Internet Engineering Task Force 99, 18 Jul 2017

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF99-MPLS-20170718-1330

Description

MPLS meeting session at IETF99
2017/07/18 1330

https://datatracker.ietf.org/meeting/99/proceedings/

A

So with afternoon, MPLS is about to start. We will have two sessions, one is today the other one is on Friday morning and there will be also a joint session with C camp on Thursday. So note: well, it's a new note. Well, yeah I think.

B

So Lu put it in the t's meeting. Ask if you, for example, hum on one particular the raft, and you know about now. If you're on that raft, you have to disclose so everything you do say or go about here. It's under the note well and please take care and disclose if you name and read it carefully, yeah.

A

So a couple of administrative a--so, please speak to the mics. There should be pink squares on the floor, in fact, but I haven't seen one. Oh yes, one! Okay! So you stand inside the things. We are all materials uploaded to the MPLS working group page and we have two minute takers, I guess: yeah mideco is active and, as I said agenda and so on on the web.

B

So we have had a couple of smaller problems: George can't travel, so he won't be here. This meeting tarik was about to do the airport in Toronto thinking it was raining, so the flight was canceled. He should be arriving about this time, but he we haven't seen him yet. So if there is anything in.

B

Have a misspelling on agenda or slide tied to those things on the uploaded again, there's probably I. That's in there. It's not tired.

B

Okay, so we have again draft for today. I, don't think it's too much idea to bash the Friday agonda, because we will do that on Friday morning or actually I should say that first before I start, what we have done is actually we swapped the mpls slot on Friday to 10:30 borrowing, time from routing area, routing working group.

B

We haven't. You have a green sir, so we distort the mpls meeting on Friday at 10:30 and hopefully and 11:30.

B

We were doing this because we were in the same room, so it was just swapping us up in time, and some of us had something we needed to do where we could do during the mpls slot in the afternoon. So it was very peaceful. Okay,.

B

Anyone that song maganda one thing we have a longest Lord 30 minutes. That's not a lot number r8 begin to discuss bhp directed thereafter. It has been stuck for a very long time. We want to see if we can come to a situation where you can move with the head.

B

C

The mic jeff has so my request is to maybe bump the BFD topic a little bit further up on the off hand chance that other topics want to push it out. I will not be here on Friday that participate. I couldn't hear you close to mark and loud. Could you please move the DFT presentation today? Further up the schedule, so it doesn't get pushed out into Friday. I will not be here on Friday. No, it's not on Friday. It's today, I I am aware what the schedule says: I'm, also where the topics overrun. Okay,.

B

We actually have a ten minutes buffer at the antis should work.

B

So, let's do it like that we put it as slot number. Four, that's fine! Okay! Is that one? Okay with that Mac you just got pushed down. Is that fine.

B

Okay, great, so we had a slot freebie that we do the adult direct in and let me see, we have Carlson room, we have Greg in Rome and we have low in the room. So we have all everyone here, good.

B

Anything else: okay, we give with this agenda.

B

Let me see so blue sheets has been started passing along and when you sign them as the last one last one on the last forward them to the share a check. If there is anyone that is not signed yet.

D

B

So when you're done exactly coming down in the middle coming, this way, okay.

B

We have an errata that actually been approved and it was kind of the way I'm reading. It is just a grammar, change and I, don't have any opinion on English grammar, but other than that, it's hard.

B

So it's a problem, it's good. It should be kept for update. We have lace incoming liaisons, I, think everyone here. Everything here is under control and we will kind of work on a response with the other working groups and.

B

Yeah we haven't sent any liaison. However, we are preparing a liaison at the moment that will go out as soon as we can get the sound and it's just telling a stereo 15 that the ampulla TP shared ring protection has been approved. Okay, we have new oral seas, I think it's free of them. That's kind of can kind of normal between two meetings. The next time we have one new working group document and lots of working with documents that has been updated.

B

The blue documents in this, this ceaseless org document that will be on the agenda.

B

We have discussed the tempura temporal hitless PSM document today, I had his custody Deborah and we will try the one loss, Carmen sleeping at it, No.

B

Okay, there are a number of new IDs and, like half of them, are on the agenda later today or on Friday. So the list here is not that I do need to talk about each and every draft. If you have questions on draft you can you can ask them. Otherwise. This is just a bookkeeping, so we can go back and check what was that? What was that?

B

What was up at the meeting in in Prague 2017, and we also have a number of documents that might be interest for people for people here that are in other working groups. We list them for the same reason, just as a heads up and you can check. If there is anything you need to pay attention to and that one is.

B

B

B

So I'm, aware of updates for not updates, really it's a information on two drafts. One is the opportunistic security draft that one, if I understand it correctly, will be allowed to expire.

B

Unless Stephen can find a student to do some student work, then it will be revived again. We also have the 50:36 base still an individual draft. The next thing we will do in that trap is actually the write rewrite. A and a section I took to iron our last time and I got an instruction on what to do and I try to do it in they come out very awkward, so I talk to them again and then it will be updated.

E

You please clarify what the issue with opportunistic encryption is.

B

The the issue is that we have, we didn't, have any take it to do. The plan was to actually get students to do a student project around that and then forwarded as as an experimental drug experimental RFC. We couldn't find the student actually found the student that the student defected, so if they even actually can find the student we will arrive it again. That's no technically assumed.

B

B

B

F

Good afternoon my name Shaolin from juniper on behalf of all the out cars are here today we presented this unified, South, shouting use and yesterday post at here. So basically since version zero, three author encounter from Nokia.

G

F

Individual and Shaolin added into the Carson group here and then we are also modifying something on the adjacent since II, and also some text to indicate the benefit of using technology even follow up with it. I guess spring to man and this my sort of we're also working here.

F

So the first topic we will talk about.

F

It is a chapter linear as the overlay part, so the tracker, linear or the transportation pretty previously was sinking as an underlay of Transportation somehow, but the traffic engineer is actually a capability along the package size, somehow so with any transportation, if you're adding try to adding the traffic and in your capability as one of the overlay capabilities such as a one optimization another somehow, so that's why we try to highlight with sec, knocking label here with the label stack, you actually can add in an existing transportation, the similar traffic and in your capability on this one, so especially in certain successful deployment such as Google before and others.

F

So the really sinking traffic engineer is not a transportation or just underlays, pretty much the iron function. On top of that, so I think the.

F

This one is not the latest one. Actually, when we change the picture version to a different version: zero slice.

F

H

D

F

Still, okay, anyway, really I really changed a lot of a picture here, and so it's gone so the the idea here is: we use Japanese overlay here then p-pass is actually just a set of label stack. Then the transportation will be the IP GRE or IP UDP hider across given the node and so on. So in that case, first use a second label stack, not for transportation, just for traffic linear capability here.

F

So if you try to send the packet from one to another, then you want to really want to check a linear in in the middle here you can actually use GRE, UDP or even lttb. Any kind of mechanical. The certain node in the topology here is upgrated to support those assumption. So the key function here is when you first hop router receive the packet people found out the next capable of this shelter.

F

So the routing protocol updated, maybe just laboratory the existing second routing, v6, sorry section one sec, not in SSR OSPF, then the node, a and no E&G, and somehow they actually issuing the information you change. The label stack information for sec, no team, then, between the a and a certain node in the middle. They cannot support either. Mpls said no team or you cannot support like.

