Internet Engineering Task Force Interim Meetings, 21 Sep 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF-LSR-20220921-1400

Description

LSR meeting session at IETF
2022/09/21 1400

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

A

I'll handle that okay.

B

Mike check hello.

A

C

We can hear you hi guys hi.

D

D

C

Ajin we can hear you.

E

Okay, thank you.

A

You've got a little echo on your side from your speaker. Probably if you have a headset, that's probably better.

A

All right, the timer picked it's 10 o'clock. Well somewhere.

A

What are the 1400 utc.

C

Yep, that's exactly what it is.

A

Excellent wanna get started sure. Okay, let me I guess, I'm the one that should be sharing.

A

Probably should have had this up from the start. um Hi everyone welcome to the first uh interim for lsr in 2022, probably the last one as well.

A

um These slides were put together by young jen she's, the secretary, also the chair for out and working with. I think now um thank you for that.

A

Here's the note well.

A

And what it says is that by participating in this meeting or in the ietf in general, you agree to follow the processes and policies.

A

You're aware of the ipr rules.

A

And you agree to work respectfully with others, there's more to this, and you can read it on these various documents that are listed in the slide.

A

So here's the agenda- these are primarily well. A lot of these are presentations that were not able to be given at 1 14, because we didn't have enough time in our two hour block.

A

Does anyone have uh some agenda bashing? I want to bash this agenda.

A

Okay hold on let's get started peter. I think you're going to be first up.

F

Yeah sure chris, can you hear me.

A

Yeah, I can I'm going to pass you the control of the slides.

A

Okay. Is it showing you the deck I'm sharing or is it showing you a choice.

F

It is showing me the deck you're sharing.

F

C

F

It says christian flights.

A

No, nothing big slides back.

A

Now, what does it show? I'm just curious.

F

I still see your slides.

A

Yeah right now, I'm looking at the selection list, so I wonder if it would give you that, let me uh so I I have to be the one that starts. I guess picking.

F

It says the news slide: deck has been shared, but I still see the agenda.

A

I'm not exactly sure, what's going on.

G

Chris, maybe you should try to stop sharing the current one. First.

A

Yeah, so I did that and then I clicked share okay. Now it's giving me the choice. So then I picked this so I have to pick the slide deck and then pass it to peter. That's strange, but okay.

A

You should have your deck.

F

Yeah seems like I can even slide it up, but.

A

All right thanks.

F

Hi folks, uh good morning, good afternoon, good evening, whatever the time is at your uh location, I'm going to cover the simple draft for the flexago rivers affinity that we put together with the qr code as listed here.

F

So what are we trying to do so today, with flex algo? We have support for affinity, related constraints which is exclude and include and include all and affinities as such are advertised as a link, attributes and as such. These are unidirectional properties, and these affinities are being used in one direction, basically in the direction that we calculate the spt, so we only use the affinities in the forwarding direction of the spt.

F

What we are proposing is to allow the usage of the affinities in the reverse direction of the link, and it can be either set manually or could be set based on some. You know measurements or some other logic. Whatever the implementation is willing to do.

F

So well, I used to have a animation here, but with pdf I don't think this is going to work. So basically, an example of the use case here is the primary path between one and seven is one two, three seven. Now there is some input error rate uh measurement happening on the three and and that's three can actually figure out that it is receiving some input errors and, uh if it tracks, if it crosses the number of errors, crosses the threshold.

F

The three can set the the affinity in the directions three to two and then, if we use such reverse affinity as a constraint in the flexalgo calculation, then basically what the one and two and everybody else will do. It will avoid the the link. So the paths will be one two, four three and seven avoiding the link with the error rate.

F

So that is uh the high level concept. This is being done or requested for a specific service, which we called an ultra reliable service, where the user really is very, very sensitive to any any drops in the traffic.

F

So the mechanics that we are proposing, we are proposing to add a three new subtleties into the fed, which is an exclude reverse admin group include any reverse admin group and include all rivers, admin, group, tlbs or sub tlvs.

F

This is analogous to the existing ones which are using in the in the forwarding direction and obviously it will be defined for both igps.

F

And in the calculation of the flex paths, we add the three rules which are very similar to those which are used for the existing affinities. It's just that the affinities we are going to take is in the reverse direction of the link that we are adding uh into the into the spd.

F

So basically, this is a new draft. uh We are welcoming any comments, ideas you like it, you don't like it uh whatever and if the group decides this is a good idea, then we would like to take it to the working group. As a working group item.

A

Hi, so I think it's a good idea. I have a question, so the um the reverse affinity is this a new. This is a new attribute right. It's not using old one.

F

So this is, I mean, attribute from the link attribute perspective. This is not new. This is just the the affinity which is being advertised as it is it's just that today we never use the affinity on the reverse direction of a link we always use and consider affinities in the in the direction of you know from a to b. If that is the path that we are calculating, but never from b to a, but we are not defining any new link attribute.

F

What we are defining is that during the spf we will use the affinities, not only from the you know, forwarding.

A

Direction but also yes, so uh I would the reason I so I came at this from a specific thing I was going to say: we should call this the receiving affinity, not the reverse affinity, but given what you just said, it truly is the reverse effect. It is.

F

A

So if three to two changes from blue to red, uh that's going to affect that.

F

I want to explore basically.

A

Seven to one right two to three so.

F

Without without your, without.

A

This yeah go to the thing so without this right, uh not considering this extension, three is going to change three to two from green to red. Then traffic from.

F

Nothing, but one is a.

A

Seven to one traffic will change right.

F

Yeah, seven to one but not one to seven, but we want to say right.

A

And we want to change.

F

One to seven yeah.

A

Yeah yeah, but seven to one will change well.

F

Not necessarily it depends, it depends if you have affinity, so you can the the the constraint for forward affinity and reverse affinity is going to be independent. So if you have the constraint, saying red existing exclude or include red, then the three to two change would affect seven to one traffic, if not it wouldn't. So these are.

A

Independent, so the reason I mean what I'm trying to get at here was is sort of hinted at with with my name change right, receiving affinity like. I think it would be useful uh because you're making a change based on input, error rate- that's not output, error rate right and you can actually have this in the field where you have one fiber, that's degraded right where the one.

F

That's three to two yeah.

A

Three to two's transmit is fine, but two to three the receive on three is broken right and so there's a real usefulness here of being able to say my receive is, is degraded, but my send isn't right and by by over overloading the reverse affinity you're affecting both directions, whereas in I think it might even be more useful to be able to say to be able to comment on the receive we.

A

We did this um at a previous job right with something else, but you know just uh wanting to degrade things based on received error, so it is useful, but I I'm wondering about it's becoming. You know it is having a bigger effect than maybe it should.

A

It could be fine or grained that that's it.

F

C

I guess I'm up, I I looked at this and I looked at the existing things we had and this uh the reverse metric and the reverse t e metric, and this is a little bit more flexible and then extends it to the well, the flexible algorithm to the flexible algorithm. So you can only uh do the traffic reroute for certain prefixes that are using a uh a flex algorithm. So I thought it would. I thought it.

C

I thought it made sense in in the same in the same vein, as we already had, those two this is just seems like a natural extension to flex algorithm.

A

Yeah, when I said we did it with something else, that's what I was thinking. It was reverse metric.

C

That's for everything on the link within that topology. This is just for the prefixes using the flex that uh yeah.

F

Yeah, so the idea is that you can define flex algo for the specific service, which is this ultra reliable one, and then you can use the flex algo mechanics to achieve the service. You want right.

