Internet Engineering Task Force 92, 27 Mar 2015

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF92-ISIS-20150327-0900

Description

ISIS meeting session at IETF92
2015/03/27 0900

A

Genius, part-time genius part.

B

B

They'll, give you a new one.

A

B

Should start it's.

A

For Jabbar scribes, what's the difference between a note taker in a chair, escribe they're, different.

C

Oh, this is like he's cool yeah.

A

We're closing, does it mean, does anyone want to be a Jeffers subscribe? You get you get a star.

B

A

Well, we have a sign up here, saying we should hand out stickers.

A

Okay, all right, I think we're ready to get started alright. So welcome to the ISS working group we're a little pressed for time today with the two and half hour, Friday.

B

A

That's a joke! Here's our note well,.

C

You have to read it.

A

By participating with the whole thing or just a blow.

A

Like participating with the ITF, you agree to follow the idea of processes. Do are aware that a contribution of your is something you write, say or discussing. The idea of context is covered by patents or patent applications. You need to disclose the fact that fact you understand that the meetings might be recorded, broadcast and public. The archived I believe this one is ok for further reading.

A

So this is our agenda today. Does anybody have any additions or subtractions.

A

Alright, so let's get on to the status of where we're at so we've had two new working group documents. Admin tag actually managed to rev once since the 00 listening since the last meeting at Honolulu. A lot of these everyone should be familiar with. We, the ones in bold, on these slides, are being presented today.

A

B

Comments on those.

A

We were a little bit the lacs on our key metric extensions, so that's completed last. Call we just need to push it through and extended sequence number is, should be completing last IETF last call soon I think we have some new IDs. So look for these on the you know list and maybe give some guidance.

A

All of the updated IDs are being presented today, and then we have some existing stuff. Some of these are sort of just you know, waiting to time out like the critical sub tlv and with it. Since you know, the distribution trees will push that over to multicast the source. Routing and auto comp are pretty much related to home nap right now, they're not restricted to home, that's used, but while Otto café is but sourced esperare, and these are the ones that are about to fall off.

A

If you go to data tracker, anti-bias ISS shows up it's worth reading, I guess it was a reference to its. The security analysis for ice is primarily uses. I think where the extended sequence number work came from.

B

A

The yang was first.

A

Nice already nice, one.

D

Okay, so I'm stefan from orange, I will present you. The updates regarding the is, is yang model next slide. So in this new version we introduced via is is not tagging, so we are able to define with. Envy is a yes instance, a list of not tags that could be applied to a taurus nod, and it's also applicable to be in which the pragya use case next slide. We are also modifying the authentication so based on vicki chain model from AC.

D

We are introducing so the ability to use a keychain reference so toward via keychain model, and also in Norder, to simplify via oughtn't equation between all the routing protocols. We are currently using some groupings coming from vikki chain to not redefining all the autentica possible, I authentication algorithm, so it's inherited from keychain and as AC already mentioned during the LT gwg. If an implementation is not supporting, for example, che one for specific routing protocol, it is expected that the back end will reject the configuration next slide.

D

We are also introducing first reward I PFS throughout configuration. There was a lot of debate on it, so what we are proposing years, so it's a support of LFA on the matter, if a remoter, if a is seen as a child of LFA. So if you want to activate the motel fail, you will have to activate RFA first and the activation is only done / interface.

D

So next slide. So you will see how this is a global configuration.

D

So what we done globally is just creating containers, but these containers are empty and just where for augment, so we can think about. For example, let's vendor agreement this container using, for example, LFA policies, so the global configuration is really not to activate LFA so next slide. So next slide is the interface configuration for fast throughout and you have the ability, through via enabled key word and the first word LFA to say: okay I want to activate LFA or remotely facing under via the interface next slide.

D

There was also some adding regarding spring. We are currently under discussion about what should be kept in. This is, is yang model and what should be in the spring segment routing yang model. So currently the situation is not clear, so there are some redundant information between both model, but we hope to clarify this quite quite soon.

D

There was still the discussion between the vrf, centric and protocol sandwich, which was the description of the week and to the statues. Now it seems that very clear consensus to go to add to the grf, so routing instance centric model. So where was a discussion on net word?

D

Where was a discussion on etsy gwg and sounds that the consensus is well, so we will update via model accordingly next time, so the next step internal configuration, we pretty things that we are done, but we are clearly welcoming comments if there are some critical pieces that we missed, we are starting the review of the operational state, so we are currently trying to align both configuration, trees and operational states trees, which was not the case before we are also progressing them of alignment with via ospf, mother, sorry, mother.

D

So now we are clearing your feedback because we are working in a design team, but for sure we need your feedback about. Do we miss something? Is it fine for everybody? So please comment what what we have them. Thank you.

A

B

E

Good morning, everyone I mean I love short cash, and this is a presentation on the recent is, is pcr draft. We have a new outdoor chris bars welcome, and thank you very much for your help in this draft and be in this subject. Actually, in other dots next slide, please- and this is just a quick recap- what it is. It is, in fact, an extension to SB be short as much bridging the goal is to move beyond the shortest paths, capabilities provided by SPB by means of having a so-called explicit trees.

E

This figure illustrates this to type of trees. The blue one is the shortest path and for exceptions, traffic steering. In some cases we may need extra seat trees. This is for small networks and for rare usage.

E

One important thing is that there is no protocol change included in this work at all. It is just new theories. Actually next slide, please the status it is. We are developing the working in two SDOs here and the United Plato two dot, one, the corresponding dot. One document is a QC, a dot on qca, which is entering the sponsor ballot, which is the highest level ballotine in I. Typically, we expect to have two cycles at this level and if everything goes optimal, the final approval could be in July.