F

Totally understand this capability, then you actually sending in cap into a UDP or GRE header from A to E. Then the next one. You actually can push a label stack to to stand folder chapel in your capability here. So then the next node e you actually not looking for you actually D cap, the GRE hider or the UDP header. Then you look at the first label stash here. So this label is not swapped to another level. This auto label the.

F

So the label neighbor G, levo G, is actually look at in node E. Then my team to the UDP header, so the lookup engines, like you, have a label incoming label. Then this label, you outgoing, is a GRE tunnel or the UDP tunnel then fund to the node G. Then you go to the node h and so on. So that's a procedure here. So the k, even between this draft and sent routing normal test handling draftees the forwarding behavior change. So previously the SEC routing you actually pop up a label.

F

Then you actually swap all follow another label, but this time use reasonable. You look helper for the next edition C, which is the tunnel somehow, so that's actually I'm working for both ipv4 and ipv6, and even even certain node in between they only support ipv6 forwarding no capability for MPLS and so on those boxes. If you upgrade to this capability, then they can forwarding as as as well for ipv6 as well. So next one sorry previously one yeah so the another capability for this one. So we're adding this label stack of ultra near capability.

F

Then we keep the viewing capability as another label here, so this measurable working with any existing between mechanical layer, two layer, three EVP and so on. So all the existing protocol will be support here, so in in that case, if you look at the label stack here, the the innermost label reading instruction that you can differentiate different people between demand here then right channels, just transportation that in the middle is T pass or the second note in chief engineer capability, add on.

G

F

G

F

As well so you follow the similar function, then the last hop or the top h. You actually look at the papian label. Then you go forward to the different reaping instance here, so certain advantage over second v6 here we saw is actually we have a fallback and, as labeling encapsulation other than the second working with six assists in bad, then we also backward compatible with all the NPS spring. Here then seamless integration with existing NPS between and byte allowed panels because of a UDP saucepot.

F

You can use a lot balancer of from first hop router point of view, but if you mean capitana from a host, then you you lost those capability as well. So another advantage here is all the existing hardware, including the latest, broken shape like trend, m3 Jericho can support engines over GRE MPs over UDP. All the major vendors box can support this one without any. How well change so that's kind of a advantage here then we also not to try to replace I'm just as rare here.

F

So it's basically the incremental deployment with an TSS brain so for certain customer, the really sink, 5g ipv6 will be the future, then get rid of the enters transportation then use ipv6 as transportation, adding the G function. On top of that the benefit for this one. So basically the solution overcome the lot bonus problem. That is, there's actually two problem with signaling with existing. How we're somehow, basically certain a seeker can only read certain apps of a label for load balance in detail, then another wise, the ingress router.

F

How many label can push that's another problem here, so you either way. We know we have a certain alternative solution, but this is a true problem for the existing, our own implementation, which is a protein. But with this new method you have a better choice here.

F

So let's come to the next step, so any comments on suggestion any anything so I'm. So sorry, the the slides are kind of a mess up. We do have a better version.

G

I

Can arrange documents first, one I'm not sure to see what this turn up track document is trying to specify if you remove the use case on the figures, it's a one-page document that you can summarize in one sentence, which is: if I have an empty list: node and I: don't have amperes connectivity to my destination effect. I can use an IP turn and that point is already specified in the MPLS in UDP RFC. So what is the new specification so.

F

The new capabilities use a transportation tunnel together with saty stack for the tracker Minear previously, if you only do the tunnel part, you cannot have the per hop forwarding behavior then, which is a very simple use label to tunnel. Imagine from.

I

Our point of view I believe the MPS in UDP draft. You can put any MP respect, including a stack of la vella, so.

J

My name is Yahoo one course or this job. The purpose of this job is to describe how to leverage the existing MPS capability, including amperes and routing and ampere, so VDB tunnelling technologies to realize the te overlays over IP chessboard IPO lies IP on the rise. That's that's why this job is information, informational, no.

I

It's track, it's not attract. Well, it is a static track. Correct, okay,.

J

Okay, oh right, it's dropped. What changed from informational to stand trunk is we want to reduce the MPS sauce pot label a triple field in that way, there no need to insert each upon labels between the MPS label stack in that case, when chance it sr rock broader, the capsulate UDP tonneau is should remember the saucepot value that is the a triple value and then, when it's perform, another encapsulation is, can reduce the cash to boom value. That's different from the existing MPs forwarding.

J

Rollin capability, so that's why this.

I

Job is it's different from MPs to impress, but it's not different from MPs to you, dp-tuner MPs in your divisional, which is already specified on the.

J

Encapsulation in now chain, but the difference is we want to reduce the sauce pot label and only it simply fail for the MPS over UDP Paquette. That means there no need to insert any super labels, bring the label stack. Okay,.

I

I think we can continue on the list, but okay.

J

I

Not seeing the the specify the parking in the draft okay.

G

I

Question what is the relationship with draft Bryant, which is called a unified approach to IP segment? Regina I, see a lot of overlap between to draft and kind of technical content on teleporters. So do we need a true document? Are going to be made or.

E

We're still talking about it.

I

But the issue will be resolved, you mean are going to be matter, okay, but you target one draft.

E

I

E

We may need architectures as whether we're not sure how many drafts of obscure.

G

E

To the mic, we're not sure how many drafts it would take to properly describe this, because it's part of a bigger problem, as you will note, with the next draft, that's going to be discussed, let's, let's let we need to work through this there's tough individual drafts. We still need to work through the process. Okay, so.

B

We need to keep it short now we're running out of time, quick question, quick responses. Yes,.

I

So this will, for my comment, is arranged now with the Supreme Court Rio, exactly all your draft, all your slides of our boats premium or Sigma routine. So why do you think MPI squawking hope it's a good home for that draft, because it's all it's all about spring it. So you will probably have better feedback. You.

B

Can I can take that with when anything sounds income? You know, that's a new draft and it has some relation to more than one group. We start presenting anything one group and then we socialize it with the groups that group wouldn't you've shared that we think would be interesting in this case. That I think you should be one of them. I think Lou and T's group could be one of them and we decide which group that you take this.

I

Okay, but from the draft for further slides it's more about spring and yes,.

F

A second inside there's a tunnel mechanical and we try to overall see the value in the existing topology and network and so on. So.

I

If you want feedback, I think you would have a better feedback in this printing. Okay.

F

I

F

About them, offline.

K

Machinery Cisco Systems: can you go back to slides one more I just want the one that has nice diagrams colorful ones? Yes, this one. If I have a IP IP tunnel between a and E and I click that the virtual interface? This is exactly what you have here right, I'm, asking the same question: what is new here? It's just you're running Sigma routing over a virtual interface, which is an IP tunnel. That's it right.

F

So take this question but in.

J

This approach there are no need to can take any explicit tunnel between a and Yi when a want to Sarah amperes packet, the top label, in a case e, but there are no outgoing label for yi. So a will pop the top level. Then it caps with MPs packet between UDP, tano, okay,.

B

I think I have to do what sheriffs often do in this organization. Take it to the list. Please.

G

Okay, I actually.

B

Take it to the list, because the discussion needs to continue. Okay. Thank you very.

L

Much Asif yandex just a short comment on pushing large labels tags, at least in control environments. Now we can extend in POS forward and all the way to Sage modern topic. Chipsets are perfectly capable of origin in pairs, and that means at least in some cases we can push label stack on the host and in fact there are good arguments why pushing long and complex label stack on the is a good idea, at least in some cases, for.

F

Sure and so pushing a little longer, these liberal status, not problem, but in the middle, the chips ad capability, how many label you can read and then make the decision for the Lord balance? That's a problem. Okay,.