H

uh Hi peter just a clarification question: uh what is the input error rate mean in this figure? Is that packet loss rate.

F

Well, it can be anything you want, so basically, whatever the three decides to claim to be an error, it's it's really a local medal on three. It can be. A number of uh you know, uh input error, packets received over a certain period of time or can be you know whatever you can even go, and you know get some uh notifications from the line card at the optical level. I whatever it is, if we don't specify what that is, but it can be anything that the implementation decides to track on the receive side.

H

Okay, honestly, thank you.

F

A

Okay, the cube being empty uh seems like uh we could have some more discussion on the list. Some of these comments seems like uh decent work, though some interest in it.

A

A

All right next up.

A

All right zhang are you in a are you in a lab? It sounds like you're right next to an ac.

C

A

We got space he's at a concert.

C

A

Yeah, that audio is pretty bad.

I

Yeah, I think someone is uh there's a background noise right. Hey! You can hear me right. This is louis.

A

Lewis, I can hear you: okay, I'm just gonna cut that there, sorry, okay, um all right! Now: okay, lewis, you're gonna present.

I

Yes, that's right! Okay, thank you.

A

I

A

I should be.

I

Able to control right, okay, let me try, I haven't tried, so this is the first time I tried to present in this forum. uh My name is louis chen, and then uh this is uh I kind of dropped as a it's, a draft at coach with christoph and juniper networks.

I

It talks about how to advertising um offset for fraxel flex algorithm, which is a target for adjacency and also for prefaces as well. So this is the ijeep extensions for that one for both iss and osbf.

I

So the we know that uh in flex, algo uh there's a there's a draft already uh talking about how to advertise uh per flex algo adjacency.

I

Okay, that is the method I mean using. I mean a new label or new index I mean per, I mean for a fractional goal in this case, but then the problem for this one is when they advertise like this way. The size could be quite big, for example, here the function is the um is related, the number of total nodes in the networks and number of flex, algo and number of links.

I

If you have a router, usually make a call router, it may have a few hundred lengths or a thousand links, and you have 10 flex algo if your vertical adjustment seat is on n bytes, which is around 10 bytes right, so it is about 100k bytes. So it's a it's a it's, not a small rgb advertisement.

I

So the question actually can we make the evidence even smaller? So what we try to do is I mean independent number of lengths. That means we try to advertise something so that it reduce independent of the number of links on router. Even your thousand of lengths or ten dollars of link doesn't matter. So the scaling of the whole rgb advancement for the whole network will be totally reduced. So in this case, for for the comparative to our favorite example, it will be only around 100 bytes instead of a 100k okay, because we in development links.

I

So we what we do is just to advertise the offset for each of the adjacent sets for the perfect algo. That means we don't advertise. I mean change the I mean the stick completely we just ever had offset. Instead, so every everyone can be, uh I mean router can use this offset to calculate predict the other uh adjacency from on the other router.

I

So every node calculator says this way, so in this case it's reductions of the 100k to hundreds right. So this way actually make the flex cycle agency is in a more predictable format. The result of this one actually also good for troubleshooting. That means, when you see I mean the agency, you can easily correlate that against it with the which flex algo it belongs to so from troubleshooting perspective. It's also very easy to achieve. So here is a kind of example.

I

I think people knows about this, one in l 0. That means there's a there's, no flex algo. The between the this note, 109 1 1 is advertised at genesis. 32. This 32 is being shared among all flex algo.

I

So when we have fact circle 128 129, when we refer to these tracks this agency, they will all share this one. In this case, that will be a problem when we're actually forming a stack, for example, from this node trying to send packet to vermont six, and then we include this uh 32. Okay. If this is the flux, I'll go zero, it's algo zero. That factor will equal zero and that's fine. But if this is a flex algo, then this 32 cannot indicate which slice I mean, which fractal it will be so become a troubles.

I

So what we try to do is just advertise, something I mean upset, which is uh same level of router 7, routing capabilities uh that portions. It will just advertise a lot of set okay, so, for example, for lgo 129 I will advertise a base. Offset of, for example, label offset is two thousand, so what does it do is which is in next line?

I

This is a label stack? Actually I mean sending from this. One was the similar, the previous one, but this one is using effects l129.

I

So we see that the the prefix I'm sorry the label uh is beginning with 402 instead of 400, so 402 1109 is the this label stack is just like is the first label, and the stack is actually continue with this. This 2032, this 30 uh 2032, actually is the kind of I mean summation. I mean the sum of the algo zero and the offsets. That means it's a 32 plus the 2000, which is the offset advertised by this router 109. Here when I, this router will advertise this offset to the other.

I

Router and 106 can use this information to predict what kind of adjacency actually between 109 and 1 1, and this one becomes 2032 in this case and then, if you, after the router, what's the benefit of this one? Actually, if your router just take a look at the top labels in the stack, they can identify the flex, algo slides and actually apply some. I mean a policy over there right. So this is a purpose of having different adjacency for a different flex algo. So this one is pretty like a predictable way.

I

So, in this case, for example, like effects algo128 or 130, they may becoming 1032 for 130 is 3032, maybe from this one the advertiser upsets. So everyone can predict the other distances actually by this brain, which is simple, so it reduce a lot of the advertisement uh from from from within igp itself and also it can be actually extended to across. I mean different areas, l1 l2 and also can save.

I

That kind of I mean the kind of uh advertisement as well, so the, but then the purpose of this one is you allocate a label in the kind of block in this case, for example, if you uh target to use the, uh for example, zero is 3233.

I

Okay for the adjacency you're going to plan to locate, then, actually you should reserve the. I mean the the labels for 120 1032 103 at the same time, and for one over 129 is 2032 and 20203. In this case, you need to I mean in in your address- is a local kind of a block, a label block. You should reserve like this way so that it won't be used by the other cases. But of course, you have some other method to make it more complicated right.

I

It was again we use this kind of label, but I think this is not all of these discussions, but anyway I mean this is how the way that it will predict the the other labels in this case.

I

At the same time, we will talk. We have another offset right, which is for the prefix. That means you can use the same method to advertise the prefix offset for flexible. That means you have a noxit or prefixes. You can also use the same method to just advertise offsets in this case. The saving of this one is a bit less, but there's other other meanings of this one later on. If people want to ask, I would try to explain what is the reason of this one.

I

The saving element for prefix it offset will not be very big, but there will be other meaning, for example, if I want to expand the flex algorithm, I mean I mean to more than 120, then this one will be useful to you. Have this one actually, so um I think this is my last slide. Any questions on this.

C

Yeah, I do, I guess um this ac lindem cisco systems up what um so it's an offset. The offset is automatically based on the algorithm. What if it's, that would mean that you would have to advertise all the adjacencies in algorithm zero, so that you'd know uh where you're taking the offset from all you couldn't have you couldn't have us you couldn't have algorithm? Zero would have to be a complete superset right.

I

That's right. That means you must have algo zero to be advertised. I think this is normal right. Everyone do that alcohol advertisement right.

C

But you'd have to include every adjacency and everything in it: okay,.

I

That's right: that's right, yeah, correct, correct and then, but uh that you can have the router can optionally uh advertise whether it uh it has this kind of capability or not. I mean the 109s router can advertise 128 129, but it it don't care about 130. Then it don't advertise 130. Then the other router will know that. Okay, you, you will not going to use those kind of. uh I mean I mean offset for effects. Algo130 130 maybe still use the ago zero kind of label in this case right.

I

So that means this is per per flex algo advertisement in this case, but once it's advertised for for that router, it being I mean the global informations right. That means that it means the whole all the links uh I mean within that router should have. That kind of offset. Being I mean added to it in this case.

