and need to get some consensus on issues. Minor point -
  E-bit is completely unneeded.
• Jie: Appreciate discussion. We've had discussion on using Flex-algo.
  E-bit can be discussed.
• Ketan: Why are NRPs spread across multiple members?
• Jie: Do you mean use separate L3 links for each NRP?
• Ketan: Just don't use bundle.
• Acee: Why do we need different metric types per NRP?
• Les: Different metric types was an anxilary discussion.
• Gyan: Are you using bundles to avoid multiple L3 links?
• Jie: Yes - L3 interface is a superset of members and colors with
  color correlating
  member to NRP.
• Gyan: Compares use of SRv6 to accomplish the NRP separation.
• Yingzhen: Continue discussion on list.

(14:43)

IGP Flexible Algorithm with Link Loss

https://datatracker.ietf.org/doc/draft-wang-lsr-flex-algo-link-loss/
Jie Dong (10 mins)

• Acee: I beleive link-loss could be used as a flex-algorithm
  constraint. Need consistency across the IGP routing domain for
  link-loss measurement.

General Discusion:

• Gunter: Agree with Tony Li that IGPs are now dump trucks and we
  shouldn't put information in it that doesn't belong there. We need
  to be convinced that the IGP is the best place for added
  information.

==========================================================

Chat History:

Chat
Yingzhen Qu
00:00:17

Hello, welcome to LSR session

Andrew Stone
00:00:28

oh no this is the room with the annoying mic sound :(

Yingzhen Qu
00:00:29

please help with note taking:
https://notes.ietf.org/notes-ietf-119-lsr?both

John Scudder
00:01:13

Bad mic sound STILL? Dang.

Les Ginsberg
00:01:57

Thanx John.

John Scudder
00:03:07

I’ll pass through the complaint for whatever good it will do. Can you
characterize “annoying” for me?

John Scudder
00:03:24

(About the mic not the AD ;-)

Andrew Cooks
00:03:42

Any interest in resuming work on Multi-Topology routing for OSPFv3

Andrew Stone
00:04:14

digital sound effect like a whistle-like instrument trying to say "zoo".
Like a futurist "zoom" effect.

makes me wonder if it's some kind of sound detection check / loopback
that happens to be coming through

Lorenzo Miniero
00:05:00

No need to report it, we're already aware of the issue: the AV team is
already investigating it

Andrew Stone
00:05:09

thanks Lorenzo!

John Scudder
00:09:42

Note status slides are out of date w.r.t. dynamic-flooding draft. That
one is on the IESG agenda for 4/4

Yingzhen Qu
00:10:14

noted. thanks, John.

Tony Li
00:13:35

And I think I've addressed LC comments. New version was uploaded Sunday.

John Scudder
00:16:04

@Tony Li right, I saw that and it’s why it’s now on the IESG agenda.
Thanks for the updates.

Gyan Mishra
00:17:39

I see this OSPFv3 MT I-D from 2007.
https://datatracker.ietf.org/doc/html/draft-ietf-ospf-mt-ospfv3-03

Lorenzo Miniero
00:18:32

The mic is still turned off, apparently, at least listening to the
remote stream

Boris Khasanov
00:18:32

still very low sound from the presenter

Mengxiao Chen
00:18:48

very low +1

Lorenzo Miniero
00:18:49

is it played fine in the room there?

Gyan Mishra
00:18:51

I think most operators would opt for SR Flex Algo instead of MT

Tony Przygienda
00:18:52

mike impossible to hear

Gyan Mishra
00:19:06

Very low

Christian Hopps
00:19:09

it almost sounds like remote is getting the sound from another mic off
the in room speakers system

Andrew Cooks
00:19:15

@gyan, that's the most recent work I could find on the topic. It would
need to be reinterpreted in terms of RFC 8362

Lorenzo Miniero
00:19:50

Anyone in the room there that can relay the message?

John Scudder
00:20:16

Might as well say “don’t think of a pink elephant”!

John Scudder
00:20:48

@Lorenzo Miniero what message, sorry?

Lorenzo Miniero
00:21:00

That the speaker mic is turned off

Lorenzo Miniero
00:21:21

(unless it sounds fine there and we're not getting it in the remote
feed)

Gyan Mishra
00:21:23

Agreed based or LSA TLV based formatting

John Scudder
00:21:36

How about now?

Abdussalam Baryun
00:21:42

now better

Lorenzo Miniero
00:21:56

Thanks! In the meanwhile we've notified the AV team about the issue

John Scudder
00:22:07

Thanks!

Gunter Van de Velde
00:23:00

is online ok now

Christian Hopps
00:23:08

the gain is not the same

Abdussalam Baryun
00:23:21

now ok

Boris Khasanov
00:23:28

okay now

Tony Przygienda
00:23:43

I encourage operators to have a close look at this draft depending how
they write/check RFPs or operate in deployment and hence want to query
the model

Tony Przygienda
00:24:34

e.g. almost all RFPs are structured on RFC numbers and hence we are
still discussing how to query for "is this RFC supported" and "what
things of this RFC are not supported?"

