Minutes IETF103: lsr
minutes-103-lsr-00

LSR IETF103 Meeting Minutes

Chairs: Acee Lindem
        Chris Hopps
Secretary: Yingzhen Qu

WG Status Web Page: http://tools.ietf.org/wg/lsr/

1) Administrivia - 5 minutes
- Blue sheets
- Scribe/jabber
- Jabber room: lsr@jabber.ietf.org

2) WG Status Update - 10 minutes - Acee/Chris

WG status is now available online.

Chris H:  for TE attributes draft, it has been submitted. what's been hold up?
Acee:     the SR draft, 8 authors.
Les:      have you done the IESG submission? We'll take care of it.

3) IS-IS TE App - 5 minutes - Les Ginsberg
https://datatracker.ietf.org/doc/draft-ietf-isis-te-app/

Acee:    anybody objects? No implementation yet. But people are using the
         encodings. We’ll take the last call to the WG list.

4) IS-IS RFC 5306BIS - 5 minutes - Les Ginsberg
- https://datatracker.ietf.org/doc/draft-ietf-lsr-isis-rfc5306bis

Les:    this is equivalent to OSPF.
Chris:  any objection? Nobody objects.


5) IS-IS Spine Leaf - 10 minutes - Les or Naiming Shen
- https://datatracker.ietf.org/doc/draft-shen-isis-spine-leaf-ext

Les:     ask for WG adoption.
Chris H: is the dynamic flooding compatible with this?
Les:     the leaf node are not part of the flooding. You use dynamic flooding
         for up layers.
Chris:   it’s like of overloaded LSP. when you said the horizontal links, I
         don’t remember that being talked.
Les:     there was a B bit. The leaf can be a backup node to the spine. The
         feedback we got considered this unnecessary. The leaf node doesn’t
         know full topology, it gets information from spine. It just needs
         to know which spine to use as default route.
Chris:   I didn’t see anything talking about horizontal links. what happens if
         there are?
Naiming: We had a B bit if we can go through leaf to spine. Now we don’t flood
         our neighbors LSP back over there. We will need to put something like
         high metric there.
Les:     this is not really useful.
Acee:    I think this has little overlap with other proposals. It does a small
         subset of RIFT. We don’t need to worry about overlap for adoption call.
Chris:   other than the one I mentioned. I think the work is done other than
         the point I mentioned. I think we’ll be collapsing all flood reductions.
Les:     this is a complimentary proposal.
Chris:   this is only for bottom layer.
Les:     you could use this for up layers, but that's not our primary target.
Tony
Przygienda: we should push 7356, then it has more potential.
Acee:    7356 is for Link-scope LSP.
Les:     right now the draft allow you put this into hellos or link scope LSPs.
         the link scope lsps are better solution. I’ll consult with my
         coauthors. I have no objection.
Les:     the answer before was wait. Can we ask for WG adoption now?
Chris:   yes, we can do that. shall we do it at interim?
Acee:    I disagree at this point. this draft doesn't have overlap, we don't
         need to wait for interim. we have an implementation going. I don't
         want to delay it.
Chris:   I don't object as a chair. does anyone object?
Tony P:  if I were to run this group, I’ll try to get to 7356.
Les:     I support your suggestion. We have two mechanisms, you're suggesting
         to use one of them.
Tony P:  this optimization is optional.
Les:     I don’t think any changes required to this draft.
Tony P:  they have to proper deal with it. I'm not objecting, I'm suggesting.
Acee:    why don’t you put it down during adoption?
Naiming: I agree with Les and Tony. We may move to link-scope LSP. In this
         case, the special case is to not flood back. You may only need 100
         lines to support this. Really simple.
Chris:   we’ll take it to the list.

6) Update on dynamic flooding - 10 Minutes - Sarah Chen
- https://datatracker.ietf.org/doc/draft-li-lsr-dynamic-flooding/