E

Qca refers has no to normative references, 22 ITF, dots, working Rudolph's, the one of them is here, the is is the metric extensions and the other one is the MRT FL algorithm and for final approval. These I mean qca cannot be approved without literal referring ongoing work, so I chilly, you say, can be only finished and they become RFC's.

E

Also they need ice is tre good points, and this is one of the main goals of this craft I'm. Presenting here.

E

Somewhere unknown on this draft, this is in sync, with the most recent QC a draft and actually has moved beyond the next slide. Please, and we will see the details this is, is pcr draft has been evolving as QC evolved at the last ITF, I presented the 00 draft, and this is 0-2. There was an intermediate draft which was for bringing it sink. The idea here with the qca updates this one is a includes. The updates of the QC draft 1.4 and actually have moved beyond the first bullet summarizes these updates.

E

That happened in the QC, a document that we wanted to simplify and find the way to simplify the topology sub theory, which is the main main subtly we are defining at the price of burning, the so-called explicit algorithm IDs. So that's why the first to change is the ECT algorithm for the loose tree and the loose reset has been changed.

E

The topology sub theory has been changed and there is also change a new new flag in the bandwidth assignment, sub div and the new sub div, the timestamp sub div, which is for sort of conflict resolution. If something goes wrong and as I said, decided, I say, as pcrid has moved beyond the QC, a draft I've got comments from a number of people which are included in this document already. So we wanted to bring the draft for this meeting as in as good shape as possible.

E

There are updates for clarifications like how do we achieve the symmetric parts? This was not made clear in the former versions that, in case of different league matrix in the two directions of the link, we use the maximum one and we use the tiebreaking used by SPB for a cloaca spot, and there were also further updates on the Amati aspects for the clarification, and there is one new thing.

E

The previous documents only described the case of an so one of the mechanisms is when the topology sub div is used to describe the so-called Cadoc, the generalized almost directed acyclic graph and in a simplest case it. It consists of a single block, but for instance, if the physical topology has cut links, then it is split into multiple blocks, and this new revision of the draft covers the multi-block case as well. Next like this, so this is a summary on the on the sub t alvey's. As I said, this is extension to spb.

E

We reuse spb theories as much as possible. These are the top funds and summary on the on the new tier is being specified by this draft, and the next slide is the last one. Just a quick summary, I I, pretty much said what, in the top in the slide that this is extension to SP, be reusing theories without put a code change and the bottom is is quite important. Qca is under final review and we would like to move forward with this is spc our draft here.

E

It would be good to have a working group adoption and really allocation for the code points of the sub theories.

C

Okay, um so he would be a bit the road map on progress in this further, so this is not really a classical IDF document, so I would like to start the working group. Adoption call right after this meeting will give it some period the usual two weeks, and then I would like janish and friends to initiate the earlier location procedure. Such they get their code points which are going to be valid for at least one year and throughout a proc and the Oklahoma. We progress the document further and well send it to IHG for publication.

C

So if you have any concerns about the way progressing things that way, please speak up on the list, of course, as well.

E

Okay, thank you.

A

Tony here, which one which one do you want to do first.

C

A

Auto be of our ID or beer.

B

A

A

F

Morning, everyone I'm 20p, are two questions. First, who read the update on the beer eyes is the 02 that was published in meantime, okay, huge response who r at the auto allocation procedure, oh and ready? Okay. ah Next one please. So this is basically what happened between 2001 and 02. To support beer via is is so basically was all aligned with architecture worked at ice and a couple of other people arrogant needed so subdomains have been introduced. It's the biggest one subdomains our map to topology and other authors must agree on the mapping.

F

So the supplement belongs to topology. Subdomain has a single encapsulation I'm, still poking d'architecture people, to spell it out specifically on because theoretically, we could do in capsulation. Translation on beer within a subdomain r sub domain has a single tree or forwarding strategy, so that is implicit in the architecture on but I'm being poked and poked by more and more people, whether we could run anything else but SPF. So I wanted to keep the draft open enough.

F

So it's not excluded and what was LEDs that any rotten all advertising enough labels for all the BF ers will be ignored in the computation. That's something I basically wait for implementation and deployment experience. Whether people want that or not good. So basically, the new stuff is that the tv's has been reordered on the market topology when I have to sub domain, and within that we have the encapsulation enough labels and we have an optional sub tlv to indicate the tree type and when it's omitted it means SPF implicitly today.

F

So it's spelled out, but it can show up later and the routers that run SPF today and see a tlv. They don't understand for the tree will know that they cannot participate next. One, please um all right, and there is I introduce something which is an experimental idea.

F

um So one of the things that I have some discussions with people about is that when you get enough deployment in terms of size and we need a lot of PVR bits, whether an automatic assignment makes sense- and we were not doing that until these, but the idea would be that we can put on a tlv a minimum maximum version number of what protocol version de still V represents and with that without introducing a completely new TLD.

F

We can progress the tlv and indicate two previous routers, and very precisely in which rate range they have to work version wise to to be able to interpret this tlv. The biggest advantage is that routers to run a newer version could rather steal, via the newer version, indicating a newer version with new information which all routers do not understand and fall back once they have enough information, publish it at the lower version of the tlv, allowing old all the routers to participate.

F

um Okay next one please so the first one is the sub tlv beer sub tlv. So we have tagged length now. The version with a compatibility bit then maybe mean max version. No I need discussion on the stuff sub domain. Id is under it and the bfr ID, which can be empty under that we have the encapsulation. That's the MPLS encapsulation, which has a label range size. What is a bit mask length because they are hit actually change in the sense the sub domain can support multiple business line.