B

F

B

Have to do on. Thank you very.

F

J

Good afternoon they talk about how to leverage the unified sauce, locking Magnum to realize service chain changes since a lot like a meeting three closers and Eden, and some tensed about how to carry metadata of adding MPs package is ending as well, and you can see from this picture. The service chain is real life, add overlay, just as what had been mentioned, the pipe so on.

J

In the previous presentation, the MPS stack, which is used to indicate servicing or service path since vast fortune, as in encoded and ambience labels that that label stack is then encapsulated over UDP tunnels.

J

So, with this approach is fully meet the transport in pendency requirement for the surf chain, which is listed in the searching architectural requirements RC, since there are no explicit protocol ID field within amp a packet, so it will be hard to indicate whether on method is contained or not. Reading our MPs packet. There are multiple.

M

J

Address this issue, why a tree along the service function for water to ink to allocate true labels? Wine indicates the presence of metadata another a to indicates the absence of MAG data here matter date. It contained the ink.

J

The matter could be contained in network service handler. That means the service network service hinder playing role of metadata container. Only in this case, of course, no details need to be specified in future versions. Advantages over message. First, land states on sales function, forwarders the same with claim benefits as our v6 based solution. Approach. Second, is leverage the efficient MPs network programming capabilities. Third is build on the existing MPs hardware. Forwarding capabilities comment sessions.

B

B

Thank you, okay, thank you.

B

So we are at the BBC directed slot you're gonna start with low talk about IPR, and things related to that note. Well, and other things then begin to make Carlos go over his review and then Greg finishing up, and there should be time for discussion.

N

Hi I'm Lou, burger I, was asked by the chairs to talk a little bit about IPR, not sure exactly where the group or the chairs want to take this so I'm, just gonna give you some high-level background. If you and then open the floor to questions, hopefully only take about a couple of minutes. This is the note. Well the new note. Well, it's basically the same as the one we've used for the last couple of years. The biggest difference is is that we now highlight that RFC 81-79 is available.

N

81-79 provides an update to the IETF site, Dr rules, disclosure rules, and notably it clarifies what is a contribution. What is not a contribution? Basically, it says that anything you do on any of the in the public arena of the IETF is an IPR contribution. That's whether you do it electronically on a list, whether you do it it within a design team, whether you do it in person here by saying something or if you participate in any way in a working group activity and a working group activity- and it goes into.

N

This is certainly giving a presentation like this, although obviously I don't have any idea on it. Talking at the mic, raising your hand, humming is part of a a poll or even expressing some personal opinion through whatever means like crowning. When someone is presenting an idea, that's even considered a contribution.

N

Basically, the rules also say that if you contribute to the IETF or contribute to work in an IETF that you're subject to the disclosure rules and the the I think I have this on the next slide.

N

The disclosure rules basically say: if you participate in any way, you contribute you're required to disclose any knowledge you have about IPR on the topic. Sometimes it may happen that you contribute and then find out about IPR. Then you need to disclose it at the the most honest, a convenient but timely fashion, as you could you can, if you know about IPR and you want to contribute, but for some reason you can't disclose that IPR the rules say you can't. Basically they say you can't contribute.

N

So if you know of IPR and you're not willing to disclose it don't participate. I had this happen to me personally, I just never walked in the room of that working group ever didn't even subscribe to the list.

N

There are lots of ways you can handle it individually. You should take a look at the the duck of the new rules and decide how you want to deal with them. Clearly, some conversations that happen while you're at the IETF are not subject to IPR disclosure in it, and the document 81 79 clarifies if you're having a private conversation, even if it's in the building, but it's just the one-on-one thing- and you know maybe it's vendor to customer, or maybe it's to a colleague and you're talking about some design- that's not subject to IPR.

N

So it's not like the rules or trying to be crazy. Here there are trying to be reasonable. So what do you do if someone doesn't disclose? First of all, from a working group standpoint, it's a working group chairs responsibility to deal with that situation. Anyone who's participating in the working group. Anyone is participating.

N

The IETF is welcome to help identify that that situation is going on, but really it's the chair of the working group to figure out how to deal with it when it's in the context of a working group, the ADEs play a role, particularly as you get to the sort of most more severe responses. You definitely have to end and the ad can certainly be the person who's alerted to about the IPR.

N

There is a way to do third-party IPR disclosures and that's also covered in 81-79, the there's, an RFC from Adrian who's, not in the room he's chairing a different session that talks about what are the sanctions available to violate through the violations of the policy. The sanctions are actually I, wouldn't call that call them all that heavy they're related to our process and the the draft lists tend items ranging from. If a chair hears about the IPR or even an individual can do this.

N

You could talk to the person and say hey, you know, there's this IP right here. Please dispose it as quickly as you can and sometimes you'll, see. They'll, say, yeah I know, I should have gotten it done. We just didn't I, just it's before it's working its way through legal. That happens. We deal with it. If the lack of disclosure becomes repeat a repeated offense or it's pretty egregious, you know the the person who's.

N

The author of the lead off of the document is also the lead author on the on the disclosure I'm, sorry on the patent filing and the patent filing was five years earlier. You know clearly the person should have known: that's pretty blatant. You might start looking at some more some of the more extreme responses, and basically you can.

N

The chairs can end up removing someone as an author moving them down to the acknowledgement section or even if it's really a repeated thing, you can just stop them from participating in the working group and to do that by the way you have to involve the ad I think I've covered this at a high level. I may have missed something I'm sure I did. But if you have questions please so.

B

N

By the way, I'm, just speaking as an IETF participant and someone who's been a chair and had to deal with these issues over the years, I happen to also be on the IOC, which makes me a trustee. I am NOT speaking as a trustee in this forum. Okay,.

O

Greg Merson city I.

O

Experienced that many times with the early disclosure, while before their application being published, it creates some confusion because no.

I

One can really evaluate.

O

The relevance of their IPR to the draft, if.

N

You notice there's no statement here about evaluation of relevance, so once an IPR is disclosed, it's up to the working group members to decide how they want to do what they want to do with it. The IETF no IAD member is G. Member working group chair is going to evaluate the statement. They say whether whether it is a reasonable claim, as a working group chair I've, had documents which are sort of like problem statement and someone does an IPR disclosure about them.

N

I won't say whether that's reasonable or not, I say what texts does that apply to, and they say these paragraphs I said: okay, those paragraphs are removed from the document right.

B

N

That I've certainly done that as a as a chair, I've also seen working groups that when an IPR disclosure comes in, look at it and say you know what we don't want to work on that that direction of that technology. In an extreme case- and it's talked about in the draft- is you can empty mentioned on the slide intentionally? You can the workgroup can say fine, we're, not gonna work on that technology at all. You know, maybe that's not our main, our main business.

N

This is something we want, but because of that disclosure, we're not going to do anything in that area that that's a reasonable option for an available option for working group, but it's up to the working group to decide. Basically we have information, and then we follow our consensus process right.

I

But again for somebody in the working group who wants to do.

O

The research himself, their existence of the published application is.

N

Important, it sure is very convenient, you're right absolutely, but it's not a requirement. You don't have to. You. Don't have to wait for the information to be published the patent application to be published in order to make an IPR declaration. In fact, as I personally interpret the rules you shouldn't. So if me is a individual, go write, a patent application and then go write, a draft that that patent application relates to I believe I have to disclose at the time I publish the internet draft.