Tony Li: so if there is no discussion, We’d like to ask for WG adoption.
Chris H: we have to resolve one issue. I think we’re ready. We have another
         draft out there.
Huaimo:  we have two different solutions for flood reduction. Last IETF some
         people proposed to merge the drafts. I’m wondering whether we just
         want to progress only one. we need to have more discussions.
Chris:   I look at the timeline. Tony published his draft in Jan.
         the distributed one was in March. Both were presented at IETF 101.
         There were some problems with the distributed one, it wasn't fully
         baked. During transitions to next IETF, Tony brought in the idea of
         distribution. You added the centralized solution. I think one way
         going forward to have you work on distributed algorithm, so that
         won't conflict. this is my opinion as a member of the WG. Tony’s
         draft is well written. Your proposal is very complicated. That’s
         my opinion as WG member. My suggestion is to have you work on
         distributed solution.
Huaimo:  Jeff had a requirement draft in DC. Then people start to work on flood
         reduction. After both drafts presented, some people thought disctributed
         is more practical, we should have more discussions. Maybe we should
         combine together or separate them.
Acee:    speaking as a chair, I like the way Chris positioned the draft. I
         support progress both distributed and central controlled.
         In Manet, 3 different solutions, they all came in as experimental.
         we don't want to lose this work.
Huaimo:  both drafts resolve the same issue.
Chris
 Martin: in terms of the splitting, that's the piece that makes the most
         sense.

7) IS-IS Sparse Link-State Flooding - 10 Minutes - Gunter Van de Velde
-  https://datatracker.ietf.org/doc/draft-hsmit-lsr-isis-dnfm/

Acee:    is that the only thing reduced? parallel links?
Gunter:  I will explain later. I even have pics.
Chris:   doesn’t this end up computing spinning tree?
Gunter:  it’s like a sparse tree.
Les:     in the spirit of Chris proposal, do you see we can take your idea
         into other draft? Do you see it reasonable?
Gunter:  we actually envision is as simple implementation. how to move it
         forward, we’ll discuss offline.
Tony Li: my concern about the diameter, also the degree. It’s key because it
         controls the speed of flooding. If the degree of the nodes is too
         high, and multiple anchors, that could cause instability.
Gunter:  we have a solution for that, it will be in the next version.
Chris
Bowers:  I support this. It doesn’t seem to fit in Tony’s draft. It's not just
         distributed, it has signaling, not just a subset of it.
Chris
Martin: back to to Tony’s point. The nature of flooding tree doesn't
         seem to be have resilience. I’m not sure this is the best approach.

8) OSPF Flooding Reduction - 10 Minutes - Huaimo Chen
- https://datatracker.ietf.org/doc/draft-cc-ospf-flooding-reduction/

Acee: no time for comments. We will discuss offline.

9) Distributed Algorithm for Constrained IGP Flooding - 15 Minutes - Dave Allen
- https://datatracker.ietf.org/doc/draft-allan-lsr-flooding-algorithm-00/

Tony P:  I looked at the details. Will that produce the lowest diameter tree?
         I saw this tie breaking, but didn't understand it.
Dave:    the diameter is corresponding to physical.
Tony P:  how to choose root?
Dave:    we looked it while looked multicast a while ago. I didn't consider it
         in this case.
Tony P:  you don’t have degree control. you probably need to address it.
Acee:    we’re not going to standardize all these. Implementations will be
         preferred.
Chris:   I think not just implementation, but differentiation. They're not
         exactly the same but similar.
Alvaro:  what’s gonna happen next? Straw men? Interim? On the list?
Chris:   we have to have a discussion first, then interim based on the results.
Acee:    we decided to put spine-leaf aside, since it’s incremental. we're
         gonna put in the straw men for the centralized and distributed algorithm
         framework.
Chris:   we need to settle the two drafts covering the same technology. after
         we solve that, we'll have the interim to discuss the algorithm.

10) IS-IS Invalid TLV - 10 Minutes -  Les Ginsberg
- https://datatracker.ietf.org/doc/draft-ginsberg-lsr-isis-invalid-tlv

Tony P:  thanks for doing it. If you need co-author.
Jeff
Tantsura: very needed. This clarifies what happens when security is asking.
         Looking forward to ospf as well.
Les:     one reason for writing this so new draft can reference.
Michael: I had experience with this, the LSP got dropped because no v6
         enablement. I support this work.
Shraddha
Hegde:   question about allowing unknown TLVs in the purge. are you suggesting
         if an implementation doesn't understand the TLV in the purge you should
         accept it?
Les:     you have version N, you understand some TLVs. Then new TLV is
         introduced after your implementation. you don't know whether
         that TLV is allowed in the purge or not, so you have to accept it.
Shraddha: what if new draft allow TLV in the purge?
Les:     we’re introducing a new backward compatibility issue.

11) IS-IS over TCP - 15 Minutes -  Gunter Van de Velde
- https://datatracker.ietf.org/doc/draft-hsmit-lsr-isis-flooding-over-tcp/

Acee:    no time for questions. How may people read it? Many.
Acee:    Cisco has an IPR, I’ll check.
Tony LI: I support this work. I suggest we use quic instead of TCP.