C

J

C

Put their comments in.

A

Peter europe.

F

Yep, I just have a single, simple command. We already have a white link called the flex algo hc. Now, if you are going to take this, then we need to stop the other one. We do not want to have two ways of encoding. The same thing please so for the working.

I

Group is I mean they, I mean, I understand right, it's it's very logical to use that way, but the problem is advertisement size right. You talk about 100 writers or you know.

F

No, no, I'm not arguing, I'm not arguing your your use case. What I'm saying is that the working group as such needs to decide which way of encoding are we going to standardize. We do not want to have to.

I

ah Okay, that's all sorry right! Yeah, sorry, I mean understood right. Yeah,.

A

Great thanks peter bless.

B

I appreciate the scaling issue and- and you know your attempt to to solve that because, certainly if we have a node with you know a thousand neighbors uh and you want to support multiple flexalgos.

B

um You know that this can take up a lot of space. I'm just wondering: have you vetted this idea with any operators, because it seems to me that this creates a deployment problem for the operators because they have to reserve.

B

You know x, number of sids depending upon how many flex algos they think they might deploy in the future. So today maybe they're only supporting two, but they think in the future. They might support five in order to avoid disruption they're going to have to reserve, uh you know a range of five for each of the adjacencies.

B

um You know. Have you discussed this from a deployment.

I

Yeah, actually not yet right, but but then actually the label space. I mean local labor space. Is it's not big? Actually um I mean here is uh I mean 10 facts about a thousand links? How many labels you have? Can you have me a 10k right? So I mean the the the the the label. Space is not a big problem. Maybe it also adjusts adjacency, so you can change it. I mean I mean you can change it by configuration. Yeah.

B

I'm not talking it's fine.

I

B

Sorry, look I'm not.

I

B

That the you know, the number of labels that you're going to need doesn't change. What you're trying to do is economize on the advertisement uh which, which I think is, is something you know that would be good to do, but it creates a problem for the operator, because now they have to anticipate gee.

B

How many, how many consecutive labels do I have to reserve for each neighbor, given how many flex algos I might deploy in the next five years? You see what I mean.

I

Yeah I understand, but this really depends on the size of the router and also it's because the local label blocks right. So it's only per local router's configurations. It's not.

C

Something I don't want global.

I

Is a problem? Actually, I think, if a global, unless it's a problem, but it's local configurations right so local label right, you can change it.

B

I don't think it changes anyways, I'm just suggesting I'd like to hear input from operators on this.

I

Yes, thanks no yeah, sorry, no yeah, I discussed with them yeah that's right!.

A

Thanks les uh g.

K

Hello, hello, hey hello, is uh I have one comment uh on this document. I think uh here you assume that the offset is like a per algorithm parameter so that it means for each flux, algo the same amount of the adjacent seed as the number of the seeds for the algorithm algorithm. Zero will be reserved in once, uh but actually not every flash algo would require adjacent seed on every link. So this means some of this reserved seeds will be, uh cannot be used for other purpose right.

I

Correct yeah, okay, I mean I can answer it in just two two ways. Right, first of all, if you have hundred flex algo a thousand links, you only consume 100k labels. So it's not a lot. Actually, you have one billion label space right, so I think from that one it's fine, I mean from the local. I mean label space perspective, right, yeah and then the second one.

I

Actually I mean, although says you're correct, actually so when, as I said, when we allocate the label blocks right, for example like this way, okay, I mean supposedly the 30202033 cannot be used. Okay because we served it right, but, as I said, actually it will be. You can use income complicated kind of uh I mean I mean I mean um collect the label back right. So this you know that this one's not gonna use right. 203 is not going to use anyway.

I

You can put it back to your local label space if you want, but this over more complicated kind of label management. Your local routers have to manage this one. This label can be put back into use actually because you're not going to use this one. Actually, the facts are going to use that link right because affinity. You you avoid using that link right, but you need to put it back into your your lower level space. So that is the solution for that one. If you really want to do that.

K

Yeah, but that would be more complex for the uh label. Space management.

I

Yeah but I'll say there's a hundred k.

A

Let's move on to the last person because we're kind of over time.

K

Yeah: okay: let's talk more often.

A

Why claim I don't know if I'm pronouncing that right, quaint.

L

uh Hello, uh this.

E

L

Channel mobile, uh I think from operator point of view. The solution is valid. In fact, we also have a similar draft for srv6. Here I have one question.

L

How much can this solution uh reduce the uh the advertisement to smaller and make the advertisement smaller how much it have well? Thank you.

I

Yes, this is this is already estimated here in the in the apartment statement slice right, so it will be reducing from 100k to 100. That means the order is a 1000 if you have 1000 links right, but it really depends on number of notes in the whole networks and number of links in the whole networks. So it is a function of number of link and notes as well.

I

So if the one case is extreme is one thousand, but if you have an average length in the network, maybe maybe 50 right, then you reduce around 50 in this case, so because this one in the damn number lengths for for the routers. So this one I think that will depend on average length per node.

I

I mean per per, I mean per network, actually, I would say every modeling per node per network. In this case you can estimate that one. So it really depends on the network size.

L

Okay, that's good.

A

All right, thank thanks. Louis uh I'm gonna move on to the next presentation.

I

A

A

Again, uh people interested- we should continue this on the list.

A

Oh dear okay, I don't know why. Oh let's see if it works.

A

Who's presenting.

A

Oh wait: this is not the right presentation. Is it.

A

uh Yeah, I think we skipped over common address so one second sorry.

G

Okay, um I have updated the slides. Can you share the new one.

A

This is broken by interface.

A

Common address here it is.

A

A

About that hello.

E

Can you hear me? Yes, ah uh I'm I'm from huawei. My position is about the comet, just igb friends, algorithm next page.

E

Next night, please.

A

You're controlling.

E

E

I can't see okay currently flash eggos are made are manually identified using different ibmgs in that packet. This is no general approach for different flash angles. Using shared iphones and applications of network slicing increases. The number of flash echo in the network self-development complexity would also increase due to the management and config for path. Less lego, aviation, an interface may belong to multiplayer flash eagle. This problem cannot be slowed by associating supplied interface with different flash angles.

E

This document instills a generic approach to allow different floor cycles to share the iphds common change, algorithm sub.

E

Okay, subscribe is defined to indicate that that's a locator has the ability to be used for positive calculations in multiply command.js flash algorithm.

E

uh Sorry is slow.

E

This page is about the commerce jazz flash auxin protocol. uh It extension uh comment. Just uh augustine. Subtle away is the subject of the easiest load capability uh used to uh indicate which command checks for exagger to know the precise path in uh c flag uh commodities for rise against some flag service to indicate the locator.

E

Has the ability to be used for past correlations in multiplayer communications, flash algorithms.

E

This page is about a common js phrase: echo counterplay an example. Things with different colors belong to different flash angles: 128 129 and 170..

E

Each node describes participation in combat.js, flash orgasm for each flat circle, each node calculations and spf2 independently and generalize. An implant leave and flip are using the same ipsgs.

E

uh In the dead plane, uh they already have a draft as a six-man, enhanced, vpn and vtid has defined a vtn id hope. I hope option to identify the network's rising sources. We can use the vpn option or use the topic id define shaft six month top range id a topology associated with frasergo is encapped. On the head's end, the chance node makes the forwarding table selection based on the publisher id 10-fold package. According to the cross-spending folding table,.