N

That's the way. I read it I'm talking just as an individual.

B

N

That's how I read the dress.

B

I agree: that's the by.

N

The way this is the exact thing that I was referring to for me personally: I had a patent application in area, something came up, things were in public, I, didn't want to file, I didn't wanna talk, I didn't want to disclose it because things were still working its way through. So I just didn't touch that area whatsoever. Okay,.

B

O

Agree because the thing is that personally, that's the policy I follow and try to encourage the holder of IPR to do the right thing. Okay,.

E

So, whilst this is an interesting area- and actually in my view, is that the process is fundamentally broken because you can't have a detailed discussion in the working group about the details of the IPR I'm curious, why are we having this discussion here of MPLS were a grouping, particularly bad, because this seems to be a discussion we should have in the area meeting tomorrow rather than in a technical meeting. Yeah.

B

That can take a party response to that. I have a situation where I need to deal with the document we will discuss and in our PR disclosure against that document and I want to know what I can do. What I can expect the different parties to do and what I we can do if they don't do what they, if they don't do what they should do. So we have a live problem that I think we had the problem and I wish that people would sit down and say: okay, it's gone away.

B

The situation is like this: we had a draft through working group loss in work, new law school. We got an IPR a blocking arc. We are comment on that both of those person that actually made that comment at that time has now said that, okay, that they don't they have Whitlow on their block in common, so I'm happy with that. Now.

B

That is partly because the draft has changed, but the IPR holder has not done anything and I, don't think we can actually require the IPR holder to to do anything. So it's up to us to evaluate and do ahead from that from that point, you're.

N

Correct there is no nothing that the ietf can do to compel an IPR holder to disclose to update a disclosure to remove a disclosure. There are certain recommendations in 81 79, but they there's nothing that will compel them. It does also mention that failure to disclose IPR and then bringing work into a standards body and knowingly not disclosing that IPR in certain venues has resulted in the the patent claims being invalidated.

N

That's what it says that in the text that it notes, it notes that so there's some good incentives out there for IPR holders who want to enforce their IPR to disclose, but those incentives aren't imposed by the ietf they're.

B

Coming from a legal framework that we actually had the quite different situation here, we don't have a failure to disclose. We probably have a a over.

N

Disclosure, so you, as a chair, can't correct that all you can do is as a working group look at it and decide whether or not you want to continue with the document continue with the technology change. The document change the technology and whether you want to submit for publication request. That's all you can do I.

B

N

It would be nice to do more, but.

O

Cut off peanut arrow Cisco, Lu I have not read. 81-79 I have read previous versions a to see every letter. I still do have a question which is not an offer. Oh you.

P

O

Didn't so you're a better position to answer these than me for sure, there's a draft right, hypothetical situation and IPR gate is closed against it. The working group works around the IPR right. Changing the document. Moves on a different wave removes the paragraph that you're needed in your personal example.

O

What are the requirements to the to the IPR holder or to the person doing the disclosure? If they know the IPR doesn't apply anymore? If any like? Are there.

I

O

Or can that be forever ad infinitum in there? So there are no requirements on IPR IPR holders. In this.

N

Process whatsoever full stop to do.

G

N

All right and there are obligations to a disclose, disclose e, but it doesn't include that they must remove IPR that they believe it does not include that, whether that's an oversight or not that's a good thing to bring up at in Singapore, because the authors of 81-79, notably George who's, the paid attorney for the ITF he's going to be talking. If they're, putting together a presentation and we'll be talking through this, so that'll be a good place to ask: why that's not there or should it be there I can tell you it is not there.

N

There's no obligation to say remove in that personal example that I refer to that draft still has that IPR yeah that it's a still draft. The draft still has an IPR reference against that, even though we know tax been removed, reality is well if it never goes to court we'll deal with it. That.

E

Sure brawny I was going to basically reinforce your very last point, which is that no one knows the validity of the IPR or the applicability or the applicability to the change texts until the judge has looked at it and decided in court and that's way beyond our society. That's right.

N

All we can do is a working group is decide whether or not to progress. A document simple as that.

B

And actually the licensing condition actually comes in if it's a free for use, then that.

N

That may be a discussion in the working group. There is no position that I know of within the ietf management structure or process structure. That says that, but.

E

Wait a moment we're explicitly excluded from discussing licensing terms. So that's a piece of the broken process that I pointed out the scott-so paradigms.

N

E

We can't discuss those, we can't actually have the conversation.

N

You are I'm, not sure, that's completely correct. For example, the document 81 79 talks about that. There are certain areas where the IETF has expressed an interest in certain IPR licensing terms, for example in cryptography and that's the example uses in the document. There's an expressed bias by the IETF to do things that are I, don't remember what the right term is, but there's some license that they prefer. So you know, for example, if the iesg did come up and say we prefer this. That would be. That would be okay.

N

The document does go into a little bit of an excavation of why the ITF currently doesn't do that. So.

L

N

As a working group as a working group, you know you can look at a document and say I, don't like this disclosure I because of this disclosure I'm not going I, don't want to support this document right.

E

But you can't, you can't say: I, don't like it, because I've got to pay money or I've got to pay this much money or negotiate the amount or anything like it's a whole load of stuff that we're absolutely forbidden from from talking about you to add to trust. If.

N

We're getting into what we're allowed to talk.

E

N

Allowed to talk about I think at that point we need George to come. Talk to the group I'm.

E

Not a person who's I'm, just reflecting what George said to me on a different occasion: okay, okay.

B

Go down! Thank you. Sue Carlos,.

B

So what I think I have at the starting point here is actually a bit both Carlos and the colonists comments on the document and I think I had a mail where Carlos and the author says that they are prepared to work jointly to resolve the issues if I correct, yes, okay, I think.

O

That should be the starting point for this discussion. Yeah, absolutely so I think, probably a more precise way of saying that is I was asked to do a review for this for the RTG directory Directorate part of that uncover a bunch of issues. I would love the working group to come to terms with the issue. Insist I mean this is not a Carlos conversation.

O

It's not like Carlos has issues I I'm, just highlighting you know a few deficiency that I see on the existing text, but I have no say in anything other than doing a review for the Directorate for dat yep super. So what am I doing here today in front of you Louis and an email saying? Look we have this document is being through three unsuccessful. Last calls Carlos yeah.

B

Strictly that talk, I had no problem with you disclosing it, but that mail was not sent to the working group it to a smaller group. So what Cora says is correct as done for it, but it hasn't been sent to the working group. I did.

O

Not send the email to the working group I'm just explaining what I'm doing here. Okay, so he says, there's current issues, let's dedicate time on the meeting to talk about him and we're going to allocate whichever amount of time to discuss that and I need you to present, and the main reason for me to sign here to including this is because I was requested to present.

O

So the background on the document deep diving into the into the technical issues. Now first working group last call I had not even read the document but, like Laura, said IPR blocking unsuccessful on the safer work working group last call ihe is when I first up a first set of comments and that that was also a an unsuccessful last call.

O

The result of that was removing a whole section on segment routing from the document, which is like half of the document I think, and that was because he was actually using label instead of fakes, to refer to a return path.

O

On the third working group last called another review, another set of documents, another set of comments, the net result of that and the bump up in version after that is because the authors decided to allow multiple sub TL widths, which was one of the one of the. You know initial, very early comments.

O

When we look at this without allowing multiple sub tlvs, we don't have the necessary degrees of freedom, touch it direct the return path through things like a tunnel or through things like a nested set of LSPs and right now, the last one is a routing directorate reveal which again link are posted there, and what I want to talk about right now are the specifics of their technical issues right.