F

So it's a list and the very right squeeze in is what tree type on this sub domain supports when it's missing. It means SPF, but in the future, when this tlv, well-known sub sub sub tlv will show up, routers will understand that they cannot participate because it is not SPF. So if people start to build, Steiner I also basically indicated that there put enough opaque data to actually all agree. You know what kind of Steiner they're running, so that's pretty much the update. um Let's, let me take the questions now towards that.

F

If there are any and then I quickly, basically just run through the idea of the auto election, because if people who didn't read the draft, it's pointless to lie going into more depth, questions.

G

Ac lindam, a cisco other. The only thing I'm wondering about is the terminology. You know the Eggman routing draft uses algorithm and this uses treat Type O I.

F

Was the world ok, it.

G

For four somewhat the same thing it may look, maybe they should be aligned. Okay,.

F

Yeah point taken: I will align this stuff all right, I mean there's no particular preference.

C

And it's great let jennifer! I was actually first time that I saw using versioning inside subkey. Elvis I was under the impression that the versioning tool is actually a tlv. Your security code point. Yes, any any reasoning why you ended up to that solution? Yes,.

F

Imagine that I want someone to implement the absolute minimum set and then I want something which does a smart election after the election ran. I can show people the old style tlv, while the election is running, I want to tell the other routers back off. It looks like the format you understand, but I have sub T all these inside. If you don't understand them, you screw up so I use exactly the same format, but I bump up the version.

F

201 I say you, don't speak 01, don't look at the TLB the moment the election ran and I have all the information that makes sense in 00, I just drop the version of the TLD 200 and all the old style routers participate happily. So that allows to note every time you put a sub tlv, which may other people old-style people screw up. I don't have a new TLD type as.

C

I see you right, you're using it for atomic nessa cement, yeah.

F

Okay, yes, so you know it's an experimental idea, but since we have this Auto election and I don't want to force, people today to understand, implement, do anything about it. I want the future hook to people operating on the same key of E format being able to participate it after the smarter guys figured out their stuff. I mean it is it's exactly the same ideas shared libraries of lineups right, you load the library, and you tells you I'm at three, but I can really also talk to to you, but please I cannot talk what.

A

So I mean that my first thought was: wouldn't you just use to TL VZ and then back up to the old man after.

F

That's the other possibility. Yes, so basically you have even maybe the same format, but you change the tlv type and you talk to steal these and then you publish and worst case you end up with two versions right, like all the version with all the routers new version for the newer routers. Yes, so that's kind of the simple naive way to do it right, which is more implementation, more TL, these more data published so okay, Chris.

G

So it seems like an interesting idea, as making an experimental seems reasonable, but at actually sort of sharing the the data that you to my concern about. It is that you actually sort of brought in the testing space of for compatibility testing that you you actually you know, even when you don't have versioning. Often a lot of problems are discovered during interrupt testing, even with things that are supposedly the identical version, and now you've basically created a matrix of of cases. You know version 3, &, 4, &, 4, & 5 that need to work together.

G

F

Okay, all right, the argument I would make it is absolutely equivalent to having to tlv types. The TLD time 135 will indicate version. 11, 3620, Kate version, 2, I.

A

It's not it's not quite that vary because you, because you're sort of taking over and putting a value- and it can be a lot of different time. Yes,.

F

Okay, if you go wild and do it 124 and you want to do all the permutations yeah- that's why I only gave two bits to a version: I think it one. But but then we.

B

A

Notice one that's.

F

About it, sorry we're out of runs runway. Les mots.

H

Ginsberg, who are you you owe me say it again, he's.

F

A really nice guy.

H

Thank you so two things, and this is something that you know we. We are still kind of actively discussion discussing even among the the co-authors you're, essentially making the version number a new type that has to be controlled. That was where we farm.

B

Immigration, yes,.

F

Allegation electrical noise, the.

H

Other thing that concerns me is how you define the transition strategies as old or new routers appear and disappear in the network and you're able to move forward or backwards on your version. Numbers I think this is always problematic. Okay,.

F

Fair observation and.

H

If we're going to do this, I think we need to do a better job of specifying how that happens. Yeah.

F

Fully agreed so I mean this is something yes, it's a completely new idea, I'm throwing it at the wall. I won't actually people to go. You know beat me up and see how bad it gets right and also the idea I socialized they. Maybe they for they find the beat still liking the that don't like, and we basically have the trade-offs and otherwise it's there's always the fallback like. Okay, let's get the new tlv time number. Okay, so yeah the discussion.

F

You know, I want to pick up the discussion right and I want to show a feature that will benefit from it. So people understand that it opens possibility to do those things on the protocol. We couldn't do currently right today. We just basically say well all this up till these, you don't understand them just keep on going, and if someone introduced the one that breaks things, he better explains very well and then ultimately, somehow at certain pony time cuts over to a new tlv tied. Okay.

B

Right enough, okay,.

F

So then, I think next presentation, I I very coarsely, explained the idea of this author election. How, although that would run with it and.

F

So, where is not that familiar with beer? The very short summary the route the receiving routers need a bit allocated in a bit mask, and the idea currently is that, as long as the deployments are small enough and stable enough, it will be a manual configuration if there's a collision. There's notification people take care of it or may a central controller.

F

This proposes something akin to a designated router right, so people have a central authority, they can get beats from or it's more maybe a better analogy is a dhcp server all right next one please! ah So now that we have some domain at apologies, okay and those number of those be f. Ers can go pretty big because we have two sets. So basically, we are not constrained by the bit mask length up to a reasonable size. We may be talking, you know low number of thousands for sure of receiver beads.