E

And the command chest for us first orgasm benefit is one seam flying flashlight developments and no need that attention. Ipsgs planning each flashlight can share the same ipad js to a reduced protocol packet overhead, no need to advise iphs and srv666 preference ego.

E

Three. It can support a peripheral traffic, cheating.

D

E

The vivian seed has determined which flashlight is selected. Therefore, the english p cannot select different flash flash angles based on the flows and four. The flash ergo failed can be recovered to the basic igp igp top bridge.

E

By uh changes, the total ig top publish id to the to the zero it can switch over the basic igb top range.

E

ah And comments are welcome.

A

Peter go ahead, peter.

F

All right, okay, so uh the problem I see here is that this requires the packet classification on every hub, and that means you need to do that in the hardware. So you know this has been tried with the mtr in the past. It failed. I'm not sure we want to repeat the same mistake here and just one other thing when we did the flex algo design from the very beginning.

F

One of our requirements always was that we didn't want to add any forwarding changes. We basically want to use the existing forwarding plane being a label for srm pls or, being you know ip or ipv6, so this basically makes the flex algo requiring the changes in the forwarding plane, which I don't really am very optimistic about. So that's my comments on this.

A

B

In addition to what peter has said, you cannot use toppo id in the forwarding plane to mark the packets, because topoid is scoped by the protocol. It's not a protocol independent identifier, so, on top of what the concerns peter raised, I I don't think this works thanks.

C

C

uh To peter and less is that you're we're solving a control plane, uh we're we're solving a configuration problem in the control plane by adding extra data plane, separate, uh separate fibs, which seems to be rather excessive. Also, I'm saying if somebody could do this without I mean I don't see why you couldn't do something like this independent of using, if, if you supported multiple topology, why you couldn't do this on top of it running flex, algorithms, any each flex algorithm in a topology and reusing the addresses?

C

I guess you wouldn't do anything to reduce the size of the addresses, but that wasn't explained how to reduce the size wasn't explained very well in the draft anyway. That's my comment.

K

A

K

I'm yeah sorry I'm to interrupt I'm trying to uh help to people to answer some of the comments from the previous participants. The first about peter's uh comments on the classification of package on the national nodes. I think yes, we have tried this with multi topology routing in the past, but I think the the capability of the hardware has changed since then, and as jupiter mentioned in the presentation, there have been uh proposals to introduce some new data plane identifiers, either in ipv6 or the mprs, so that it can.

K

Since we're already considering to change the forwarding plane, there is a possibility to introduce this uh topology identifiers as a one option for the data plane to help to the packet forwarding so that we can save this addresses or seeds and also the control plane resources for the advertisement of this year. Information. So well. Maybe we need to reconsider this, since the hardware has changed since the mtr routing uh design right and regarding less the comments on the topology.

K

uh Actually, uh I'm not sure whether that means that we may need to introduce a different identifier for like a topology and algorithm. Is that if that is the case, maybe you can consider it further offline.

K

Okay, that's all.

A

I I had a question specifically to what last said: two less. If you want to jump back on the mic, um could you click? I I didn't understand too. My my question was about this topo idea. Is this being done in six man? Are they changing the v6 header to change forwarding based on a topo id.

A

Is that your understanding less.

B

I I think the authors could speak to us. I believe they have a draft in six man that that speaks to this um yeah.

K

Here are two drafts: police did.

B

I would make the same comment on that draft. Of course, a topoid is a protocol specific identifier. The ranges for topoids are protocol. Specific ospf supports 127.

B

isaiah supports 1024, there's no requirement that the toppo id for ospf be numerically equivalent to the topoid for isis. If, if the two protocols are operating on the same topology, it's simply not an identifier that can be used for classification of packets.

A

But they okay yeah. I guess I didn't understand that. But but there I mean that draft. It says ietf6 man, so it's been adopted, so they, I guess that's what gu is saying. Is that there's an option now in the v6 forwarding plane to scope addresses based on a vtn id. That's interesting.

A

uh Okay, there's a lot of people in the queue so peter did you want to do a follow-up.

F

Just a quick uh response is that you know the fact that we have the more capable hardware doesn't make the mtr problem any easier, so uh yeah, I don't really buy that argument, but sure you can do it. You can try it and but the problem, I don't understand here. Why are we worried about the number of addresses I mean this is ipv6. We have plenty of them. What's the problem they are trying to solve.

A

I had the same question.

K

A

K

Sorry to interrupt again- actually it's not about the number of the issue is not only about the number of addresses needed it's also about. We need to advertise these addresses in the protocol. Like we'll, like the previous presentation mentioned, this is not one issue. This document tries to solve and there are also other issues. This movement can solve like the coupling between the services and the upper algorithm locator, so that is, we can have the flexible steering and also the reca switch over from one I'll go to the other algorithm.

F

Well that that it's in itself has issues I'm not going to comment on that. That is very dangerous thing to do, but, but I think the solution here is into more problems than the problem that we are trying to solve.

K

Yeah, maybe we can elaborate on what other problems introduce.

A

Yes, we should take. We should take this to the list at this point. um There's some interesting, uh bigger concepts I think to discuss here, uh not not, and I don't want to lose peter's point of that. He said that you know the original design of flex algorithm was specifically not to do something like this right. So the working group should also consider that.

A

Ac, did you have something to put in before we move.

C

A

C

Quick yeah, the the the the vtn uh algorithm in six man, is to use it for a vtn. I don't think you you could just say: okay, let's hijack this and all of a sudden it can be a vtn or a flex algorithm. I mean that would have that that that's then what what? If you want to have, uh there's there's this multiplicity thing: what if you want to have flex algorithms in that vtn?

C

You know, then, then then you're talking about maybe adding multiple identifiers to the packet instead of one I mean you just can't make that extension. Just because there's a draft, that's called a tunnel yeah yeah, just because you're doing a draft for vtn doesn't mean that you could use it for flex algorithm as well.

A

A

Okay, I think my agenda is different. That's why I got confused last time.

A

So next up is exclusive link.

G

Yeah, I'm here.

A

You should have control of the slides.

G

Yes, but sorry those slides has been updated through the email. Could you please share the new one.

G

Because this this draft has been updated, so we update the slides the recording.

G

A

Is this not the newest? Is that what I'm hearing should we do something crazy.

A

Did I screw up.

G

Okay, how about we do the next presentation first and then come back to this one and.

J

Meanwhile, I can upload, in the background, see what we can do.

A

Okay, that sounds great. That's what we'll do.

G

Okay, uh the next one is also b, so you can skip the two to these two.

A

Okay, so up, we gotta update the extensions srv6 sids using sid block and ever exclusive links so on to the service.

C

Side so they said.

A

Who's presenting this.

A

ah And yeah, okay, let me transfer the controls either.

J

uh Hello, everyone: this is museo chen from new h3c technologies. This presentation is on isis and ospf v3 extensions to advertise srv6 service seed. This is a 0-00 version, individual draft.

J

J

Okay, uh as described in ifc5565, there are situations where it is required to establish connectivity between ipv4 islands across an ipv6 core bgp based airstream service is often used to solve this problem. The bss, srv6 services draft defines procedures and messages for bgp based r3 service over srv6, which uses srv6 services to interconnect. Pes and form vpns, for example, router. A and c are both bgp speakers.

J

Router c sends the ipv4 prefixes of islands 2 to router a along with srv6 and dt4c when forwarding the ipv4 traffic from island 1 to island 2 router a encapsulates the payload in an outer ipv6 header, whereas the destination address is c6srv6 seed.