O

The the net summary of that from my experiment view is that the document is underspecified, there's no enough to actually go down and sit down and implement an interval specification from it. So the issues the first one is: it's, not it's pseudo technical, perhaps, but it's about the motivation that is actually lifted. There is actually listed on the document. The document talks about the main motivation being the return path needs to follow the exact same path that the forward direction of the LSP is taken, and he goes into texts like this.

O

This is verbatim from the document. A failure detection by the ingress node on the reverse path cannot be interpreted as bi-directional. Failure unambiguously on trigger switch over bla bla bla. To address the scenario we gonna specify we're going to allow the specification of a particular return path. However, doesn't that does not solve the problem? I mean you can still not unambiguously interpreted by directional failure and, in fact, there's reasons why we traffic engineer traffic to not take the shortest path.

O

Various reasons. Here's you know a extrapolated example, but really if we want to have a forward path in red which is going ABCDEF and the shortest path on that network is g2a, it's probably more likely that we're going to have a better chance of having a quality return path on the shortest path, as opposed to trying to back engineer all the way.

O

Exactly part of the point is that even if I use the same note or even if I use the same interfaces, there's no guarantee that the programming is happening on the return path that is chosen as opposed to a different one. There's no I'm, not you know, passing opinion on whether this is useful or not. I personally think it has some applicability, potentially I. Don't think the motivation is, you know well, defining the dogma is that strict sauce rating is.

E

That strict source routing, which on which any I to the HJ, the red one yeah eco divino, if it's not the nice EMP and the you CMP selection, means that it's impossible to guarantee what the part is in either direction and.

O

I agree, I agree. That's part of the reason why the motivation has laid out in the document is bogus right without even without going to, even if you specify inner face by inner face strictly it's now exactly and and if you don't, then you still have a lot of other issues. Now going more, you know known as early on the motivation on the technical things, the structure of the effect sub TL B's. This is text that is totally confusing because it doesn't really specify what to do.

O

It really says we're going to include a number of sub tlvs. None one or more may be included, and that's how the return path specified. If none are found, then you revert to a default which is our IP, but that doesn't really explain what if some are found, and what can you use an ill fake is allowed on the document. What do you do with a pastor has an ill effects, a specification. What if you actually have the first two sub tlvs understood, but not the third one.

O

You know what, if some are found, what, if the LSP that you specified on the return path, actually goes in a completely different direction right. All of those things are just not.

E

Contemplated sorry, colors can take you back. Why is the problem? Why is there a problem with the thing we discussed just now, because what you're gonna do is you may force me detective, which is a fail-safe thing to do, and probably what people would prefer so I, don't understand why there's actually a problem if you declare a failure due to a failure in either the forward or the reverse part, the univ it's different, so you're.

O

E

It the other way around this.

O

Let me let me penetrate the question: are you saying why.

E

Do we need these solution yeah I'm, asking why we care whether we get a false, positive, post, positive detection, because it's at least piling safe.

O

Is more of a question for the authors you know, I I think is probably important to okay. I had I think it's probably important to have the ability to select multiple paths to actually correlate and but there's nothing that you can really say unambiguously.

O

This is, you know, one of the one of the other questions that I raised a while ago, which is confusing. The document says, if you don't understand this up till visco to a default path which is over IP, but when I go to RFC 584, there's no default, there's no default path, there's no default via IP that takes from 58 84 is copied there and by the way.

O

All of this is on the mailing list already right so I'm repeating, but it basically says you can go over IP or you can go over an LSP and there's a uppercase May for both cases under Sun or in between the two paragraphs.

O

There's. No, the word default doesn't appear on the document.

O

One of the reasons why this is a little bit more of a a problem that is a little bit more complex at the over simplified solution that is presented either following this is mimicking what was done in RFC 7 1, 1, 0, 4, LS peeping, and that is fairly straightforward in the sense that you have a command response protocol in which the properly selves carry the instruction of how to return. So an LSP echo pocket is essentially saying send me the reply right now to this packet, using this path. Right and and there's.

O

You know, complexity in doing that. However, we have these a different monster in here, because because of a couple of reasons, the first one is because there's persistency this is a protocol. There is no request. Reply is a bootstrap configure any runs ideally forever if it doesn't detect a failure, so you have to do with the persistency element associated with changes in topology and how that state gets maintained on the VM on the VFD endpoints right. So you know questions like if I specify your return path.

O

That is not available right now shall I say that and switch over whenever it becomes available or if there is a return path right now, and it goes down right, I mean, let's not assume topological. Invariably T I mean things change. You know, that's the whole purpose of having BFD what happens if I am actually on a return path and then that effect goes down or that release pecos down. Should it there's no default to go back to so that that is another area that is under specified.

O

In my opinion, there's they're still after a few iterations, a number of issues which I don't think fully understanding. So, for example, there's a return code that says failed to establish a VFD session and right after that it says, vigorous may establish the VFD session, so you're essentially receiving a response code. That says I failed to establish the session and then the protocol says you may establish it right. So there is self-contradicting.

O

In terms of state that this import- and this is you know, a question for Jeff- perhaps that this potential impulse additional state considerations on BFD should the end point have to somehow have a new.

O

Variable like way of of saving the return path or not so.

C

I think this is operationally useful and interesting, but I don't think it is operationally required. The PFD RFC requires knowing whether the session is up or not you get that stayed out of thing. Behavior.

O

O

Switching to a different return path is similar to what I was talking about. If there is a need to switch to a different paths, but how how that exactly happens there is in 58 84 a provision for if I, remember correctly, optionally, send LS beeping messages after the bootstrapped, so not for bootstrapping and and Jeff is saying no with the head.

O

But you know, even if there is I'm, pretty sure, there's no requirement for it and when you take that down to practice, I'm, not aware of an implementation to actually tacit- and there may be, but I don't know them. That.

C

What is actually in 58 84 is strictly about how to bootstrap a new session of the ERC is silent about how you would actually provide changes to the session and I think the most substantive BFD comment on this is if it is desired, for these EFT sessions to be able to not like the bootstrapped, with a fixed return path, but be able to be up able to update them. This is an update to 58 84 procedures, rather than strictly a application.

O

Thank you, and that was the last question. The third update the procedures in 58 84. If we want to have you know, updates to the bf the incision state while well is running and the last I think and before that does torque.

E

Why do you want to sort of modify an in-flight BFD session as opposed to starting a new one up which, after we're only trying to find out where the other end is there, so have two sessions and kill one of them when it's not fit purpose.

C

So I could speak to the motivation of the authors. The the general impression I have on this document is we're.

G

C

Bfd is bi-directional: yeah, the authors are really trying to do a unidirectional test, but the bi-directional technology, a consequence of this is you know. Even failure in the return path will falsely flag know that the forward path is the problem. The technology that's trying to be specified here is an attempt to get the return LSP as closely coded to the sending path as possible to try to reduce the problem space I, like all the comments that they made thus far, all good reasons why this does not always work. So it doesn't repeat it that.

E

Didn't the PSP version 1.0 have a unidirectional negative operation, no.

C

It does that it's always bad by rationale.

O

Thank you and they last the last piece of technical comment on the review of the document is 58 84 cause at some particular requirements instead in terms of the UDP port usage. So when you send something over IP, you use a different border if you send it over on LSP that this is, if we're talking about a bf this that can change the return path from an ASP to sender over IP there's also a change of UDP port and the implications of that are also not explained.