12) IS-IS SRv6 - 5 minutes - Ketan Talaulikar
- https://datatracker.ietf.org/doc/draft-bashandy-isis-srv6-extensions/

Ketan:    ask for WG adoption.
Acee:     how many read? Not many.


13) LSR Discovery of PCE Security - 10 Minutes - Qin Wu
- https://draft-wu-lsr-pce-discovery-security-support-00

Acee:     I don’t see no reason we can’t adopt it.
Chris:    if no more comments on this one. We can talk about TCP.

Break 10-20 Minutes

14) Level 1 IS-IS Area Abstraction - 15 Minutes - Tony Li
- https://www.ietf.org/id/draft-li-area-abstraction-00.txt

Acee:     so you're going to tunneling the traffic?
Tony Li:  no.
Acee:     that indicates level 1 router will have all routes.
Tony LI:  level 2 is going to exchange LSPs across the tunnels, and the level
          2 area border routers are going ot have to be able to compute
          transit across level and area, and provide that transit.
Acee:     that means level 1 only router will have all the level 2 routes.
Tony Li:  I'm suggesting that we tunnel from area entry to area exit.
Acee:     that's what I originally asked but you said no.
Tony Li:  we're not using the tunnel for area leader, but from entry to exit.

Chris H:  I could benefit if there are pictures.
Tony Li:  yes, in next draft.
Acee:     we really need to think about it. in ospf, we got rid of it because
          of added complexity.
Tony Li:  shared links cause scale problem.
Acee:     I'm just questioning the use case of needing an arbitrary area for
          transit.
Tony:     that's going to happen sooner or later as network scale up, in stead of
          flat you will need hierarchy.
Cengiz
Alaettinoglu: you want to be able to transit. How do you set the metric?
Tony Li:  I didn't explain this in the draft. There are no metrics that are
          advertised outside of the area. This is abstraction, and it has
          the pain of suboptimal routing.
Cengiz:    will you have loops?
Tony:     no, because you're getting the shortcuts computed through the area.
Cengiz:    you will have to have encapsulations.
Tony Li:  yes. inside the area, there are l1 router.
Jeff T:   two questions: MT and traffic engineering
Tony LI:  I don't see a problem with MT. but traffic engineering, it’s an
          abstraction, you can do TE within the area, even for the tranist
          traffic. but external traffic coming in this area do not control its path.
Jeff T:   remember for RSVP TE, border router wouldn't know what's inside.
Tony:     Correct. that border router can manage traffic within the area, but
          not from the head end.
Jeff T:   probably need to do some extension.
Tony:     there is nothing for CSNP to compute. area entry is the best place.
Chris
Martin:   imagine a leaf-spine router. The root router will be the ABR,
          and the leaves are more likely to be the inter area connection points.
          To your points, Jeff, you can just hide that information. I hope this
          helps to clarify the use case.
Tony P:   choosing an ingress will be difficult, you may end up with a bad
          ingress. you may try to advertise metric, but you end up advertising
          this full mesh with some approximation. It's inherent limitation of
          abstraction.
Chris M:  we don’t have TE between line card and router. everybody agree?
Tony Li:  if links fail within l1 area you can compute a new route that's
          gonna happen.

Acee:     I encourage people to read it. This is an extension to base
          mechanism. This is definitely interesting. Let’s have a
          discussion on mechanisms and deployment cases.


15) Hierarchial IS-IS - 15 Minutes - Tony Li
- https://datatracker.ietf.org/doc/draft-li-hierarchical-isis/


Les:       still wrapping my head around this. If we do this, we’d prefer to
           use 7056 encoding. There is no such thing in 10589 as an L2 only
           router, so circuit type 2 is actually illegal. I’d like to see how
           this is being used in discussions.
Chris:     I’m shocked. I didn't realize there is no just l2 operation.
Les:       by definition. Level 2 is level 1 router in that area. there are
           many implementations that has L2 only CLI just for convenience,
           so you don't have to got to every circuit and say the circuit is l2 only.
Tony Li:   no level 1 circuit only.
Tony
Przygienda: my only question you assume the highest level get connected, right?
           we assume people don't do stupid things like partition area 0 in OSPF.
Tony Li:   I don't intentionally suggest people do stupid stuff.
Acee:      you’re keeping all the same semantics, Level n manages level n-1.
Tony:      yes. in theory, you can have a router from level 1 to level 8.
Acee:      a circuit would be in how many levels?
Tony:      as many as you want. but those levels must be contiguous.
Chris H:   I'm wondering whether we have to configure the area according to
           the level?
Tony Li:   yes, it would make sense for the end sap to be hierarchical.
Stewart
Bryant:    I see case you might want to build a level 4 router with its sub
           component in level 1. this is a special case with discontinuous.
Les:       this is interesting discussion. Deployment cases. So people know
           the good ways to use them. we could debate.
Stewart:   this is a special case of level 1 being special.
Les:       let's talk about it more.
Tony:      I haven’t find a case to use discontinuous level yet.
Acee:      if we were to adopt it, we will need implementations.