J

Since the packet traverses ipv6 score and reaches router c, the outer ipv6 header will be encapsulated and the ignore ipv4 packet will be sent to island 2.

J

However, for the ipv4 ipv6 core, not deploying bgp, like the campus network using isis or ospf v3, there may also be requirements to interconnect. Ipv4 islands, migrating igp to bgp or or at routers may not be the first choice. So this draft describes an igp method using srv6 service seed to solve this problem.

J

Sorry, it's a little slow. Okay, here is the solution. The procedure is similar with bgp based solution, see edge router advertises, ipv4 prefixes, along with an srv6 service seed in igp. The ingress router encapsulates the ipv4 payload in an outer ipv6 header, whereas the destination address is a ipv.

J

Sorry, ac, srv6 service seed provided by the ingress egress router, the egress router decalculates, the outer ipv6 header, and perform ipv4 table lookup to forward the inner packet.

J

The edge routers must be srv6 enabled, but the transitionals only need to perform plain ipv6 forwarding.

J

Another problem is about the traffic engineering in igp, only networks. In the current shortcut solution, the head-end uses the terminal as a shortcut to the tail end in its local spf tree. However, if only the traffics to prefix one needed to be steeled via the t-tunnel, the traffics to prefix two are forwarded along shortest path.

J

In this case, a shortcut may not work well, as it will steal all traffic to c via the tunnel.

J

In addition, a shortcut will influence the path to downstream nodes which may bring troubles for fine granularity flow control.

J

uh Here's the solution, uh the tail end advertises prefix with an srv6 services and a color.

J

In the example network, router c advertises srv6 prefix of site 1 in igp, along with an srv6 and dt6 seed and color 1, and the prefix of site 2 is advertises advertised with color 2.

J

on router a there are two srv6 policies: the color of site, 1 matches policy, 1, so router a uses, the segment list of policy 1 and the end dt6 seed of router c. To build the path the packet is encapsulated in an outer ipv6 header and an sr header.

J

Then the packet is forwarded along a to d, to e, to f to c router c decalculates, the outer ipv6 header, and sends the inner packet to size one. Likewise, the traffics to set two are still the via srv6 policy. Two and the path is along a2b2c.

J

This slide shows the extensions for isis protocol, a new srv6 service c. The sub trv is defined in the draft. It is used to carry the srv6 server seed when advertising prefixes.

J

When the subtree appears it means the associated prefixes can be accessed via the srv6 service seed.

J

If that seed or its locator is reachable, the receiver may create forwarding entries with relevant instruction as the packet will be encapsulated in the outer ipv6 header. With the destination address being srv6 service seed in the ipv4 islands, over ipv6 core scenarios, it is recommended that the ipv4 prefixes associated with srv6 services are advertised with a magic larger than max metric.

J

The isis color sub trv is also defined. The sub trv indicates that traffic to the associated prefixes should be stilled over the t pass specified by the color.

J

uh This slide shows the extensions for ospf v3 uh protocol. They are similar with the isis extensions, so I just skip.

J

Okay, the next step is to revise the draft. According to feedbacks, we have received some very good comments and a suggestion from our robot. One of the comments is that injecting ipv4 size to ipv6 igp may cause issues.

J

We think uh we need further discussion about that.

J

Any questions and the comments are welcomed. Thank you.

C

I guess I'm in the queue ac lyndon cisco systems. Yeah I mean I, I didn't know that robert had this comment. I guess I missed it, but that was my comment too. You said if it's an ipv6 core, you know in bgp. We have the comment we have. The combination of uh you know. I mean the concept of a vpn address family, where you're just flattening it all out and putting the ipv4 prefixes into ipv6. Now, if you've got an ipv6, only core you're not going to want to do that.

C

So I don't think I don't think this is the right anybody's going to want to use this solution. There's actually there's actually a a draft for ospf a draft, an rfc for ospf, where we support the translation. I know because uh I joined this later- I was a I was an offer.

C

I can't I should have had it before, but where we actually do the the translation where we advertise ipv6, you know or ipv4 addresses in uh ipv6, and then you do it that way and then uh then similar to the rather than a d cap. You do the address translation, but in that case they're not they're. Ipv6 addresses they're, not um ipv5.

C

It's not ipv4 addresses. So that's just my comment.

J

uh Yes, thank you uh very good comments, uh but I think uh in some small campus network it may be still useful and, as I introduced in the slides that uh it is recommended that the ipv4 prefix prefixes are advertised with uh max metric, so it it does not influence spf in in the core.

J

Okay, thank you.

B

I'm just not clear on why we need a new encodings for the igps.

B

It seems to me you're just you just want the igps to advertise the new type of sid that they that hasn't been explicitly specified in the existing specifications.

B

But I don't see why the existing advertisements can't be used to do that.

J

uh Sorry, I I I didn't. I didn't get to your question. uh uh You means uh we, we don't need the uh uh new sub prv to to to do. Do the uh do the questions correct.

K

A

Okay, well, that's something we can explore more on the list. I guess right. You can come back into that if you want. Oh okay,.

C

I'm just going to say real quick that rfc I was talking about that already exists is 69.92.

J

16 9 92.. Okay, thank you I'll raise that draft for details. Thank you.

A

Okay, thanks all right, let's see.

E

C

A

G

A

G

Yeah yeah, thank you uh hi. This is lian and the topic I'm going to talk about is advertising exclusive links for flux, algo in itp.

G

So it's a little slow. Okay, as we all know, flux algo allows atp to compute the constraint against the path and in some scenarios the specific links will be associated with the flex algo, and these links should not affect the defaulted quality.

G

However, according to the current spl calculation rules, these links are also included in normal spf calculation, so the final routine table of the default topology will change and the flows will be misdirected along the specific links.

G

So this draft proposed a new method to exclude the flux algorithm from the normal spf calculation.

G

This slide shows the typical scenario of the problem: flux, algo128 is newly deployed for the network sliced and it is associated with a new sub interface, which is with the red color, as the picture shows, and it is expected that only the particular flow belongs to the network. Slides should be steered along the red links.

G

However, the newly added sub-interface are also contained in the default topology and is used by normal spf calculation, so the flows of the best effort service for the default topology may be still into the red links. If the itp cost of the sub interface is smaller than the main interface.

G

G

Oh sorry, this is the still the the old version. The draft has been updated through the mailing list. Discussion and the updated draft describes the two solutions solution: a use, the exiting existing mechanism about advertised maximum link matrix and in combination with the existing draft.

G

To avoid the flux, algo cannot use the hp metric at its magic type, and this picture this slide shows the second solution. Solution b, it's yours using the protocol extension mechanism define a new subtle way to indicate the link should not be calculated by the default spf calculation.

G

So this slice shows the detail. The format of the isis subtle.

G

So uh I want to also express that about the solution, a the maximum link matrix. uh There is some points to be considered.

G

uh First, I think that for us is because the rfc 5305 has defined that a link of maximum link metric must not be considered during normal spf calculation, so I think it can be applied to these scenarios.

G

It means that the maximum link metric mechanism works well for a, as is the protocol better for ospf protocol. According the current osprc.

G

Link of maximum link metric is irritable, it's not unreachable, similar with the eyes to, as is so. We think that the maximum link metric mechanism does not work well for ospf.

C

I guess so so my comment is that, given that the biggest problem with that last draft is backward compatibility, it doesn't really make sense. I'm trying to do it.

A

Trying to ac what I'm trying to pull up I'm trying to pull up the slides. I folks. This is why you can't send the slides like the morning of the thing right I mean. Oh okay, did you what you wanted to speak to solution a is that right.