O

Thank you. Any additional questions.

B

Any question for Carlos so Craig.

M

O

I think that that's what Stewart and Jeff pointed out is that we do try. Okay with being the over MPLS LSP. We are monitoring unidirectional construct with a bi-directional instrument and, as you see here so if we are monitoring LSP a to B, then failure on reverse path will be interpreted by a as a failure of LSP being monitored, and if we have a protection switch over associated with it, it will train your unnecessary switch over.

O

The purpose of this proposal is not only to make it corrupted, but it's possible that if we deem their network that used for reverse third class as unreliable, unstable or less stable than other paths, we can direct the reverse path over the more stable.

O

Network, okay, so it's not only intention of this proposal to make to create a corrupted beam these sessions, but in as instrument to control, reverse path and based on observations of operator to make it more stable, more reliable and more deterministic in the term of detecting the failure.

O

That's the real intention, and yes, of course, in some cases it might come to creating carotid being this session.

O

So the opposed mechanism is yeah here this this ultimate carotid, but that's not the only wall of this proposal so as Carlos characterized. Yes, we propose to create new the if you reverse path, till V to be used in conjunction with a VFD discriminator t OB in LSP thing to identify the D session to wish it's applicable arm and then our in there being the universe path. Tv to have subsidies that describe the return path.

O

To some comments to the comment of our IP return path, RFC 58 84 says that the return path over the LSP may be possible, and this is bi-directional LSP, so I'm not sure really about how many bi-directional arrows piece we do have as far as I know, only in MPLS TP but MPLS TP does not use IP encapsulation at all.

O

O

What else well, the document bid we have. We have very good discussion about procedure of using LSP, pink and I appreciate the Carlos and Jeff clarifying that the concern they have is that the mechanism of using BFD reverse path to V in conjunction with a VFD discriminator and the changes so basically when to use and how to use that.

O

Yes, it was implied in the document. They need to make clarification and we offers. We agree that there need to be clear statement. That explains what is expected from LSR that supports this mechanism, how to process us being with the same being the discriminator and the University will be so basically that they must be prepared to receive repeated LS beeping with the same beanie discriminator. If the reverse path to be is present, and we are ready to make this update and along other updates. That working group of suggests.

B

O

Nutella Cisco Greg in the interest of correctness. What you just said is incorrect. You just said that 50 84 say that you can use an LSP return path if they're, crowded, I, think that's what you said no means.

I

Know if it's bi-directional bi-directional.

O

Lsp, thank you. That is incorrect. That is not what 58 84 says. 58 84 and you know, I brought this just to be very explicit. It says I mean I copied on my slide price, but it says, or the BFD control packets and by the egress al-azhar to the ingress eleazar may be encapsulated in an MPLS stack, full period. Then it says this may be the case if the effect for which the fault detection is being performed corresponds to it by directional, LSP or an MPLS to the wire.

O

This may also be the case when there is a return LSP from the ingress eleazar to the ingress eleazar right so basic says you can do it as long as you have an LSP back.

O

No, we are provide the mechanism to make it deterministic.

B

Okay, any the questions.

B

B

Carlos, based on your review and the responses you have, do you think there is a possibility to actually converge on a viable draft in the color near future?.

O

Carlos pignataro, like I, said before law is not up to me. It's up to the working group. I mean that working group care is there support. Do they I mean the comments for from the directory that I that I made or as much for you? As for the authors? As for the least on the line by line response, that I saw half of the things are an answer and the other half are either being understood. The comment or or the some additional disconnect I.

O

Don't know how else to answer the question: okay,.

B

C

You Jeff I was I, I have three comments, and while these reiterates other things Carlos said, but in the slides problem number one is whether the mpls working group thinks this is a useful problem to solve. No, obviously, this is something you all have to do amongst yourselves problem number: two, no there's technical issues within the document themselves in terms of how this is actually no address, no feedback from Carlos and the other people can help the document be better.

C

If the first one is the group says yes, if not, why would you spend time if you want to take it to the independent submission track? That's one potential, but if you're trying to do this in the MPLS group answer that first question. The third point is that minimally, with what you're looking to do this, it would imply fifty eight eighty for procedural changes.

C

So even if you get the things resolved and the second piece we're probably going back to bf to get a 50 40 50 80 for update, so there are three things that have no increasing levels of specificity.

C

My suggestion is to tackle them in that order.

B

Okay, thank you.

B

Okay, so thank you. Everyone, yeah, I, think the working group one group chat had to go back and think and kind of formulate a question to the working group. How we want to proceed on this?

B

Okay, Mac you up next.

M

Okay, hello, this transition is about update today are speaking for la interfaces.

M

Okay, here's a history of audit rap Darrell, Iverson individual submit in 2014, and it was experienced several generation before it was adopted as a working contract and there's none of knowing understanding the issue and Commons addressed a burger. The draft has adopted by the working draft in that innocent year and the job after adopted or in court wrapped. It has been there for over two years and this year has requested by the co-chairs I'm joined to don't try to move to move this work forward.

M

Well, here's a recoilless wrapped. So the trap is motivated by an issue that is encountered in a in an I blame work and it is required to do validate wider and it's being traversed over specific member link of a lab tends to in to isolate. Where is a filler point and the socialist question multi straightforward and a first step we can.

M

The two are: two servings is used to discovery, to discover all member link of lag and relevant informations, and then we can use the relevant Commission's to validate whether our experience traversed over age or over a specific member link by using the DDM app at curvy.

M

So yeah, yes, as I, said, there's no substantial update since it was adopted and the changes made to the perhaps the main editorial changes and some object to the references so yeah. We really appreciate more what incorporate the news on the trap. If there are any comments on issue and we were each orbit and familiar for the altars as in we've lived at the drug may be, is it stable and he's ready for looking good? Let's call.

B

Okay can I ask this is the draft that is about three and a half years old, something like that. That's a working group that how many sites you read this stuff.

B

We have a few and four five six people, so I agree with Mike. We appreciate much more working group review on this, though.

B

We probably have to figure out how to do that. Maybe she has triggered an early review routing Directorate to get comments on the list. That's a probably best to do. Okay, okay! Thank you! Okay, thank you, and now it's harsh! No, it's good!

B

So one thing about Stewart presentation: I think you do need to talk about four different drafts. Yes,.

E

B

The procedure will we only list it free but begin to discuss for anyway, so again ah the agenda. Yes, that's only listed free drafts idea, so.

G

B

Has an objection to that? Please stand up and kind of walk out for something. Are we doing to talk about the old four draft anyway, I'm.

E

Gonna talk about his status actually, so the appeal is flow. Identifying draft has finished we're here at last. Call lower picked up some minor points when he was writing the Shepards report, a stale reference and a minor issue with with Ayanna for completeness I've addressed the the issue in the text and I will upload it after this meeting or before the end of the week. So that's pretty well done. As far as the working groups concerned, the SFL framework went through MPLS QA and an interesting problem was raised.

E

I'll go through the status, then I'll come back to what that problem. Was it completing the IPR poll and it's now in working reproduction polls? So please comment the RSV RFC 60 374 SFL draft again, as have been adopted as a as a working group draft. There are seven in Section seven. There are a number of methods of making measurements, which I really would like some review from the working group on and in particular, which ones to pursue to to completion.