F

um So you manual configuration certain point in time. If this technology successful will become invisible right, otherwise, we'll be all still assigning. Ip addresses by hand to I worked like a charm or ftp et Cie host, and so it that it always ends up on a centralized and distributed solution. The centralized solution is probably something akin to SDN controller right, which is just fine single point of failure. If you replicate them, you have split brain scenario, sets its own limitations and a distribute solution is far more resilient.

F

I mean that's why we were here right working on protocols like this, of course, always additional complexity. Next, one, please all right. So the whole thing is the main thing. It is completely backwards compatible to the drafts that to support beer / is, is so people can do they implement what what the draft eyes eyes beer says and when this thing is slipped into the network that won't realize the change there will be all-star routers, not supporting anything new and they'll be participating. Just fine and their assignments will be fully honored.

F

Okay, the trick behind the whole thing is pretty much that there is purse up to Maine. Whatever night we spread out a designated route, Rebecca designated routers lots of that stuff is stolen from John, because it's so stable and this thing will act as central authority. So the routers can signal on the tlv saying give me an assignment and that's the one I would prefer and what they get back from the designated router will be basically bit assignment for it or an error or whatever all kind of conditions.

F

Excellent, please um I, won't even go how many sub T of Easter things and titles, but the dr basically puts out something which is the assignments in it and on the normal info. Tlv people set a bit to signal, give me an assignment and or they say I have an administrative asst assignment the version number as long as the assignment is going on, the router will say: I'm at version number one. If you don't understand version number one. Please don't look. Look at my advertisement. The moment d assignment is stable.

F

They put the assignment into the four minutes, I'm with versions year, which means all routers can perfectly well work with. It I think this last one next one. Please I think it's it yet. Alright! So that's coarsely the idea. If there are any comments, don't towards it, you have one more.

G

Ac lindam Cisco, ah the only thing I'm thinking is: if you're setting up subdomains you're going to have to decide what routers participate in every sub domain and when you're doing that, when you're doing that, you could just as well give it an ID because I don't think I think this is only half of the setting up a sub domain in anything unless subdomains.

G

Unless you envision a deployment scenario where subdomains come and go dynamically.

F

It will depend how people deploy services of a sub domains, but basically, what this thing takes is one configuration step out. So you say, I'm running a certain service now came to me. I would provide d for subdomain met. Mappings I run a VPN service I end up on this sub domain by default, okay, which may or may not happen, but when you mapped onto a sub domain, that will be the only configuration step today. You have to provide a non colliding, BFF be fer, so the bit for the sub domain.

F

So that says this configuration step, no, no deployments the deployment, simplicity versus protocol complexity, the usual traders I mean this is nothing. You know that that you know any kind of magic, wand or whatsoever.

B

F

I'm done thanks for your attention. I hope is with the team I.

A

Took some notes.

B

A

H

A

Night I said eject we're just waiting for it to eject.

B

B

A

What what order would you like we registered.

H

Prefix attributes less as.

A

We can see the other two I, don't care, how thing? Let's go on reverse order.

H

Ok, this is this is going to go pretty fast because all of these drafts have been presented before so these are just quick updates, spfd discriminators. Next there was one change which was: there was a conflict in the suggested code point with one that had been suggested in the second rounding draft, so I've resolved that conflict. That's the only change since Honolulu. As far as we're concerned, the draft is complete and we're going to ask for last call once the base spfd draft goes to last call in the BFD working group.

H

We don't have to do the rest, yeah.

A

H

I left this in this has been presented before any questions about the spfd draft.

B

H

H

Okay, Rob preference draft next slide again. This is complete. There have been no comments since Honolulu.

H

We're delaying last call, because there is a dependency on the prefix attributes draft, which is the next thing I'm going to talk about so again, that's all I have to say about this. Unless anybody has any comments about raw preference.

H

Okay, prefix attributes next slide. Since V 0, there's been some editorial clean up. The biggest issue was there were two proposals. There was a competing proposal to as far as advertising multiple node addresses.

H

There has been a discussion between the authors of both drafts and we've reached agreement that we're going to use the solution in the prefix attributes. Draft next slide, so I put this in red because there are two documents that are dependent on the progress of this document. One is the reference draft that I talked about before, and the other is the segment routing draft, because some of the bits that were defining for prefect attributes are also defined in. They were originally defined in the segment routing draft.

H

We move them here because they had applications outside of segment routing. So the plan is to eventually deprecated the do the duplicated bits in the segment routing draft, but clearly we cannot do that until people can actually go ahead and implement the prefix attributes trap, which means we want to become a working group item. We want to get early allocation code points as quickly as possible, so I would really like to progress this as quickly as possible, and that's all I have any questions comments.

A

Is there anyone in the room right now that objects to this becoming a working group document hum raise your hands and thing? Okay, I! Think we'll take it to the list and ask for adoption.

A

Make sure you stay in the pink box.

I

Hello, everybody, my name is Uma. Today, I am going to present encapsulation capabilities in my assess next like this, so this is not specifically for spring. Personally, this is envisioned for spring incremental deployments. We have fused cases my interest in the strap first few use cases where small routers, where we don't have mpls, but we want to do traffic engineering in mobile cell stations, where only you have is routing capabilities, but on the edge notes you can put a spring and a controller can put the paths, explicit paths and divert the traffic.

I

So that was my interest in this trap so later on, there are a couple of other use cases where incremental deployment is needed and like beer, where you don't support, mpls and just pure IP, but on edge notes. You have mpls capability with the segment routing with beer, so in those kind of situations advertising the d capsulation capabilities of that I Gress node will be very useful. So this draft is about advertising.