G

A

Yeah I'll, let her finish. That's.

G

The newest, yes, thank you. Okay,.

C

Oh okay, okay. I see that the draft has been changed to just my comment or not. The draft draft hasn't been changed. The slides have been changed.

G

So this is, this slice shows the uh some points about uh solution, a so the first one is the sis it works well and the second one. We think it does not work with ospl protocol and the third one when we use the maximum lincoln metric mechanism.

G

Combination of the other solution should be applied.

G

According to the equating draft existing factor, we can define a new metric type yoga metric type so that the associated flux algo can use the icp metric in the past calculation.

G

Okay, I uh I can't control the slides that, could you help to yeah yeah, okay, yeah.

A

Which one would you like this one.

G

This this is okay. Yes, also, this is the solution b about the backwater campus uh compatibility, uh as mentioned essay. Thank you for your suggestion and we updated the draft, so the mechanism was is similar with the existing rfc8024.

G

And use the capability to our way to negotiation and decide which one we should advertise.

G

Next, please, oh that's all for the chapter and uh thank you for all the suggestions through the mailing list and any further comments are welcome.

A

Okay, uh less syrup.

B

I've done my best to try to kill this draft on the list. um The draft was created to solve an alleged problem with flex algo. There is no problem with the flex algo. We have ways to exclude links. We have ways to define algo, specific metrics, that's also being greatly enhanced by the generic metric, that's being proposed in the in the bandwidth constraints draft.

B

There is no problem to be solved for flex algo. The only remaining question is for base algo0.

B

Is there something that needs to be done for ospf, not for isis, because ospf doesn't have the equivalent of max metric in as robust or fashion? That's a completely separate topic. It should be raised as a separate discussion. This draft should be abandoned.

A

um Yeah, you just stated that a few times on the list- and I think the last thing last comment I saw was- should we address this- I I I agree as a working group member uh I this doesn't using plex algorithm or putting it under the guise of flex algorithm or you know I mean if the problem is. We want to exclude links from uh the base forwarding path right, let's solve that problem, but let's not do it under the guise of flex algorithm. That's weird.

G

um So I want to confirm uh if you're agreeing that the eso proposed in the draft does exist.

A

A

Oh, I was just I mean it's. What les said right the issue, the only issue that seems to be there is to be able to exclude links from algorithm zero right from the standard reporting. If that's the issue, then yeah, I think you have agreement that that's a problem worth or at least worth looking at whether it's worth solving or not. That would be a further question.

A

Okay, let's move to the next.

C

A

C

A

Queue, oh oh you're. In the q ac, I thought you were still in there go ahead.

C

Yeah, okay, yeah, I I agree it has nothing to do with flex algorithm itself. If you I mean uh and- and I think isis should be removed because because of backward compatibility issues, even if we were to do something for ospf, there's no sense in doing something for isis just to have a uh an analogous mechanism.

C

The since there's already a mechanism and there's no backward compatibility, leave isis alone for ospf. I think the capability. If, if we were to do this, the capability, we could relax that so it would be up to the um deployment whether or not they want to revert if everybody's not supporting it, because in many deployments, because the max metric discourages use of the link.

C

That's good enough and that's my comment.

A

Okay, uh so sid block- I don't know if this one is gonna, also not be updated. I keep hitting the refresh thing, so I I don't know if it's just not refreshing meeting materials or not.

G

A

We'll give it a try, let's see.

L

Yeah uh uh I mean the queue, uh could I say something chair.

A

uh Yeah quickly.

L

Yeah and as a one of the cause of the draft, I would like to confirm that so the problem for ospf is there right.

A

Yeah, I think we've already covered that right.

L

Yeah, so uh my question is that if that is a problem, should we continue to update the current draft? All we should write another new draft to address or as a pi.

A

Problem well, you've only gotten extreme, extremely forceful feedback from les so far saying abandon the draft. I, as a working group member, I haven't said that, but I agree that I don't think that solving the ospf issue is should be done with that draft with this draft, so the one that was being presented a minute ago.

A

um Other people are going to need to chime in on the list. Maybe please people other than less maybe could throw some commentary on the list, but it does seem odd to solve. You know like yeah anyway. Let's move on to the next presentation: okay, okay, so uh same presenter, right.

G

A

Well, hopefully, this is the updated version. I don't know.

G

Oh, it's not. The update version is.

A

It going to be as problematic as the last one.

G

It's okay! I can use this.

A

G

G

This topic is the is about the isis extension to advertise. I service takes the safe using the block.

G

Sorry, it's a living, slow, wait, a minute.

A

We can see the background slide.

G

Oh, I cannot see oh okay, okay, I can see it and the itf draft has defined the advertisement of the seeds, including and and and each subtle way contains a complete 128 bits.

G

So in some scenarios there may be a huge number of in the unlocks, so a considerable amount of issps will be advertised and we think that is a highly burden for xp protocols. Protocols.

G

This is the detailed description about this problem.

G

It should be noted that this is just a simple example to describe the scenario. Clearly in the updated version, we give a complete complex example, so you can refer it after the meeting.

G

I know I will explain this using this slide.

G

So, as we all know, uh multiple, as we srv6 and access says, we will be associated with the same p2p agency, all the same physical land neighbors. So the number of the n-axis is related to the number of the neighbors and also the number of the flavors.

G

If the flex elbow is applied, a node will advertise flex, algo, specific locator and, along with an axis for every every link.

G

So in this scenarios the nx, the number of the attacks will be increased into a large amount.

G

So we can see this picture uh if uh the flux echo is newly added between a and b for network slicing, there should be two more unaccessed advertised for the newly added subtitles.

G

So when f2 actually deployed much more flex, algo will be used. Turning 20, even 128, and each physical interface will be sliced into multiple sub interface to support best effort rules. So the more I go deployed and the more one router has the links. The more an x will be advertised.

G

For example, if there are, if there are 10 flags elbow and each rotor have about 100 links, the final size will will be about 48 student, ops,.

G

So this draft proposed a method to reduce the lsp number. First, we use the concept of the smart basics, said local block, it's composed of a number of containers within the address range of a locator and the second, the seize ascend in srv6, the seedlocal block is represented by its index in the block and the third. The off-site value of the index is used to determine the production of the state for the index.

G

Finally, instead of advertising, the complete 128 bit speed and 16 bit index value is advertised in igp.

G

This is also an simple example of the solution. From this slides, the thing the seed can be replaced by local block and indexed with the offsite.

G

This lesson shows the detail format of the new subtle way. First, is the local block slab tailway? It is carried in the srv6 locator tla.

G

It contains the block id and the status the understatement and the other is the indexed subject of an x and it contains the block id it refers to and the index also associated with any point. Behavior.

G

um This is the the format of the land x and the understated is similar with the and the extra seed. So I will skip it.

G

Take this simple scene narrow as I example this slide- shows the detailed tray of roots advertised for the top. This topology and first, the two different local blocks battery are included in different locator tlv and the second separate index value is used to represent an exit separately during the under x index. Subject.

G

This is the comparation of the tla size after the optimization, the size will be reduced to about one third.

G

So the overall optimization can reach one third. This means that, as the number of the and the x increases, the comparation will be more obvious.

G

Next step of this chapter first, the terminology see the block may be confused with the locator block, so here uh we would appreciate and use the dry stone for by turning and welcome and the other for the review and the feedback.

G

That's all for this draft. Thank you.

A

F

All right, I have one comment here which is similar to what I said for the other job, but here it's even more important. We already have the are it's almost an rfc srv6 draft is, in the rfc editor queue just waiting for playing some of the preferences.