Tony Li
00:25:17

Put the RFC number in the leaf description?

Tony Przygienda
00:25:33

yeah, but that's just an unstructured comment

Tony Przygienda
00:25:55

so a possibility is to have multiple ways to group the leaves, one of
the things we were looking at

Tony Li
00:26:04

You could add a numeric attribute to each item.

Les Ginsberg
00:26:56

I would not like a BIS to trigger unnecessary updates.

Tony Przygienda
00:27:27

@Tony: yeah, a possible idea

Tony Przygienda
00:27:54

again, input from operators' experience would be good, their thinking
about tooling they'd deploy

Tony Li
00:28:56

@Les: Features could update the YANG model, so no BIS would be
necessary.

Les Ginsberg
00:29:35

I meant a BIS for the RFC defining the feature which is being described
in the YANG model.

Christian Hopps
00:29:36

@Les Ginsberg hmm or the opposite if an original feature RFC get's bis'd
don't you maybe want to know which RFC is supported by the router?

Shraddha Hegde
00:29:54

What about LFA mechanisms and related calculations? Is that taken into
account?

Les Ginsberg
00:30:09

@Chris YOu want to know whether the new features defined in the BIS are
supported.

Les Ginsberg
00:30:47

The RFC # isn't the key - as others stated it is just a comment.

Christian Hopps
00:31:29

makes sense

Tony Przygienda
00:32:12

hmm, how will that work with ECMP on reverse path, both interfaces are
valid?

Tony Przygienda
00:32:39

@Shraddha: yeah, the LFA comment is +1

Tony Przygienda
00:32:57

pointless having 50ms failover to wait for SAVNET blackhole ;-)

Tony Przygienda
00:33:32

fix the mike pls. I ran out of volume on my hearing aid ;-)

Acee Lindem
00:33:59

It is very loud for me.

Yingzhen Qu
00:34:06

the current presenter is remote @Tony P

Tony Przygienda
00:34:32

hmm, ok.

Yingzhen Qu
00:39:19

@meetecho, please point the camera to the presenter

Les Ginsberg
00:45:34

@Tony: I think we have already used the MSD advertisements for things
that don;t match that name. Give us a bit of slack here as regards
naming...

Tony Li
00:46:14

I will give you slack now. I will repeat my complaint to those who
defined MSD incorrectly in the first place.

Christian Hopps
00:46:22

\:)

Les Ginsberg
00:46:28

That's fair.

Tony Li
00:47:32

And for this case, it would be nice not to compound the error.

Tony Przygienda
00:47:42

calling things names that don't mean what things are and calling
different things same names are strong indications of confusion ;-)

Srihari Sangli
00:48:11

\:)

Les Ginsberg
00:48:15

It is conceptualy similar - hate to waste a codepoint.

Tony Li
00:48:17

And causes confusion for other implementors and operators. Our job is to
NOT do that.

Tony Przygienda
00:48:30

and Tony was more successful in his valiant defence of IGP not becoming
a dump truck while the packets by now are a full dumpsite it seems ;-)

Christian Hopps
00:48:41

what's a hippity anyway?

Ketan Talaulikar
00:49:00

MSD value is 1 byte, which means we cannot go beyond 256 bytes for IPv6
EH ... so that is another aspect to consider ...

Tony Przygienda
00:49:31

this could be a feature actually ;-) surely, as almost always in SR
case, unintended but nevertheless ;-)

Tony Li
00:50:12

MSD is maximum SID depth. So 255 indicates 255 SIDs. How this
corresponds to an IPv6 header is still unclear.

Tony Li
00:51:04

You COULD treat it as the number of /128's in the extension header, but
as Ketan pointed out, there can be other extensions.

Ketan Talaulikar
00:52:51

https://datatracker.ietf.org/doc/draft-ietf-6man-eh-limits/ ... just to
give an idea of where this work might be going

Boris Khasanov
00:53:37

@Ketan, thanks for pointing this work

Yingzhen Qu
00:59:12

please move camera to the presenter

Liu Yao
01:02:15

the summary of previous descussions about AHL in LSR and 6man can be
found at
https://mailarchive.ietf.org/arch/msg/lsr/iEMsMFntQceI3y67dOd2fdcq0g4/

Liu Yao
01:11:20

@Ketan Talaulikar yes, there're diffrent EH limits,AHL is one of
them,and not all of them need to be signaled. Diffrent nodes may process
different EHs, HBH/DOH/SRH,but for the intermediate nodes between
intermediate endpoint nodes，they wouldn‘t process SRH, just IPv6 header
and at most HBH,so the total size of the header in the packet they need
to process is not that big. But for intermediate endpoints，maybe it's
HBH+DOH+SRH with many TLVs。

Ketan Talaulikar
01:12:56

@ Liu Yao, what I am looking for is what all info is needed and who is
the consumer that is going to use them. As an example, I mentioned, ALH
has no relevance on an transit IPv6 router - it only needs to process
HBH.