E

If the answer is all of them, then we'll presume all of them, but it would be good to have sir a triage review to make sure that, with their all needed and, of course, feedback on anything else will be welcome. So really it's a request for please comment on the draft and and help the authors in terms of direction to steer it SFL control, and this still describes just the basic standalone control plane. It really needs to be integrated with the other MPLS control planes.

E

I've said this several times and much to my shame, haven't done anything so in easily integrated with LDP rsvp-te BGP and the signal routing control plane.

E

We also need to put some components in it to control it via Sdn and if anyone's interested in helping with the text their input would be welcome. So the issue that arose you.

B

Know, can you go back yes for a clarification point when you see BUP? Is it 3107 you're talking about yeah.

E

When we use yeah.

E

So the issue that arose during review was around ecmp and this thought to a my certainly to my attention and I should have spotted it earlier: a limitation in RFC, 67 90, which is the entropy label draft, in that the inclusion of an entropy label still allows an LSR to include other label in other labels in the entropy calculation. Although it does not prohibit the it does prohibit the use of SPLs from the calculation.

E

So the point is that I think I had kind of naively expected in my sort of dyslexic reading of this at the draft over the years that when you had any ale in there, that's how you did the the low balance. But there are implementations that take a more aggressive approach and load balance on that and other things as well. And this is a problem because any path subject to a CMP, the packet path may change the inclusion of the exclusion of s as SFL in a way that I hadn't been expecting.

E

So there was some discussion and I'm Matthew helped with this that we we made two sort of changes to the to the text during the review process, and the key text is that if this approach is adopting the intervening, MPLS Network must not load balance on any packet field other than the entropy label. Note that this is stricter than the text in our of in section 4.2 of RFC 67 90. So that is to say we would have to verify about one means or another that that condition was applied before it would be safe.

E

To use this, to be sure that the go and return paths were identical.

O

Cuddling the ferrule see Stuart. What do you think in Brock? These are the chances of that.

E

Happening well of you, knowing that, although nodes are conformed to this.

J

E

I guess, if you were doing a new deployment, then you would know- and you would probably have to take a judgment- call if you didn't know as to whether you cared or not I.

O

Guess the question is so: let me rephrase how important of a limitation it is to the to the overall document, because I think it might be hard to actually even know.

O

You know whether this.

E

Is the case presumably Arusha would know whether it did it, and so we could get the Richard advertised. What is whether it was strictly entropy label only or was maybe otherwise so.

O

In order to signal you're saying, there's an opt-in signaling, which needs to fill this requirement, yeah.

E

You probably have to announce that you fulfilled this in order to allow you to do that. Thank you.

E

So anyway, it works. If, if you've got Reuters on the path that only do entropy label, it has difficulty when you have reuters that decide to you the entropy label, as a with, together with a bunch of other information and I'm, not sure where any, whether it makes sense to have Reuters do both, but some of them do.

E

So we would need continuing from Carlos's point. We would need LSS to signal whether the whether they were using ili only or they were using other factors as well, and we would need to consider whether we need to look at the text in RFC, 67 90, since I'm sure there are cases where we would like PL to be the only director of of the CMP process and I think there was something, certainly some discussion by Eric along those lines through Rosen during the in this discussion along the lines of.

E

Why does it make sense to load balance on anything other than the Yale when the eel is in there? So.

A

That we ran with it, yes, I think, there's one question from.

A

So go ahead. Yes,.

Q

I just wanted to make famine or routers reporting whether or not they do these things, at least in some older Cisco equipment, whether you do such things or not, dependent on specific line cards and, in the very very worst case, it's dependent on both the ingress and egress lines. So so getting some things to report. This could be very difficult.

E

But Jewishly would be on.

G

E

Those riches also be unlikely. He.

Q

E

Know, balancing anyway.

Q

Yes, and that's precisely the other comment that I wanted to make, is that Stewart said I, don't know why he would load balancing something the other than the e Li and that's because that's the only thing they know, but if the stack depth that there operating on also includes the Li than the e Li Mei, you know, depending on the stacked up, D Li may or may not influence the path.

E

No, it's a bit more complicated occasion for blekic. My.

H

Understanding is that.

R

E

Reuters that Li.

Q

K

P

E

Apparently, that there exist, which is the and.

E

Other labels, sorry in most cases.

E

Ok, so that's what yeah behavior that people weren't expecting.

B

B

Any other questions, thank you, George. Thank you. So next up is Harris.

S

Hi, my name is Hari Sitaram, so presenting an update to this one version of this draft.

S

So we presented the first version of this draft last idea and just to give a quick recap. It basically was coupling our SOP t control plane with certain aspects of said, marauding, MPLS, forwarding, plane to quickly recap what it did basically allocated unique Poplarville for te link of an interface and then it utilized label stacking on the ingress. So the RSO pingers would actually stack a set of labels for each labels. Operation was to pop and forward at every transit link. We also introduced the concept of delegation of labels label stack in position.

S

We briefly alluded to the fact of how one could automatically delegate label stacks, but we didn't have time to add more text to it. So this draft, this update, actually adds a lot of text about delegation. We get into details of automatic delegation and explicit delegation. We also have section separately on constructing label stacks how a transit could construct able stacks how in just the construction of stacks, what would it produce?

S

We have new co-author directs and we also have George holo, both of them of contributed to the draft and based on the feedback in how authors, so with the delegation of label stack in position. So the idea is to manage the stack and position depth of ingress. So platforms may have limitations of how many labels it could push if the platform can push up all the labels that can carry the packet all the way to the egress.

S

Then much of what we are talking about today is in as relevant because well you can push all the labels so might as well push them in the even that you are unable to push. We have two different approaches. One is called automatic delegation and the other one is explicit delegation. So with automatic delegation you actually allow the transit routers to actually automatically select themselves as delegation hops with explicit delegation, you think Russ actually ingress or any part compute engine can actually specifically dictate certain transit hops.

S

The B delegation so take their example on the diagram, so D and G are like delegation hops, and they say it's okay, for instance. Let's assume for a moment that it can push. Only three labels can actually push a packet with 200, 300 and 350. Packer gives to D. D can push another three levels: Packard gets to.

S

Gigi can push another three labels packet and get to J, so the labels are D and G, which are marked red, are actually what are known as delegation labels, so des da d and g allocate these labels for the LSP.

S

So here, from a signalling extension point of view, the rro label sub-object that we have we've taken a bit out of it and asking for a delegation label bit and in the we also have a bit allocated and attributes flag TLV that if it can be carried either in the hop attributes or it can be carried in the LSP attributes. So if you were to carry it and LSP attribute, it would be automatic delegation, so the ingress is actually requesting automatic delegation throughout the LSP.

S

If you were to carry it in the hopper attributes, the ingress is basically saying I pick D and G I am going to specifically it's a part, hop attribute. So I'm specifically asking D and G to become delegation hops right, so two different flavors on how you do delegation automatic and explicit.

S

With respect to now that we have two different models of delegation, you could actually start the labels in different ways. So this basically is a slide talks about stacking only up to reaching a delegation half so the ingress only wants to reach to the first delegation of any transit delegation hop on until he wants to reach up to the next delegation up. So in this example, you have a person, you know puffin forward tunnel from a to J. Let's assume the delegation. Hops are DNG our ingress and at all the delegation hops.

S

The label stack includes the next delegation illegal. So, if you look at the stack, the red labels from a it only reaches up to D and D in fact takes the packet up to G. &Amp; G takes the packet up to J, so this is one way of stacking and we have another way of stacking. Where you can, the ingress can actually stack a packet all the way to the egress.