F

You already have standardized way to advertise, uh you know and exceeds, or whatever type of adjacency are we going to standardize another method? Well, I would add, or not.

G

Oh okay, I understand that we think that we can use this solution as a supplement for the current to advertise.

G

Our current draft.

F

Well, we never have two ways of encoding the same information, so unless this is really really really then I would rather not go this past to have two different ways of encoding. The same information just my opinion. I leave other people to speak. Thank you.

G

Okay, thank you.

A

C

More on the line, peter said I was just to say as working group group chair and given the number of uh deployments.

C

We would not do this in either the either of these two mechanisms for uh reducing the size of the advertisements. We wouldn't do either of them instead of the deploy um deployed solutions, because even even I don't even I I mean that's particularly true for srv6- and I- and maybe this isn't, even though we haven't published flex algorithm. Yet I mean there's multiple vendors that have been uh have implemented it and have uh you know, have have have published uh interop and whatnot or press releases.

C

So I don't. I don't think I don't think we. I don't think we have either of the any of these mechanisms to reduce the size that would completely supplant. So any documents going to have to really address backward compatibility.

G

Yes, thank you and I am very understand your concerns, and here we prompt this problem because it's that's very. uh It is very seriously when the srv stakes deployed and that lsp number is really a great burden for the idp protocol. So we want to resolve it and.

G

About the solution, I think uh this is the first version of the draft, and uh maybe it's not the perfect, and uh we will leave that through the discussion. We can find the best way to resolve it better. The problem we still think is very, very necessary to resolve it.

A

Is that sorry? Is that really highlighted in your draft like because the the feedback, I'm hearing so far, that you've gotten is that this is not critical, but it's pretty disruptive.

A

So if you do have a you know, if you have use cases that show that the solution that we have that's uh sitting in the editor is not sufficient. I think you're going to have to make that case pretty strongly.

G

Oh, as the slice shows in the new latest version, we can see that even there is no flux algo. The problem is also very seriously because the number of else p is a greater number.

G

You can refer to it and discuss it. Offline.

A

Do you mean the lsp and isis the number of lsds.

G

Yeah yeah the, as I also.

A

Maybe this is a good. Maybe this is a good uh time to do our extension anyway, we play we've played around with making that bigger too. In the past, the number of lsp segments uh allowed uh all.

L

A

L

Okay, yeah, thank you uh so as of course I I would like it to have some clarification. So, first of all, I think it. I totally agree with you that it's a good time to have a looser extinction, because most of uh operators who have not deployed srv6.

L

So if we can optimize the solution this moment, that would help for those operators who can deploy the optimize the solution at the first time, uh the second one uh I I think it currently I I'm from china mobile. In fact, we might a critical issue uh when we hope to deploy the flight cycle uh with srv6 uh as uh discussed the mprs as our draft just now. In fact, we have already hoped to deploy the solution in one thousand more than one thousand nodes network.

L

So uh the conclusion is that uh it cannot work because the advantage, the advertisement.

L

Load is too high so that that doesn't work. It's not the performance. uh It's a the. uh It can cause a critical issue so that uh the source it doesn't work. So I think.

A

So I hear your point: I'm I'm curious where where was this discussion when we were doing the srv6 document in the working group, I mean we got it all the way to the outer queue and now you're saying it's undeployable.

L

uh So uh I think it currently that the prime teacher, the station, is uh not the latest one if you can download the latest one.

L

If you consider a network with 100 nodes, each nodes have 100 links and each link supports two flavors. You can calculate uh yeah.

A

No, I understand I'm just I'm asking I mean the srv6 stuff. That's just about to get standardized has been worked on for a while. Why why I just? Why do we have a new solution showing up now? You know like why wasn't this work? This is a working group document for a while right. We should why weren't we addressing this problem earlier. It seems like a failure.

A

C

Yeah I'll just go back to what I said: we're not going to pull isis off the rfcq, so any solution has to address back the srv6 or any solution is going to have to address backward compatibility.

L

Yeah, so I think we should consider the backward com compatibility, so if you don't deploy the i76 and the flight cycle at the same time, I think that that's okay, but if you want to deploy the both of them in a large network at the same time you will might you will meet a critical problem, so I think problem there. We should solve that. Of course we should consider the back uh forward compatible, but we have to solve the issue to go forward right.

A

Okay, uh peter europe and then, let's move on to the last presentation.

F

I guess I just want to mention what you started chris is that we have rfc 7356, which solves this problem. Maybe it's time to use it.

A

Is that the extended lsp yeah yeah cool sounds good to me?

A

Okay, let's get to the last presentation.

A

G

Okay, thank you.

A

Oh this is this is a funny one, I'm sorry I pushed off to do it. Okay,.

D

Hello, everyone hi.

A

D

Can you hear me.

A

D

A

Working group yeah.

D

Oh talking about that basics, plus equal protection.

D

Let's have an overview of sr eagles protection, so, for example, here we have sr pass from english, node p1 to equals note pen pa, has a locator and then has a seed. A this city may be a vpn seed to provide protection for pa.

D

We select a bike, hop egress, node peb on bike hub equals not peb.

D

We configure mirror seeds to protect pea. After this configuration by cap equations, node pb will create a folding table for p. A this forwarding table will contain an entry for a seat such as seed a so. This foreign entry will forward the package with cda to the same destination such as c2 as pa.

D

After this configuration, the igp on bike club e-western pb will advertise the information for equals protection. This information contains new seed and pea, which will be protected.

D

This information will be added advertised to evernote in the in the area. So after p1 as a prr receiving this information, p1 will compute a bicon path from p1 to bicarb equation. Node peb, when p1 detects the failure of pa p1 will incap the package to pa with mirror id and send this package to bicap equation peb when bicarbonate equals peb receives this package.

D

Peb will decap the package using the mirror id to get the forwarding table for pa and then using the seed a to get the forward entry in this forwarding table and then for the package with the a to the same destination c2 as pea.

D

So, let's look at the extensions to support supported protection, extensions to isis.

D

Basically, we define a mirror system trv in the existing plv.

D

This sub trv contains mirror c and then the node to be protected, for example, in where this subtype of v will contain the c mirror street configuring backup equation of pb, and then the pa pa is very eager to be protected.

D

This pa is indicated in a sub trv. We call the protected locator node subtle, v.

D

Optionally, we have another separate subatomic forces.

D

So the extension to ospf is similar with the final ocpf mirror see the subject of e, so this mirror separate v will contain your seed and the node to be protected. This node to be protected is contained in protected node, subtle, v.

D

And then optional subtle v for system protected.

D

And then I would like to receive comments from everyone.

A

So this is, um this: work is being done in the routing working group right right. Yes,.

D

Yes, these are represent many times in a routine working group and then we would get us comments from the lsr working group because we have extensions into igp to support the equal protection.

A

Right yeah, I just wanted to call that out to everyone. This is sort of an exposure thing, we're exposing you to work going on in rotting, working group that relates to lsr.

A

I

Yes, the question: actually, why I mean you can use any cards to achieve the similar thing, but of course you, if you're any cast, then there's under complications, you might need slt. I mean some kind of a tea for some protection so that it can swing. I mean primary and secondary so choose a primary secondary right, but any cars can solve this trouble as well. But why not consider any cars, which is I mean, do not change any floating plane in at all of a control plane at all yeah.

D

Yes, that's a good question uh for equal production, the different ways. One way is you mentioned that using any task and remember that for the solution using any task, there are some issues and also those issues has been raised in the in a working group of meetings.