Ketan Talaulikar
01:13:57

The point is that we need all sorts of limits depending on who is
parsing what.

Ketan Talaulikar
01:14:49

My suggestion, please put together the whole story - then the WG can
better understand what this is where it is going. Then comes the point
if IGP is indeed the right place to put this.

Shraddha Hegde
01:19:19

IP flex-algo can be used as another alternative solution instead of new
color extensions

Mengxiao Chen
01:22:46

@Shraddha, do you mean to use IP flex-algo for TE-tunnel shortcut?

Shraddha Hegde
01:25:19

@Mengxiao the IP flex-algo has ways to advertise prefixes with
associated IP flex-algo which can match to the color of the tunnel to
shortcut. As long as the number of colors are limited it's a possible
solution

Liu Yao
01:27:19

@Ketan Talaulikar I can provide some analysis in SRv6 network (who will
process what, and what are the limits on them)based on
https://datatracker.ietf.org/doc/draft-ietf-6man-eh-limits/ and take it
to the 6MAN as well，I think some packet dropping issues are not easy to
solve, but we'll try and see.

Mengxiao Chen
01:28:15

@Peter Psenak, could you please help to provide the document you
mentioned about coloring IGP prefix?

Peter Psenak
01:28:44

Standard IGP tags

Acee Lindem
01:29:55

For OSPF:
https://datatracker.ietf.org/doc/draft-ietf-lsr-ospf-admin-tags/

Changwang Lin
01:30:11

It is possible to use tags to map colors, but tags are typically used
for management and may be utilized for policy filtering. When multiple
tags are present, it can be challenging to determine which tag is being
used to represent a color.

Mengxiao Chen
01:30:37

@Shraddha, thanks a lot. I will check the IP flex-algo document.

Christian Hopps
01:31:23

I don't understand why it's difficult

Changwang Lin
01:31:45

@shraddha, the flex-algo is fundamentally for BE forwarding and cannot
replace TE. Therefore, flex-algo and IGP shortcut are two different
technologies that are not inherently related.

Changwang Lin
01:32:39

@Christian Hopps:Multiple tags could be attached to one prefix. There
must be a mechanism for the
receiver to know which tag is used as color.

Christian Hopps
01:33:07

yes, but that doesn't seem challenging to me

Peter Psenak
01:33:37

it is not

Shraddha Hegde
01:33:57

@Changwang SPF is BE forwarding and it stitches onto shortcut tunnels so
I dont understand your concern

Randy Bush
01:34:55

when, or why, is a bundle not a bundle?

Christian Hopps
01:34:57

i mean typically you are doing things based on a tag being present. you
don't really have to do anything more if that's the case.

Mengxiao Chen
01:36:53

do we have such mechanism now? sorry, I haven't followed the discussion
on tags.

Christian Hopps
01:37:21

there's no mechnism if it's just "do this if tag A is present" "do that
if tag B is present"

Changwang Lin
01:39:18

＠Shraddha Hegde: Each prefix corresponds to a flex-algo, and the essence
of flex-algo computation is the BE forwarding calculated by all devices
based on the same constraints. Meanwhile, the essence of IGP shortcut is
TE forwarding.

Srihari Sangli
01:39:22

@Ketan, bundles are extensively deployed today, so opening up the bundle
members as L3 interfaces is not a good option. I know you are not
recommending it.

Ketan Talaulikar
01:40:08

@Srihari, no, that is definitely not a general recommendation :-)

Mengxiao Chen
01:41:51

it seems tags can be used at will, and the receiver needs to be decide
how to handle them through local configuration. is my understanding
correct?

Peter Psenak
01:43:35

you can split the tag space and use some of it to signal your color

Changwang Lin
01:43:46

If we use tags, multiple tags are required to accommodate management
needs, and consistent constraints and definitions are necessary for
interoperability between different devices.

Randy Bush
01:44:39

crickets

Andrew Cooks
01:44:46

reset here too

Ketan Talaulikar
01:44:56

meetecho reset

Lorenzo Miniero
01:45:35

Looking into it

Tony Przygienda
01:45:55

hutis_ cut more cables ;-)

Yingzhen Qu
01:46:03

we're waiting, please let us know when it's back

Acee Lindem
01:46:34

The remote people are able to communicate.

Tony Przygienda
01:46:39

well, it's only a draft about a new fad ;-) so cannot be that
important ;-)

Christian Hopps
01:46:40

move to the slide which shows the one sentence change :)

Les Ginsberg
01:46:41

You are back

Gyan Mishra
01:47:03

@Jie I understand the idea and it makes sense to be able to color
different member links with different SLA queuing characteristics per SR
Algo sub topology makes sense. I can see you want to have a single high
capacity bundle with many member links but want to carry multiple Algo /
NRPs on a bundle. Also you may want 1-1 or 1-many member link to
NRP/Algo color mapping. Thank you