Internet Engineering Task Force 93, 20 Jul 2015

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF93-IDR-20150720-1740

Description

IDR meeting session at IETF93
2015/07/20 1740

A

B

A

B

A

C

Started with the focus.

C

B

D

Is this one any better.

E

C

Folks, we're going to be free, we've already lost a couple of the minutes. Please make a note: well, that's it.

C

Ok, there are 0. This is much better. This is the status of the draft. Read it. We've done really well in dealing with working group. Calls thanks to everyone. It's now time to bury of their Alvaro with grass, so I, don't think I think he'll enjoy the challenge.

C

Administrivia I've cut everything out of the agenda that have more face time. Don't wait, go on with there's a status online go on. That's yeah, you were right, I was wrong and online. There is something that anyone who is a otter I am now.

F

C

People up based on feedback from Alvaro to put implementation reports online for anyone who's got a written draft I'm using those is my test case. So I will do that for you, but before you go to the isg, anyone who has a protocol must fill out the implementation and put it on the site. If you're doing an administrative draft, you don't get out of it. I'd like to have you. Do a testimonial, basically get a quote from Jeff I and the I I Anna folks and saying it's cool I'll.

C

Do it for you Jeff for you? Yes,.

C

Yes, um I was trying to just it's a wiki guys.

F

B

C

Your friends type it in yeah, yes, that we good but I, intend to do the summary of the stuff in the draft. Stefano I'm done.

A

Yes, mistake in the you.

C

Want the clicker, ok, so.

G

I have three minutes right, so thank you.

G

Ok, BTW pls for epx 10 chains. It's not working group document next slide, just a very small reminder of what it is. We want to advertising BTW pls, the topology of the peering points. So what are the piers? What are the interfaces without appearing yeses and so on so to feed the controller with this information using BTW pls extensions, the next slide, so what we changed very little things. We just use different objects for the piano teeth and the peerage agency said in the previous version both had the same object at the india encoding.

G

Would that was the adjacency said now we have different so different code points. The format, of course, is always a set so that the formats stays the same. We just called them differently.

G

We added some text on the way we identify interfaces through addresses or identifiers, and we really thanks ac for the in depth, review and contribution to dollar to the draft next, which was the last in fact, so I think I stayed in the three-minute budget. Thank you.

F

Good afternoon, everyone here is a brief update, updates t as TR SPD state distribution using be GPRS.

F

Okay, this a brief recap: this truck provide a mechanism for the collecting of the tears these days. It is based on the BGP are as architecture which provide a unified protocol for the network layer, information distribution and this mechanism is complementary to the pc based SPD report.

F

Since, in some scenarios, some areas may not be the PCCs, so they need another mechanism to clad to the aerospace days and also the BGP route. Reflector can help to reduce a session overhead, okay, nice line. So here's updates and working group adoption, we add the support for the segment routing TRS piece is it. This is achieved a base on a new protocol ID for the second minerality, and we add an operational consideration, extra section to be aligned with the be GPRS base document and also the TSB object.

F

The list is updated to include all the relevant objects about the T aerospace state. Also, we update the inet section. That's right here are some comments we received on the mailing list. The first one is how to specify the switching type of a non packet RSP so country. Our proposal is to use the generalized. The label request object in the are spaced a tre to carry such kind of information, and another comment is how to distinguish the object defined by RSVP and the pc.

F

Is that a proposal on this is too maybe we could define a top-level tlv, the top lab theories based on the source of the TSP information, such as one for the RSVP and another one for the psf. Actually, the officers are open on this, such a compliments and the solutions. So we would like to discuss further along the solution to these comments. Then we can update the draft accordingly. That's like okay, so the next step is to solicit more comments and review from the working group, so it can revise his draft.

F

H

So the date on BG p.m. tomorrow, water, hot affection next, is a recap of the goal. The goal is to provide optimal patil routine, without bothering, where is the water factors or where, as a client Oh? What is the network topology between the boss.

H

Next, please, so there has been a significant changes: induction 9am antenna there used to be multiple solution: adoptions in the draft on and we'll remove the many options on refinement and we get a single solution which is a 212 support, vectors computer, it's igp location, better than an arbitrary or location IGP location, so that location may be a one location for the world of other factors or one location, /, a group of rogue group of peer or one location per client, typically, the location of the client.