I

Those capabilities in high gps, and especially in this case, is s so the new use case was remote LF a case in pure IP networks, where PQ node wanted to ties. It's d, capsulation capabilities to all the piece pace notes this is not documented in the track. This is a still in discussion.

I

Bruno suggested this, and we will work on this next right-click. So, as I said, this enables ingress ip-based tunnel to know I'm capsulation type to use, and this has a huge advantage. I in operations basically need not config any time any type of stuff. You know on each node, like admin tax, that's kind of things expended next active.

I

So what we are asking is a sub tlv in the 497 one which types which specifies the encapsulation capability are. Basically it's a V capsulation capability to be specific. After I Gress node and the associated parameters for that the gap / incapability, the current we only in the draft current version of the tab. We only support only mpls encapsulation, we didn't put the IP encapsulation set, so the list will be updated next, like this.

I

So we thank Bruno for his thorough comments and new use cases will work on this to be done. Adding a remote LF, a use case, encapsulation types review and subdial ways and encapsulation subtly type to indicate the parameters at and encapsulation types for certain encapsulation. That's like IP address GRE key and we look for more inputs and comments.

J

I

J

A basenji cisco systems so woman- I talked about this and I had a couple of 2 comments. First of all, this is a general problem that is not really related or specific to segment routing. It is more of a how do I bridge to mpls Islands over an IP island, and it is not specific to a remote LFA or how to decide the PQ note again. It's a general problem. How do I choose the peak you note?

J

The third comment is that this is about a 20 year old, maybe 15, year old problem, since the creation of mpls, when it was called back switching- and I am sure people have solved this problem- how to a bridge to mpls islands over an IP island without the need to extend the routing protocol and one way as I suggested to do it is to use admin tags.

J

There are other node admin, tags or prefix admin, tags or configuration, in other words, I, don't see a need for such extension for a problem that existed 15 20 years ago and nobody has objected to its existing, even mention its existence. Thank you. A.

K

Shaka from poway another culture is dropped and, in fact Upsy 5512 healthified in approach for dynamically advertising, encapsulation capability, where PGP now this dropped it to extend IGP I sighs to do the same thing but listen L for this is in some scenario. Pgp is not you'd, so it's straightforward to extend I sighs to do that. Work for that are in the incremental deployment of the beer.

K

There are no biggie p na bola. I sighs use instant, so drop it to support a snarl. Thank you.

H

Unless Ginsburg, um just in a vein of brainstorming, Ockman, has suggested that prefix admin tags could be used to do this. One potential advantage of using prefix admin tags, which I think is something dear to your heart by the way Mumma, is that it would also implicitly tie the endpoint address that you want to use to the tunnel type. It's just a thought so.

I

I'll answer both the questions. First, for your question: we can use admin tax, but it requires changes in the code to also said typically to the admin tax use, but basically for redistribution. You stick a tag on the prefix and that prefix is the attribute for your stuffer, for redistribution between the levels are leaking that will be used. Thats all is there for today and if you want to put associated parameters, probably if we were tunnel pipe, it requires particular IP address. You can use it, but it cannot be generic and it requires code change.

I

It's not straightforward. Actually, this is a straightforward approach. Any way that also my code changes required. This is a more straightforward approach, more generic approach, I.

H

Mean we can take this to the list, but for the record I. Don't.

I

H

It's so difficult, so.

I

You cannot put a GRE key, for example, an admin tag actually.

K

Shampoo from poway, in fact, I have noticed there are attempt to use a different way to advertise. No capability, that is to use the prefix by Stan, keowee I, think both the size and the ospf should the sails away. Consider whether we need two different approaches to do. The same work think.

J

The point that I'm trying to make is that we are trying to solve a 15 year old problem that nobody has complained about. I am not sure why it popped in the past few months, so.

I

So I'm a question: actually we discussed amadon this very elaborately, so, as she mentioned, non-bjp environment is useful. For me, the signal trotting use cases when explicit path is meant is crafted by a controller and put it at the ingress node. When a node encounters the neighbor is not spring or beer capable. We don't know what Newt whatever. What we need to do. Bgb is not there actually. So immediately. If you know the tunnel endpoint and its capability is with you at dingus node, it's quite useful and automatic. Actually.

C

um I would like to repeat this is honest. I would like to repeat the comment that I gave in the ospf working group. It looks to me that this is the formal equivalent of the tunnel safi RFC 5512, so I do not really see that this is a unsolved fifteen-year-old problem. It's actually a solved problem and my.

B

C

Is that human tries to put the tunnel encapsulation semantics now down to the igp level? Maybe.

J

I wasn't clear: this is ahmed basanti again the problem of I lending to mpls Islands over an IP island in an intra domain and an igp domain existed 15 years ago, and nobody has complained about what you're referring to is BGP and I agree bgp was solved, but it's a different problem.

K

I wonder whether you have noticed that there are section in the pair architecture drop, which talks about incremental deployment of a beer technology in that section. Is that if the next hop to the final be fer is a noun VFR that the upper could tunnel? The air package directly to the VIP are either where r SP or IP based hah knows if I PETA know is used in that case, to use such a to use DGP.

K

Do you suggest to establish bgp sessions between all nodes within the beer? Now the domain.

F

20 p, so whether a router supports beer can be very easily derived from the fact what it announced a beer info, sub, tlv right. That indicates to you. The reading you know supports team caps as well.

I

There is no support in segment routing like that.