16) IS-IS Mult-Topology Deployment - 10 Minutes - Uma Chunduri
- https://datatracker.ietf.org/doc/draft-chunduri-lsr-isis-mt-deployment-cons/

Acee:      is this standard tracked?
Uma:       that’s a mistake. It’s meant to be informational. I will fix it.

Les:       this is a well known limitation of the protocol, nothing new. If
           you look at MT, you can support whatever address family.
Uma:       you said address family, but it's a topology.
Les:       there is not limitation what address family can run in a topology.
           This is well known.
Uma:       I'm saying 5308 can also be used for IPv6.
Les:       you're correct. I'm saying this is well known.
Uma:       well known to some people, but some people don’t know.
Les:       there are some people who don't understand this, it was clearly
           stated in 1195. there have been proposals support tunneling
           between different address families, it never got a lot of support
           but this has been talked about.
Mikael
Abrahammson: there is a mechanism if I forget to configure ipv4 address on a
           link to bring it up, it won't use isis for ipv4. If I’m isis adj,
           and I have no ipv4 or ipv6 address on the link, I'm not gonna use
           the link. if you have MT, do you need both IPv4 and IPv6 addresses?
Acee:      we don't want to require IPv4, the goal is to go with IPv6 only.
           do you think this draft is useful?
Mikael A:  I haven’t read the draft. But so far I think it’s useful.
Chris H:   this may be useful across area, outside of routing, like ops.
Tony P:    I was against putting reserved topology values. We should have
           called it IPV6 transitional topology? so people know the toplogy
           concept is completely orthogonal to address family concept. I'd rather
           reissue the draft with 20 bits putting this not not not.
Jeff T:    this draft is not meant for isis implementations, it’s meant for
           customers. To help people understand it and how to deploy it.
Chris:     like writing a paper in a magazine on how to deploy it.
Tony
Przygienda:it may get lost.
Chris H:   he is trying to make clarifications.
Les:       I'm sympathetic to people get confused, but this was very well
           defined. But if we had thought about MT in 1998 about IPv6, we
           wouldn't have two TLV types. we didn't anticipate it.
Mikael A:  you have to write different kind of document for operation people.
           you don't get pitfalls from reading the protocol.
Jeff T:    we’ve been reading tutorials in routing. It’s good to have a
           reference.

17) Preferred Path Routing Graph - 10 Minutes - Uma Chunduri
- https://datatracker.ietf.org/doc/draft-ce-lsr-ppr-graph/

Zafar Ali:  the problems listed are not problems. You are looking for a
            problem to fit your solution. slide 3, we have solution for
            data statistics without additional label. it's not a problem.
Uma
Chunduri:   how do you do it? the problem I see is if FRR happens, you will
            mess up with your accounting, you need to approve it.
Zafer:        Please read my draft. I can approve you. you need additional state
            in the network, OAMs are complicated. You think it's simple but
            by the time it’s done, it’s complicated.
Uma:        binding Sid is also FIB entry adding network state.
Acee:       why don’t you start a discussion on LSR list?
Uma:        yes, I’ll.

18) OSPF Prefix Originator Extension - 10 Minutes - Aijun Wang
- https://www.ietf.org/internet-drafts/draft-wang-lsr-ospf-prefix-originator-ext-00.txt

Acee:       the changes were to remove different encodings, and some are not
            backward compatible. we only kept ospfv2 extended lsas. Ospfv2
            prefix attribute lsa, ospfv3 extended lsa. We took out the
            non backward compatible ones. Aijun had a use case if you
            have all numbered case and it was moved to appendix.
Jeff T:     regard the use cases, usually it's not meant to be propagated
            through the area, that's why BGP LS gets it to controller. I'd
            suggest to change the use case.
Jie Dong:   that’s why we put it in appendix.
Acee:       our compromise is appendix. we need to discuss it in the list. I
            think it's a good idea. You have a BGP ls draft? That one already
            adopted, we may also do adoption here. I may not be an author.