H

Other choice is up to the implementation, so it's very open or other alternative solutions have been dropped. So this includes the angular distance approximation as a senior in between the client on the rotor factors of the group ID, the Mon dating of a per client IGP computation, and also the perc IM bgp policy. So or is it all? Those options are removed from the draft, but as a reminder, we still need the water factors to have also pass prior to AGP a tiebreaker. So typically, we need a bgp I'd pass between the water factors.

H

Also, obviously, with many the editorial changes on textfree white 22 under that those changes next one. So as a result, the draft is a is much simpler. It's also in line with exiting implementation on its applicator. When you move, we want to move your your work affect awesome. So, given all users jones-drew those changes you may want to to reread the ratification of the draft because it hasn't changed at any question.

I

Good so I'm talking about the internet draft, making what soffit aware of data link failures that I XPS next, please it's a typical scenario is: if you have a pitch possession that data plane and the control plane are concurrent. So if the data plan breaks, the control plane is able to detect that next place. That is different in an IXP environment where you have a rod server there. Usually you have different data plane passes so for the PG possession and for the data plane passed between two client rotters yeah.

I

There are different passes and if the data plane breaks the control plane issues, we not able to detect that, and so data traffic is lost so ever black hole. Next, we came up with a solution to fix that. First, we we need blind. Rotters must have a mean verifying connectivity among themselves, and for that we can us BFD, which is in our proposal at least configured to a synchronous mode and data rate of one packet per second and detection after three meeting packets.

I

Second, we need client orders, master, have a means of communicating the knowledge so getting back to the rod server, and for that we move. We use the p2p link state and there we models IP network, as notes so for the client models and for the rod service and the links between the client rollers. If there is a connectivity between them and we need a next top information bay- is to store this reachability information and for each appear next.

I

So let's have a look: how this works. Let's assume PA set up a PG possession with routes over then the rod server sends out a p2p update to puree with the next top information based updated. So it says that it sees note B and the client rato receives that. So in this case a and start automatically verifying the connectivity between a and B using bfg. In this case and after the connectivity is verified, it updates its own extra information base and sent it back to the right server.

I

So you see that here with the link is, is available in this message. Next, if the data link breaks the client rato detect set because BFD tells it so, the client router updates its next stop information based sister I'm. The link is not well the links, the link section, ops of the next stop information based. It's not filled and sent it back to the right over and the rod server use this information during the route selection process and yeah remove all the next hops that are declared unreachable for that pier next.

I

The status of this internet-draft is that we have already achieved the backing probe adoption and we changed the next top information base. Yeah the way we transport the next stop information base from the client rotters to the rod, server and back from next top cost of each billing state. So we are looking for yeah for feedback from the working group and also input how to improve the account in her draft sexy.

C

It is Jerome Jerome here or David. I promised them a chance to present this a second time, instead of just in the interim going once. Okay.

B

C

Next person is Stefano.

G

Okay, an update on the BGP, perfect sage draft.

G

The big change is the merge with the hannahs proposal. So now we have a single proposal and change. A major change after this merge is that now in the prefix said, we also carry an optional TV, which is the originator srgb tlv. So we changed the format in the previous version. We had just what what was called the label index attribute. While now we talked about the prefix it attributes inside which we have multiple tlv is one of those. Is the originator srgb?

G

Another tlv is the index that was already before, and we also added the support of segment running ipv6 data plane, so we had a tea leaf or for the ipv6 case we address so two use cases: the labeled unicast prefixes 3107 and the ipv6 graphics, with a multi-protocol BJP. We have multiple implementation, so I think we are ready, at least for asking. The adoption is working group document next slide. Those are the different tlv. So the first is the label index tlv.

G

This hasn't changed, so it's the same as the as the encoding that was already in the previous version. Next slide, the ipv6 Sid tlv is in fact just a flag because in segments running ipv6 the SID, the segment identifier, is the ipv6 project itself. So you just need a flag to tell that the router is capable of doing segments running ipv6 for that prefix next slide: the originator srgb.