J

All this, this is ahmed russian team one more time, but the point. The point is that we are extending the protocol with new sub T LVS, which is not a straight for subtly. It will require configuration, so the same configuration effort is needed or actually less configuration. Effort is needed by using admin tags.

J

You can add, you can achieve the same exact result with admin tags you go to the routers, configure them for admin, tag, number ten use, GRE admin, tech number, eleven used, l2tp so on and on the end points or the D caps rating points, 10 for GRE 11 for l2tp, same thing and I've in taxes been around a coma for how long so, at the end, the amount of code. If we need code for this, we definitely need code for admin tag that has been ten years ago. We definitely need more code here, I'm.

C

At some, there is actually people who are doing T controllers these days, who keep telling me it call, can be solved with static routing, so you don't really need then imma cropping protocols at all, so so so obviously well tags generic tax, no tax, prefix tags are sort of last resort exits for solving a problem. The problem that gets a bit lost is that they are context three, so those texts do not really have any semantics and I. Think for since this is data plane implications are they would like to carry on those semantics.

A

All right well, this is an interesting discussion that maybe we'll publish on the list as well. Everyone's opinions out there, but I think less. You were suggesting something similar with tags.

A

I

Right thanks thanks Kristi thanks thanks. Thank you.

L

Good morning everybody and shad da will be talking about the link overload feature in ISS. A similar feature was draft was described in ospf working group as well. So this is a same feature being protocol extensions for the same feature in a sense.

L

So we have the node overload feature wearing. You can set a note to overload before picking it down for the maintenance, so here we are trying to extend the same functionality to the overlay networks, so this diagram shows a typical overlay Network, where in the seas are connected over or in overlay, so the pier devices are not visible to the sea devices. So when the t's go for maintenance before taking it out for a maintenance, you need to divert the traffic from the link to the other link, which is an alternate link.

L

So if p1 is going down for maintenance, traffic from cee blue one has to get diverted yrp e to the link, which is why a PE 2 and similarly the traffic from see you too I need to be get diverted. Why a PE for so, whereas it's easy to avoid the traffic from see a blue one, it's difficult to do the same on the from the remote site. So generally you you need to do manual configuration on all the remote ends to say a to divert the traffic to the other link. Next time. Please.

L

So we just described in the problem that you know manual configuration is difficult.

L

And- and the next slide please, okay- this is a this- is a similar problem described for the overlay broadcast networks. So this is an overlay network over IRB interfaces where in the sea red devices here they appear to be in a broadcast network. So here also the problem is same. If p1 goes for maintenance, you probably want to divert the traffic before the p1 is taken out for maintenance. This is useful.

L

If you know, operators want to give it as a service to the customer that they will make sure that traffic is diverted before the P is taken for maintenance makes like things. The motivation is ease of maintenance. Automated upgradation and minimize traffic large make circles.

L

So what you're suggesting is to carry the link overload built in the link attributes tlv sub tlv, which is RFC 5209, so just define a bit in the link overload. Sorry, a link overload bit in the link attribute your game, carry it along with the.

L

It's is extended, reach tlv, which is tlv number 22, so the receiver it identifies the connecting link increases the metric to max metric. So this metric needs to be a usable metric, so that in case there is no alternate paths. Traffic still continues to flow on this and then refluxed, I, LSB and traffic is directed away from the old, ordered link and then link is ready for maintenance.

L

It's like please. Similarly, for the broadcast links the D is for the link just removes D is for the link updates, the match metric for the originating node and then reflux, the Sudan or LSP. That and the traffic is diverted away from that link and then the link can be taken off for maintenance.

L

So this is the RFC 5209 define link, attribute, stop tlv, so the which is defining a bit in the flags.

C

Questions comments.

H

Less kids Berg, so a couple of questions one. Why would you put this bit into the LSP advertisements rather than the hello, given that the only person are the only know that's of going to do anything about this? Is your neighbor yeah.

L

That's true only the neighbor would act on this, but since we can reuse the existing theories and it's just a bit I think it's not very difficult. If we define the procedure that only the remote end of the link is going to look at it, so it's the same thing. So the only the neighbor is going to act on it. Well.

H

One of the consequences of putting it into tlv 22 is, if you have parallel links to the same neighbor, you're, now forced to advertise and they're they're equal cost, you're not forced to advertise all of them, which, in some scenarios is not necessary.

L

Interface address and the neighbors.

L

H

That's quite true, but if I have multiple links to the same neighbor and they're equal cost, unless I'm doing traffic engineering there's no requirement to advertise both of the neighbors, because when the point of view building the SBT just saying I have a neighbor to you know to be with a cost. 10 is sufficient. So the me that's another reason why it would be much better if you put this in the hope that the follow-on to this is I, don't know if you're familiar with the reverse metric draft.

L

H

So I mean did, was it just that you weren't aware of it, or did you talk to those authors? It's.

B

L

H

Fair enough, but a jeweler Reese erected well, in my opinion and and you know, I was never a fan of the reverse metric idea.

A

But and now you're gonna play.

B

No, no I'm, not fighting for what I'm saying I'm.

H

Saying if we're going to do this, I believe the solution to find in the reverse metric draft is the correct solution and oh by the way, maybe those authors deserve some credit. So maybe you can collaborate, but you know that's some. My problem um I just really think. If we're going to do this, we should definitely should look at the reverse, symmetric solution and I think we should use it. Oh.

I

Ma from erickson, first I have one small comment on the draft: just a clarification, basically in ISS version of the Star. A draft you still talk about, dr and bdr probably need to correct that language.

B

I

That's one comment and, regarding the reverse, metric, the power Traverse metric has a lot of concern for a lot of people was you are changing the metric of the other side on the fly, so people may not like it operational account with a lot of discussion in the mail. English probably might want to see that. So this is another approach as.