C

Okay, this is ace ac, I'll, go real. Quick, I see, uh lewis is off the off the queue uh you got my comments, the one that I think it would be good for people to hear is why you need both of the sub tlvs now granted. If you only had one, you once said.

D

That's right, your favorite good question coming in to rename.

C

D

The one the second one is optional: I'm considering removing that one that one will make things either you're right.

C

Okay, thanks and chris, you said this was the last there's still one more presentation.

A

I I screwed up and went out of order: um okay, okay, we're jumping back to the net extensions; okay,.

C

A

A

Your song you're out.

H

Okay, hi uh stressing from huawei. I will give a very quick introduction about our work on uh isst extensions for uh enhance.net.

H

uh This slide is just twice uh just twice: to provide some background for this work, as defined in rfc uh 8 86 55 danette is designed for applications which request extra high quality network transmission.

H

The performance provided by that net cues could be expressed as three aspects.

H

The first one is boundary latency, which means the minimum minimum and maximum end to end latency from source to destination could be expected and the second one is extra low packet loss ratio under different optional states and third, one is upper bound on out of order packet delivery, which could be re-altered by the technologies introduced by the net, and so three technologies are used by dyna to provide these qualities of services, including resource allocation service protection and explicit routines and now internet a working group.

H

Then enhancement is under discussion uh which make people pay more attention to introduce new requirements and solutions to provide boundary latency internet. So this document um we just try try to introduce the necessary attributes for dynamic boundary latency. Besides the t metric that have already been defined in rfc 53.

H

53 505, if I remember it's right and rc, 78 and- and here is our proposal- uh some attributes uh new attributes are introduced, including node attribute tlvs for the net, including that net processing delay subtle, v and also link attribute tlvs for enhanced net, including maximum uh data, reservable bandwidth, the net available bandwidth and also the net time results subtle. V.

H

All these attributes will be used to calculate a path which could satisfy the delay and resource requirement of a dent flow which has been introduced in the previous slide, and here is then at processing delay subtle, some background about this internet. We think perhaps delay could be divided into three parts. uh The first one is link delay. Second, one is uh processing delay and third, one is queuing delay in rfc, 78, 10 minimum and the maximum link delay. Subtle has already been defined and the queuing delay is calculated based on the queuing mechanisms.

H

So this document just defines the missing part. Minimum and maximum processing delay subtly, and also link attribute lvs for enhancement.net, including uh maximum and meat and available. Then at reservable bandwidth, subtle, uh actually, uh maximum link bandwidth and the maximum reservable link. Bandwidth. Subtle, we have already been defined, rc uh 5305.

H

This document just defines the bandwidth that could could not be shared by other traffic engineering flows, so new subtleties are introduced here and also because then that request uh boundary latency. So we think a time dimension of results should be considered in link attribute lv. So we we defined then at time results subtly, because there are a lot of different methods of implementation, solutions about how to provided boundary latency.

H

So we try to just abstract the the common attributes requested by different methods and dividing them into three or two types. The first one is time results for logical, cues, subtle, v and the second one is time resource for time, scheduling, subtlety.

H

Basically, the difference of these two parses, the first one is um when the underlying technology is logical: q, based on scheduling mechanisms, their request, a branch of attributes and another case is we introduced some uh time scheduling mechanisms, some some corresponding attributes will be introduced, and that is the existing uh work we have done and we request comments from our working group and if people are interested in this work, we could cooperate and work together. Thanks comments, please.

A

Hi, uh so speaking as a chair hat on, I spoke with uh the chair of dutnet um and he he didn't think this had been presented there and to me it seems like you know this isn't just um I mean you do refer to to some of the dot nut rfcs, but there's some new stuff here right and I think you need to you know it needs to be presented there and the concepts need to be agreed upon in dotnet, so that really, when you're coming to lsr, we can refer back to you know: ietf agreed upon last call so forth and so on.

A

um You know, concepts that you're bound into things and your cue measurements and all this has been agreed upon as the right way to do it right the right way to model these things that that has to be done in dutnet. I think, and then the work in lsr is really just reduced to. We look at what's been agreed upon in dutnet, and we say okay to deliver, that we use these sub glbs, but it seems like what you've got here.

A

Are some sub kill these that are also creating new concepts right and I I don't think you can. I don't think it's right to do that in lsr,.

H

Yes understood, I think this is uh reasonable. We will bring this work back to that net, of course, uh because we want to introduce it in our first days, because it is very related to the previous work that have already been defined in traffic engineering metric extensions in lsr.

H

So I think maybe people will be interested in this work when it's still in in its maybe a start point so just to collect more uh attention here if uh to see if people are injured, interested or rather think whether this is reasonable yeah. But uh of course we will go back to that net. First.

A

K

A

Less you're out.

B

Chris, I agree with you, but I really want to expand on on the point you're making, because it seems to me that I I haven't spent all you know a lot of time looking at that net documents, but I have looked at them somewhat and I do not see that.net has def has a defined role for the igps in a net network, and yet here we are jumping ahead and going. Oh, let's use the igps to flood a bunch of link, attribute information that that may be useful to detnet.

B

um I really think we need to have a much better liaison between the two working groups and come to some agreement on if there is a role for the igps in debt net and what that should be, and I'm kind of hoping that you, the chairs and and john who, I believe happens to be a d for both working groups- can facilitate that.

B

That's my comment.

H

Yeah sure, because uh we also were uh in that net, we are working on a control plan framework document and in that document that will introduce the you know, general requirement for uh control plan protocols, so maybe igp will be also included. I think that is another opportunity to connect these two working groups and yeah.

B

Which, which I mean.

H

C

B

Yeah sorry, it reinforces the fact that this draft is very premature.

H

uh Yes, yes, uh I agree because yeah.

A

Although perhaps it's it's done its job and got the attention of less than others,.

H

A

That's right, that's great! Please.

H

A

Okay, uh okay, oh I'm done, then I think lessons done ac euro.

C

Yeah, the only comment I had is I haven't been following: debt met net much at all, but the um the unit seemed a little bit strange to having it uh bytes per second and a three and a three octet field. That's not consistent with igps at all, and I'm wondering where that came from. You don't have to answer that. Now you can answer it once you get the question of uh vetted and vet deck net and established that it needs to be at this that net information needs to be advertised in igp's.

C

But that's something you should think about too is. It seems pretty strange to do bytes per second and only a three byte metric, because that.

A

That only allows for only 160 bits.

C

Per second, or no not hundred and eight times, 24 meg, whatever that is so.

A

Something like that ac, it's not enough. You're right is that.

K

Okay, what I mean it's.

A

Bits and bytes right, so all you have to do is is say: it's a 32 bits per second field, which means you're at four gig right. We have 400 800, gig links coming.

A

It's like a couple of magnitudes, too small.

H

uh You're right you're right well, I think this could be considered in the next version. We will consider the whether the the field is reasonable enough.

A

Yeah, just just on time,.

H

Thank you. Thank you. Thanks.

A

Yeah we did it's noon well, somewhere, yeah 12.,.

C

Okay, so we got a lot to discuss on the list. We won't keep it people here beyond beyond the end time, uh thanks everybody for uh participating and thanks to all the presenters for uh preparing their slides and presenting today, I'm glad we could catch up on all these uh drafts that we didn't get to in 114.

A

Yeah, okay and hopefully see a lot of you in london. It's nice to be seeing people in person again.

A

All right, everyone thanks thanks.

C

A

Hank, I hope you kept good notes.

A