G

This is what comes from a harness proposal, so you encode the originator srgb in the form of ranges, and you may have multiples, so you can encode multiple srgb ranges next slide.

G

This is the same yeah! Okay. It was just to mention that you encode the originator srgb, and this should not change across the propagation of the of the update next. So we take always the the reference of of the data center by the way the use case, the prefix eight addresses is documented in spring working group into the MS DC draft next slide, so those are taking from the presentation that key were made last time, so nothing has changed it really.

G

So the idea is what is to propagate in the prefix eight a hint for the receiver of the update on which label to allocate in order to do the segment, writing, mpls data, plane, forwarding and so the label index tlv now contains the index that you advertised to your neighbor and the neighbor takes this information if it supports segment rodding and allocates label corresponding to today's. The proposal of the prefect's age assumes that all the runners are configured within the same srgb.

G

If, if it is not the case, it also assumes that you can know the sRGB of each individual router, either through by example, BTW pls, either through the srgb originator tlv, as defined in the prefix it. So the update gets propagated note by note, and each node allocates each segment routing capable node allocates the label according to the value in the label index tlv. Now next slide.

G

If, in the middle of all this, you have a non segment rod incapable node, it's still fully interoperable, because the non segment rodding capable rod will allocate it's it's own label for the for the for the prefix. But still the update is propagated with the attribute, so the next segment writing capable node can continue the allocation based on the index and this interoperability is granted because the label you have to use as an outgoing label is carried by the another ride. 3107 nlr I nothing changes there.

G

The attributes is just carries just a hint for us are capable routers on which label to allocate so by making the distinction. One is for the label, outgoing label use and it this is the nlr I and the other is what is the label I have to allocate? This is the label index tlv in the prefix it by keep those things that were rated, you can ensure a full interoperability between SR and non SR nodes.

G

I think this was the last slide, so I can go back to the first slide and and ask if well next, if the working group can consider the adoption as its working group document.

D

Okay, can I get a sense of who's, read the draft, so quite quite a lot. I would say you know we sure we can take a show of hands for people who think it's ready for adoption.

D

Quite a few hands a similar amount. Anybody want to say they think it's not ready for adoption. I, don't see any hands. We will, of course confirm this on the list, but it looks pretty promising. Thank you.

J

So hello, my name is Daniel Walton work for cumulus networks, send myself and my coworker Dinesh debt or the authors of this draft.

J

So the draft basically just defines a new capability that BGP can use to inform our neighbors of our host name. So this is just the normal host name, that's assigned to the speaker, and we encode this using utf-8. This is only to be used for basically for troubleshooting and for display in either debug output or show command output.

J

It's this is implementation specific, but you know the normal use case would be that you provide a knob to enable or disable showing the host name. So this is probably the easiest capability in history. So this is the entire thing. So there's just the linked field for how long the host name is and the same itself and then the same for the domain name.

J

Okay. So what's the motivation for this? So our customer base is primarily data center and a lot of these big data centers. They tend to do all a BGP and there is no ibgp, so that means you're BGP peering to directly connected addresses.

J

So the challenge with that is normally you know if you're appearing to route reflectors, you kind of in your head memorize, which IP addresses which route reflector and it's easy, but if you're appearing to directly connected interfaces everywhere, all of the addresses are different, so you can't really memorize. You know who is who.

J

Outside of the data center, this may also be useful. You know, ospf anais is both already had this ability to kiss you know include the hostname as part of the routing protocol.

J

Ok, so just a quick example, so this is a show IP bgp, summary from a leaf node and a data center. So the first two sessions are to a spine layer and then the bottom two sessions are to a tour layer.

J

The key here is, if you were to go to the the sister leaf node and do show IP bgp summary, that box would be peering to all the same spines and tours, but you would see different addresses, so it's very challenging to jump from one box to the other and remember who is who so with the draft in place. This is basically the whole change. You do show IP bgp summary and you just display the host name. Instead of the address that you're appearing to.

J

So feedback that we got on the list that we will put in the next version, one was to require the extended optional parameter draft, and this is just to avoid the risk of some really long, hostname blowing out the size of the the open and then the other feedback was to provide some guidance on what to do. If you have two sessions to the same box or if you happen to have two different guys that use the same hostname.