S

In this case, the ingress actually includes all the delegation labels in its own label stacks and the transit does not include any delegation label in it label stacks. So if you take the diagram a actually in this example, the difference between the previous one, if you look at a it only has three 5650 is at D here, 350 and 650 are at a and at B. Actually, 650 is already in the packet that's arriving from C, so this allows ingress to actually stack the packet all the way to the egress, of course.

S

Now that you have two different ways of stacking the labels you actually have to wait have a way to let the transits know. Please don't start the next delegation label in the stack so by default. If we only if you use the LSP attributes, we do automatic delegation this way to automatic delegation, the transit routers by default, going to take this approach. So it's the transit routers are going to each router only tries to reach up to its next delegation. If we want the ingress to actually do the complete stacking, then we can.

S

We include another bit in the label, LSP attributes bit where the ingress can actually signal telling the transits. Please don't include the next label in the delegate. Please don't include the next delegation label in your stack, because I'm gonna include all of them a name in dhingra's, so this different to different ways of stacking it.

S

So so how do the? How does in automatic delegation how to hop out at the hops first of all pick themselves up like how did they are to delegate themselves? So in this example, we can still assume from the earlier examples to stay, consistent, DNG or delegation hops. We introduce a per hop attribute.

S

Basically, we've introduced in the hop attributes, object, upper half attribute, says it's signal in the path message and ingress, basically populates it with maximum labels that it can actually push how much ever it can push and each successive hop actually decremented by 1. ah It also looks at its own limitations. It could look at its own outbound limitation.

S

It could look at its own inbound limitations and then it can decide to not increment it by decremented by 1, but it can actually set it to something else, and so if a node is reached, where you know the incoming ie TLD is actually one that node actually selects itself. As a delegation hop once it selects itself. As a delegation hop, it again follows what ingress used to do.

S

It sets the e TLD out to the maximum value going out of the box that you can push and pushes the packet out by the time the packet reaches the egress. All the legation hops have been picked already. So so that's how the delegation, the TLD mechanism, is used in the signaling to actually make this work for automatic delegation specifically for automatic delegation.

S

So yeah I mean maybe next steps we actually have have to add some text for node protection. We have a bunch of big set. We are asking for specifically six bits sitting in different attributes, so request early allocation of Ayana code points feasible and just time, based on the comments that we receive last time and the response we saw request working group adoption.

M

Hi, probably stupid, question or CPT doesn't stack labels, it swaps, labels per hop, but where does it come from? Maybe I'm missing something.

S

M

Are you going to stop labels in their CPT case.

S

How many labels can be there an orbiting planes, for example? One example could be a platform, could have a limited fifth size, so RS we could potentially still scale way past. You know he's not limited by the number of labels that are needed in the stack, so in this example for it, for instance, the forwarding plane actually is more decoupled from the from the control plane. The forwarding plane only has as many labels as a number of interfaces or neighbors on that box, as opposed to, if you had say, 30,000 LSP is across a node.

S

You would need 30,000 labels today. So with this, you only need number of labels asked for the number of interfaces or neighbors on that box, and so you could one other way is you can come or decouple it from the FIB size, capabilities of any box. The.

M

Real use case and today's heart we're.

S

Not certain a that's, not the primary motivation, though that's one of the motivations others are. Basically you can cut the amount of web state. You actually need all the chatter that hack occurs between a control plane to data plane. So thank you.

M

Can you last.

B

Slide, lastly, the Deborah.

B

Earlier location is only for working with documents, all right, I can't do it an individual.

T

And Deborah yeah it has to be a working group document, because one of the rules is, it should be somewhat stable right. You.

G

Shouldn't just be.

T

Requesting names and then with a high probability of completion, it's.

B

Very stable working group document- that's a yes.

T

B

So what my conclusion here has done, that earlier location is a little bit premature. In this case. We have to wait until we have a and accept adopted working group document and actually a fairly convinced that the document is stable enough to allocate to point four so they're appreciated you bring it up.

P

On on speaking of a quarter, so that's reason why you asked for option, because we would like to see early allocation of core points.

B

So I don't know if I got you correctly, yeah yeah, but you can't ask for earlier location for an individual draft.

P

But that's next finalist request working reproduction lurches.

B

Yeah I understand the.

B

The point, though, is that we have the high probability.

B

Hurdles to get over also, so we are not going to do it in the order we have done. The slides we are going to do.

B

You're going to updated, are beginning to go through the process for working group adoption and see where that leads, and if we have a stable working with document, damla can develop asian. That's what I what I want to make.

B

Okay, you done.

B

So, thank you. So now we should have had Cameron talking about ml DP and LDP gang models. Cameron is one of those that can't travel at the moment. I! Don't really know why so sue thang. You have lost like 10 minutes to represent the work going on on ldml, repeat yang,.

R

And Shifu, who can you provide it to the area? Dp young, an American young model? So since last ITF session, we had a few discussion meetings, but we have not got chance to submit the updated draft. So here's the discussion we had, we decided to too few changes. We decide among those authors and contributors. We want to simplify the policies specified in the model, so here we will move currently.

R

Existing in the pace model, the expressive, nam, control policy, move that into extended and also we will remove a few combinations about the policies so we'll take out the yeah police, say a global level and we'll keep the PF agile family policies, but we moved all ideas, families from the global and the from not appear global, and we also have said this document to doctors for review and we haven't contributed back there working on that.

R

We felt a few questions, so welcome to answer those questions and the work with the adopters, completely review and address any with your comments. So here are the pending items, so we need to complete there's a default values and we done completed the young doctors review to try Technic comments. Then we need to update the model to be an MVA compliant.

R

We when we need to use a state model, also the next step, we'll publish a new version of the document after that, when once we have lost rest of issues, will to ask for the last call. Another document here will be the I'm ready P.

R

So again we have an occurrence to update the document, but we also had a few discussions. We decided to change the configuration model section and also we can update notification. So here is the change we have already decided to. So we are here, is we can see, will expand it.

R

Okay, we expand the configure. The nearest thing is we used to be able to only handle one particular topic type, which is a RT. Now we will add other object, types like transit and the biter.

R

So actually in this change will also take care of the second bullet here so make the recipe, confusion and state model consistent, because on the state section we already have those the different object. Types handled and I'm also planned to change a notification again. Here we expand with the epoch types not only including previous existing LSP ID, but also covers a transit and by der some are open items.

R

You cause people value and updating to the nmda style, and then we'll publish a new version to trace those issues and then to the inductor renew, and then we can ask for Lascaux so the items we have.

B

B

Okay, so a comment on the gang doctor review. We had our wires crossed, so we recursively undock review a little bit late. We got or the gender of the Santos and male and we actually failed to recognize that they were there. What that male was so we didn't respond and that's I, don't know why we are cutting. People are too busy doing too many things really I. Think.

R

That came, the question came, is the late so problem we could be father died Thailand. So we didn't have the medium argue light. So we didn't have chance to discuss, but.

B

We are on track now and we will resolve those issues as soon as we can any other questions. Okay,.

B

R

You thank you. Okay,.

B

I think it's too late. We don't start a new presentation. Now we take everything on Friday who hasn't signed the blue sheets. Anyone who hasn't son, the blue sheets comes up here and sign the blue sheets.

B

Our than that see you on Friday at 10:30. The same room is on the agenda. Time is different. 10:13. Do you write.

A

And the second blue sheet is still missing, so where is it? Who has a second one.

B

Parang, can you look at the last row and shares if there is a blue sheet laying over there.

B