I

Pushpa sheesh juniper question too less I kind of agree. The reverse metric does solve this problem and I also agree to your comment that this should drop should be only on the hello.

I

You need not be on the LSP, but I have a question to you back again. Forgive my ignorance. Is there a reverse metric draft in ospf.

L

A

Think we're not Derby yeah.

I

That's what I wanted to actually say so, instead of resurrecting the reverse metric, we can actually have these two drops already proposed across the 22 working groups are.

B

I mean, let's other alternative of adding one in.

I

Ospf for the reverse metric.

G

Ac lindum, your colleague, has a very similar draft in ospf, where the encoding could be. We have the two-part metric, which is for certain types of so there is encoding for more than one metric. There is not per se a reverse metric.

H

Ac, the the other comment I would make. Is it's a little more difficult in ospf because you don't have a metric in the network LSA so for the case of lands? We have to find some way to solve that.

C

So um just my read of the room, how much of you are the opinion that we should have a comparable? Well, if not identical mechanism and the boeuf I GPS.

A

Yeah he's leaving.

B

A

B

A

I think he's really.

C

Trying just to get some sense.

B

G

Yeah, you are AC lindam. We agreed to talk about whether this is really needed or not in the ospf working group. Whether this is this scenario, uh why you need to do this, not as opposed to how you need to do it and that's the first I think. That's the first question: I mean ie. Ok,.

C

I sent to verizon guys into you away.

B

A

B

A

Going for is that we probably shouldn't have two different methods and.

C

A

It should be the same in both I GPS.

A

H

Skin perk I I mean what we've consistently done with these kinds of things in the past is we provide equivalent functionality, but the actual mechanisms in terms of encodings are or what's best for each protocol I, don't see why we should deviate from that. If it's and I think in this case, certainly for is, is if we're going to do this. The hello is the appropriate place to put this information.

C

Okay, fair enough. Thank.

L

A

M

Hello, everyone, and then my name is a wine merchant from Huawei. So today, I'm going to talk about is, as a party in perinton totally cool t-d-z, so give the introduction extension to ISS and then HP updates for ttc an adjacency whiskey busy next, so let's the level becomes bigger and bigger, and then we need to split up the ladder book into more areas.

M

So when we do this bleeding and then we have some issues such as the surface will be interrupted, because the note will link with old area we're going down and then those notes and linked with new area that come up so during this process, the surface that will be interrupted so also for splitting laterally into more areas. We know this is a very big madoka, a cracka change.

M

So this is a very, very challenging task, in addition, is very time-consuming when doing the speak, so whether mo may be very complex for another in the art piece. So tdz will resolve these issues in Ostia we have implementations and we have showed that ospf tdz can resolve these three issues so with ospf tdz. So we show that we can transfer a part of life work smoothly to tdz without any surface interruption.

M

Also, we do the transport transformation. The operation is a very simple.

M

So also the ospf DVD is a walking boot document right now, so, like slice, so tdz.

M

So tv3 the virtualization technology is a virtualized, a part of the network as the ages of this part for the connected. So this part of ladakh we call a TDC topology at random, so outside of this tdz, any no doubts MTDC is not aware of key disease, so it will only see the ages fully connected. We'll see this virtualized a topology, so those are linked inside the TV G knows have a link well now to see not not not be at advertise through outside of ttc.

M

So like slice next one so to support, I assessed EDG, we just need one POV called the td's et al vs, so TVs et al v contains some flags and PDC ID and then a cup of software of ease, so four flags. So this is just copying from or spear, so we have weekly find five flags, so one flag is that the flag is the edge of a trot of TDD.

M

So this flag indicate this router is a edge router of TDD, and then these four frankly is just for our migration, a part of ladakh through tdz and under, and then some of crack is used for grew back at EDD to the lomo metal, for example, for migration, a part of life to TDC. We just used two flags one flag is that PFLAG for distribution of TDD information and then an as-applied is for migration.

M

That part of Blackrock include EDD so for sabit lv. So normally we have one start. A list which is T DT is labels safavieh, so next slice so for TDC is labor support. Lv. It just consists of a lambo of PVC labels, so like slice.

M

So here we just give those how we use LSP to virtualize the TSST disease, so for each internal router inside the t-d-z. So we just ate t-d-z tlv, so this TV detail v with a flag equal to 0, which indicates that this is a internal TD router and also with a TV GID, for example, with CDs ID, 1 100. So that's so justly so for internal router is easy. Just one!

M

Kov TVD healthy with e equals 0 and the PID so for any router for 88 year old of TDC, so we just have is LLP with a pdz tlv. So this TV GTA V, where have a bit equal to 1, which indicates that I'm at EDC edge, router and also it will have a TDD, is and stop at lv. So this is a betrayal v will contain the t-d-z labels, for example.

M

Here this is a page out of TDC, so the suburb POV, the subtlety of a disturbance near we will contain the label, p, 1 and then the labeled t5 and then label ta. So, with this information we ever router insert a TD tdz can determine the pdz domain containing the TDC, because for each internal rotor we have a TDD tlv. We have a flag in turn, so this indicate a for internal router. Every a Jason router is a TD g router, so 48 rotors because we have tdz, is M. Sabbath ERV.

M

This sabbath lv, contains the information about PTZ levels of this nature offers. So with this information, the whole TDC is determined. That's for determining the t-d-z. So for virtualizing tdz we just ate the other ages, as is La Paz, but we dunno here for this edge router, we just added the other ht1. The answer: a gt3 and the delta agt 3s is labels so with those information Amy road outside can see this virtualizer that virtualizer the metal piece easy.