J

So again, this is just implementation-specific but we'll put in some. You know some guidelines.

K

L

L

Job Snyder's ntt I recall from the mailing list that there was also some feedback that this might not be the best idea.

J

That's putting it mildly, yes say. Yes, there was a. There was a lot of activity on the mailing list that that weekend, a lot of people, some people liked it. Some people didn't you know, I, don't know that we'll get consensus and convince everyone that they liked it. But you know I our customers like it and want it our stones, car and we use reverse d enough. So the problem with using dns is, if you're doing ipv6 link-local hearing there. You can't dns that.

D

So if you guys want to make comments, um please actually come to the mic and do it. Thank you.

M

Lastly, man from that node um I always a bit puzzled by that utf-8 statement, because if these are host names, there's an RFC that specifies the the certification of a hostname and there's no need for utf-8. You say: yeah, it's clean, that's fine, but once you go into the utf-8 deep hole that there are lots of various mazes in there that you need to carefully consider, for instance, having two characters that look the same and printed but I actually different differently encoded. So big, big rat nest, I gets.

J

It so you, the using utf-8, was one of the points of feedback on the mailer and it was from jobs, so be.

M

Careful check out carefully.

K

Hi Jared much from auntie T, so uh you shouldn't really be using link-local for the bgp purine was I could what there's a lot of reasons? Why, starting with you, then have to go and do some sort of trance market if you're doing like ibgp or anything like that, you have to go and rewrite all your next hops for that. Just using that, it's not!

K

It is an exercise in providing yourself a non deterministic outcome, as opposed to using a global rerouted IP address, or you know, some sort of ula on the interface as opposed to doing this, because all it takes is someone to patch a cable in on the wrong switch or the wrong location, and you not only lose connectivity with your stuff because you're not in the same layer to broadcast domain, but you have a whole bunch of other consequences that can happen when you come to debugging and usability when it comes to a network operator trying to actually interact with the device, so going, you know, is setting up reverse DNS.

K

Isn't that hard? uh You know if you need help with it, I'm sure we can find a couple people who can help you, you know, configure it in the zone file and make sure that you get the right thing there. It's really not much of a barrier. It has to be used for a whole bunch of other network devices, especially any device that talks mail.

K

I, know that this device isn't talking smtp necessarily but a lot of these new routers that have they're all linux pace and stuff like that they may actually send out email or other telemetry as well and call back and having the dns stuff in. There is really helpful when you have all these additional things to have a human readable thing, as opposed to fa 0 colon colon blah that you can't map back. Unless you go and you use the mac address decoder ring it's quite challenging to operate a device like that.

K

I, don't know how much experience you have in operating networks, but going and looking at all of this stuff. It's really important for us to be able to trace back stuff, especially when hunting down customer complaints.

J

So let me let me kind of clarify on the on the ipv6 link-local hearing. We.

F

Do have many customers.

J

That use this in our implementation, the way they can figure it they don't. We don't force them to actually type out the link local address. We discover it via neighbor discovery, so they would configure your neighbor gig 0 / 1, and then we peer to the link local that we dynamically find on the other side. So if you switch a cable, it's not you know the end of the world. So.

D

This can cut in line here for just a second and ask you to clarify whether in the deployment you're talking about the router is exposed to the public internet or not, or if it's sometimes one way, and sometimes the other. So.

J

Mostly, these are going to be with an in data center. So that's that's the primary use case yeah, so I'll follow up on that. What.

K

Is the data center connected to the Internet? Yes,.

J

At some point, I'm sure, okay, that.

K

Doesn't mean they're doing link-local.

J

Peering at that point are.

K

J

I'm so add what history beachy, summery I I didn't catch. What you said add a column.

A

You notice there if your customer actually is not leaking hose, is wrong.

A

The identification.

N

Jeff has I'm the one that said the utf-8 and unfortunately, the reason I am suggesting this specifically because if you go to any of the apps guys who are going to be the ones reviewing the stuff of died as gee, that's what the isg operations guys say. We do these days. We don't put in just a skate, because we can't do no just servicing American customers. We can't do languages that are restricted to stuff that can be represented in Roman characters. So it is the right thing to do to the second half of the comment.

N

We probably don't want to use DNS style puny code, although it's an interesting idea, because you sort of devolved down to the existing DNS libraries for expanding that stuff, but all puny code is going to do is- is going to expand into a utf-8 string anyway. So you're just moving the problem around slightly. If you do that.

D

Any further questions, comments or flames.

D

Okay, thanks Daniel.

C

Luanne Luanne, just for your information, we have a good amount of time. You don't have to hustle three or.

O

Less okay, hi I'm Louie I'm from working in Microsoft I drew networking in the svn team so move on so this craft. We started with pretty simple saying that we want to have partition. The motivation of doing partition of the network is really for scaling. When you have tens of the millions of endpoints to consider a one flat Network you happy yeah, you have a scaling issues, so we want to do partition here. We're now going to say more about how you partition and it's more when you have a different partition right.

O

Either you see in the data center environment or seen any other environment. What do we want to do with a BGP I you? So here's a concern and for this version, change different from 0 0 1 is a problem, the applicability of the proposal. So initially we were talking to to support HST and now it's saying for any network. You want to partition and you want to use labels. You want to use bttl you to distribute labels, you could use such scheme, so it's more brother. You know applicability.

O

Second point: is we modify the procedure so initially we have the coverage of using ITP. Now we say: okay, if you use vtt only that we have a procedure for it to the third point, probably the most important. We are introducing a new extended community that is to identify the partition, basically enable the border nodes to. When you have partition.

O

You have borders right, so the border know who ambiguously signal the remote bgp speaker saying with that new PDP audio procedures requesting partition, unique label should be executed and enable the border no than PB speaking label mapping server. That's a server managing the label mapping it's more like a directory thing right to scope, the label request and the response to a unique partition. So basically we're.

O

Maybe I should put up front so easier for folks to grasp this I think about in the overlay word. We use a free VPN. Basically, you partition the overlay space, so each VPN has its own rd RT. You can use that to identify your party, a particular VPN. You can use the same address in that different libyans, but you have identifier, that's how you identify isolate your particular partition in the overlay player, player evm, for example, and this is in the similar principle world doing it for the underlight.

O

So in the under layer you can cut into different partition different domains and you can do the same thing with the extended new extended community to identify that partition. So that's about it. Next page here is a topology. This is just abstract example like we were saying it can be in data center environment, we started looking the very very large scale in the data center environment, but it can be not in the data center environment. So there's no restriction and also it's kind of a just, a abstract model.

O

You may have the r1 r2 inside the partition. So this shadow, you know a B and one that's a border note. I should put another being one being to their border, no ambien Trivium for their border note and in this one that shows a to partition me sorry to hierarchy. So it think about this. You can have one partition, maybe a core, and then you can have the next level partition. Many of them may be data center, maybe some other network and then under each one.

O

If you really want to have cut the partition small, so you can manage that you can really key that and so on. So you cut another level partition. In this example, we use a two-level, so you gotta be one being one being too at the border being three pm-4 as a border and the tour there are not the border, so the border and the number has different functions. They all connected with the bgp sessions to the lms that is labeled mapping server.

O

This is well, it's a label, mapping server and we may explain later, but I would like to mention that this label mapping server does not function as normal PT p-rod reflector. Initially we're trying to use a PTT reflector to modify the procedure saying I, you do this type of a function. When you see the rouse announced to you, and then you sign the label, you don't do as a typical PGP procedure, saying I will I will calculate the past. I will make a decision you don't this one will.

O

What we intend is to have the operator already put the design inside there, so they border. Why do we have the border group? The BG? The border group is used when you're doing say ecmp you're targeting to a group of notes, so any of them can do the forwarding the normal ecmp. So you make them into a group. You can use one label to indicate you're going into a group. If you want to do t type of thing, you want to select a specific path. Then you don't target that to the group.

O

You use individual April um under Lake partition. I already explained next page, so this one is to show the typical PTP how you process you get an MRI and you have the CL 3. So I have a little explanation on the CL 3 and see how, for they were, they will be now label PHP if the same outfit is used for both UV 1 and UV to UV means underlay partition, otherwise they'd, otherwise locally allocate the contacts label will be used and b + 3 + BM for append a bgp community values bg2.