M

So this virtualized path, because this is just that normal is labels so from outside you just see these normal connections so from outside is just see. The lomo virtualized of that network is not aware. Ne ne T disease, so next slide so for agency is a symbol. So when we configure ttz, so we will add a tdz tlv with the TDC ID into hallows.

M

So on both side, if we decide that the TDI dr same and then we can form the TTC adjacencies, so if they don't same and then we don't inform that it is adjacent least so for at TDC a drover, it forms adjacency with the road outside in the normal way. You can separate that the just synchronize the database example those internal informations. The internal information, will now stimulus to our side.

M

That's the only difference so next page so this page, we just saw the operation of migrations so for for my great a part of network into a TDC. Just follow these three steps. Very simple steps: three steps: first, which has a configure PDD so for comfy, tdg, normally just one command on each other, ttg routers. So, for example, TVD of this world config aren't easy and the rods get one commands. So after we configure tdz and then we just use one command for hope for how this network.

M

But what is half this one camera is trigger. The TDC information is jewish distribution, one command for ho papa and then the third step is that we use also only one comment on any one of the note for migration of this path to t disease. So sorry, steps, that's from operation per view, is very simple and easy. So from the inside so are we configure tdz? Then we will discover t disease.

M

As I mentioned that way, I put hallo for the TV Guide PDD tlv, with a pdz ID in the hollows, and then we can discover tdz, adjacencies and then other after the first step. The second set of eyes that every TV router just add TV gTLDs into their a las veas.

M

So this additive steak, so this curve cause one to the second step. So the TDD information will be distributing in the network, so the TDD will be discovered and a determined and then the last step is that we use one command on any one of the routers of TDC migrate is part of radical 2 t disease, so for this command is, will trigger the AG router to virtualize a DVD, so the ADA router will add.

M

Those kind of virtual labels means the the author ages as there is labels and then also if they are now to do with the t-d-z information to our side. So that's time, so we can see that those three steps from outside and the three steps from the inside and then the DVD is a transform without any stepha stepha interruptions and also the operation is very first street or done symbol so next step so that staff. We would like more comments from the working group.

H

Less Ginsburg, it is possible today, with the base protocol to quite easily and non disruptively, merge and split areas. We don't require any extensions in order to do that. It derives from the fact that you can configure multiple area addresses on a given box and when you do, they become synonyms for the same area for all the routers in that area. I won't spend the time in the meeting to go through the steps, but it's very straightforward to split one area into two areas, for example non-disruptive lee with existing functionality in the protocol.

H

So with that in mind, it isn't clear to me why we would invent another extension to do the same thing.

M

You mean so you are not bring the link atang with the old area or not bring the loaded up with the old area. We just uh you.

H

Can do this, none disruptively.

M

So you can give up these multi tails and then we can yeah.

H

Maybe I'd be happy to discuss that with you, but.

B

H

A in a well-known procedure.

M

N

Long, we leave for Maui.

B

N

Actually, this can solve much more than you just described and, for example, recently I saw an operator. They have a network, so they upgrade from one continuous louder's to another company's routers, but somehow in the pub they forgot to like take some out loads out, two loaders still there, but it's misbehaving. So, but if the configure disease, something like that, so that like a loose link, States advertisement will not be popped out.

N

So it can minimize like and the combines the rail system and disruption and on other hand and this technology- and this extension can use your topology complexity, because without yet in your topologies, that tables may be bigger. So there are multiple application. Is it's just not only for like a splitting your network into a smaller ones,.

C

This is harness. um I do not really quite agree using vanilla mechanisms, it can be done, but I certainly would agree that, for example, using multi instance is, is to model a TTC song. It could be done right so either way. I think there is already some tooling that we have in the toolbox to sort of model. Those zones and I would like to see whether ye reuse of those tools than to reinvent something new.

M

Yeah we like to look at as tools but I do I, don't know whether that true can replace or all the functionalities here. Well.

A

Why don't you mean, given you know what less is mentioning about the multiple areas and stuff? Why don't we take this to the list and flush out whether we can solve this problem with what we have already okay.

B

A

A

I think that's very best.

A

Yeah, so that was our last presentation. Does anybody have any thing else to add to the meeting.

A

All right great thanks, a lot see um proc, hopefully on a different day than Friday I, really worked hard to try to have it not happen this time but I I I'm gonna, put almost zero conflicts down.

B

A

That's true, so you you're the Alpha. We were almost the Omega.

B

B

A

Afternoon, one.

F

A

Right, the penultimate.

B

A

They waiting outside the door right. This is our room. You know like.

B

B

A

Anybody didn't sign the blue sheets. Could you come up and sign.

B

A

I think there might be in a blue sheet out there somewhere. Ok.

J

A

15 hour is 15 our battle.

B

B

A

I'm going to you know, he's gonna make me I've got like a half an eye esaias I get done to like you, but now I'm going to this is pretty clean line up here.

B

A

It's not get up.

B

A

I on my night actually went. Oh.

B

A

Why he's using 3.4 yeah he's, probably using like a synchronous or whatever anyway, I did I? Did it with memory you so I'm actually doing like 0 copper pipe down, let's copy so I'm using memory, views and I did a single library extension so that I can actually poke and poke within that member you.

A

So I get the framing off from throw that link, but I never copy, like I just carry that memory buffer, because that was that's where I thought like, if I'm going to see any kind of serious degradation in performance, will Python I thought it's probably going to be from copying things all the time so I said, let me focus on minimizing that, because only really one is a Python eyes that actually performs, as well as a super or close enough single fellas.