O

This is not the new extended manager. Community pg2 means I'm in the border group to when they advertise of himself in the u.p one. So the green is the the advertisement from p.m. for and the blue is from b & 3. Ok next page ah in this page, so we show from the view of p. Ok, it's a border node of the up1 learn route to p + 3, + BM for learn the routes.

O

So, basically the top part. The four routes are to blue to green is from that's from you from a PM wines. View p.m. one on the left saying that alerts p, +, 3, + BM for p + 3 is the blue BM for is the green. They each have two ways to go: that's why they each have a two routes. So the little note on the l 113 the contact, what doesn't mean we're basically saying that you're in the partition one and you go through our one and it's from p & 3.

O

So those are the small details and the same thing as for the be into view how they be and 3pm for, announced, clue, p and two interior routing the you p 1 r, 1 r 2 in the middle here they're, not the border note, so they do not modify or remove the border group community they're, just passing it when p my nepean to receive the LBGT routes for being three and BM, for they can conclude: okay being three ambien for belong to PG to group next page, that is between the piers right.

O

The border from this end announcing to the border notes on the other end. So now all these notes we're talking about p +, 1, 2, 3 4. They are all advertise himself to the lms one. Unless one is the label mapping server one, because it's that this is a UV one and the next one is easy to write. So we just take one domain to look at it, so they announce themselves over one two, three four, so it's pretty intuitive there. So here is the point we already mentioned.

O

Ms/Ms may not run if you wanted to run other functions, that's okay, but in this case what we're doing is saying it doesn't require. But in this particular case it doesn't write, run regular bgp decision process to compute the routes. So that's what we're doing with the lms one and lms one learns the group membership, bn3 and BM for from the abbey TV advertisement next page.

O

So so far what we've been talking is pcp. Are you procedure? You can just do that with the today's PGP. Are you what is a new here? So basically we mentioned: what's new, we're introducing new extended community? Why do we want to extend a community? Initially we looked at the first version. Oh draft we didn't use extended community, we say well just modify the reflector procedures on to perform such a function. You don't acting as typical. You know in right decision maker and other things. You just say: hey I, I hold all these mapping table.

O

I can tell you what to do. I learn runs from you, I tell you what label to use and so on. This is like designed by the operator. So what here is saying that now we we looked the by doing that it may not be such a clean way because you're all reflectors and doing other things. Well. How does it know when to do it in such a way when not to do it in such a way? It's not the regular behavior and to to deal with the individual partitions.

O

You don't have a very clean way. You can do it. It can definitely do it. It's doable, but it's more complicated than just give it a new extended the community. So in that community you want to do partition. This is a way to identify the partition. Very you know I like to use analogy with the layer. 3 VPN: that's how you use a rd RT to identify your VPN. So that's your partition in the overlay Network, and this is your partition in the underlying network.

O

So you use this extended community, basically to it to identify the partition so BM 1. It remember the picture right be in mind. The B & 2 are configured to be partitioned border nose for the u.p one, so the partition represent in the bgp extended community value, that's one and two, when B and one and the BN to learn a destination that in that last picture, was being three pm-4 through the LBG p from a beautiful years through our 1 and R 2, R I. Remember there in the middle.

O

They don't change anything and that belong to you p. One partition, then be a man and beam to do not allocate a label from a platform label space. What do they do? Instead? They learn the label from for the destination. The destination in this example is being three pm-4 in the partition label: space from the label: mapping server, so we basically using the labor partition server. As the mechanism we can put the operator design in it, okay, but we still want to use vgp for label signaling distribution.

O

The reason is, if you consider the pure as the Emperor as the environment, whatever the way you don't use that you you can do in a way like you have a controller and then you have every box, you have a agent, you can do that, but that's every age of me. Every box support an agent today, we're talking about brownfield or maybe for a long time were in such a brownfield, a hybrid environment. You cannot really practically doing it, but every box, almost everybody in the field.

O

We know they do, they can do PCP, so we can use bgp to help us distribute the label, and but here the BP use is not. The traditional way of using next page.

O

So here is to show the routes. Now we added this extended community, extended community, see in the from BM 12 LMS 12 the label, mapping server, saying hey: this is from you know the bells to P and P & 3 and the BM. For so it noticed the degree and the yeah and the blue posts added the extended community are, is saying, I'm requesting and that's a up1 content, so it identify which partition you're in and that next one is saying from the LMS to the BM one.

O

In reality you can see four, but in reality there are three labels. One is unique for JP and 31 is unique for being for the last two you know one is to give to be, and three wines gives a vm to good being for, but they are not unique there because they're in a group and the G we defined as group. So we use a group too yeah, that's a group, so that's a same label and plg to is the partition label assigned for barter node group bg2.

O

So just explain what we mean by all these symbols next page, so the end too I think we can skip those. Basically, this is talking about the same process. We talked for being one now it out for being too same thing right at the end, too, we'll send a request: hey I need label for these I already know certain things, but I need you to assign the particular label for me, and then the LMS was sent here. Are the labels so including the.

F

Individual level, including.

O

The group lip next page and all the BN BN the border notes in the u.p to advertise themselves. I think the last part word we're talking about. mmm Okay, this one is the reason we put this one here is because there is a we have this node week, how, let's say their tour right: they are not part of the border nodes, so we we want to show they will have little difference in the advertisement.

O

The border nose will identify the border roop, which order group they belong to, and these guys don't belong to any border group. Therefore they don't carry that community next page and piece of the u.p to learn routes to tour one and tour to so. These are the routes passing the degree and the blue I think we're kind of okay now move on.

O

Yeah I believe we're talking similar things now, so maybe I should talk a little more. The last page yeah back a.

C

O

Yeah back okay, yeah so from from the view of a be entry, and you will have an MRI with the tour one and that they don't have a community. They don't have a reporter group there so same as a tour to sotto aunt or two and going through this way going through that way and toilet or to do not belong to any order group already set in this example. Therefore, it just the simple format: I, don't think I need to say the last, but did the last part is the same.

O

It's a be in three view and be in for view. Okay, next page so now, p, + 3 learns partition label for the tour one and four to and from Abby and three to the LMS. To so now we're talking about this partition. You include the tours right and these are the requests so a label for it and then that's a response back from the LMS to to be- and three saying here are the 10, the labels? Ok, next one, so here's our summary say.

O

Basically, the reason for partition it is a partition is a key aspect for scaling. um Ptp is considered here as a tool. It's a natural glue to connect the partitions and the is widely used and essex me. We use a heavily so the next, the new extended community allow us to support the underlay partition in a efficient and the cream way as because we try the multiple different way. So maybe we should do this rather Clint and messed up, so many other things to be complicated.

O

This will be very simple if you know I layer, three VPN well, this is the same principle, so it will be similar to the layers VPN and supports brownfield. The deployment well, mptp is used as a protocol to request and the learned operator assigned the labels. Here, it's not to decide on the that and the computer out the procedure defined here can be used for any partition technology. So it's not exactly at the intent at the beginning.

O

Right we say: oh god will have a design on the adjusting, and so now it's beyond its title right. It's any partition! You want to use such a thing. You can use the next step. We really like to get feedbacks from the folks in the working group and we'd like to know very much welcome contributions and to work together on the draft, and once we have you know new update to make it more know, to get a feedback to complete and to modify and so on, then we'll ask for working group redemption. Thank you.

D

Questions comments, flames, compliments, Wow, going once.

B

P

Lucy hi Lucy I'm from Huawei. What clear vacation is this a label allocation also for just for unicast package or multicast, the Ohio another, a location for the multicast label.

O

We at this point will consider just building the pipeline's right if multicart is something bring additional. We actually look quite about multicast thing as well, but well welcome to discuss.

D

Anything else, uh two things, then one is ICU. Oh no, actually I seem to blue sheet, so nevermind. The other one is to remind you that we have a second meeting on Friday anything else. You just so.

C

The second meanings time is: 1152 1320. We are the last slot on friday. So please we hope to see many of you here.

D

Thanks everyone we are concluded for today thank.

O

B