Internet Engineering Task Force 102, 16 Jul 2018

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF102-LSR-20180716-0930

Description

LSR meeting session at IETF102
2018/07/16 0930

https://datatracker.ietf.org/meeting/102/proceedings/

A

Or get served maybe.

B

A

See if this works, if.

B

You're gonna, if.

C

You're gonna sit.

A

A

Let's see if you see this, yes,.

D

A

E

A

More trouble with Jabbar clients, I'm tried two minutes, you don't see it. The.

D

Only person I can talk to.

A

What we're leaving and you.

F

G

Mean I don't see anybody else.

A

G

huh And you're just dead Elessar Jabbar, no idea.

B

A

H

Also, have this yeah.

A

I

Been disconnected.

H

Okay, I think it's are being renewed.

D

G

I

H

So, do we have a jabber scribe, or is that can be us.

I

H

John's doing it.

H

No, no youyou you're it and we have a okay and.

H

The blue sheets are being passed around, so here's the note well so, no well I guess we were supposed to read the bulb so by participating IETF. You agree to the following process and policies you have to make aware of any IPR.

H

You may be recorded, pictures of you may be taken.

H

And you agree to work respectfully with others. That's the important one!

H

Okay, so agenda bashing. Does anybody have a problem with these jinda? Add something? Yes, a little tiny.

E

H

Me and a/c haven't: we haven't agreed upon the common. He wants me to use PowerPoint and haven't agreed to do that yet so we have two different slide decks for her status. I'll start with the is is so since London we have one new RFC. It was a bit update and we've sent two documents to the iesg.

H

The reverse metric in the second MSD and the we're ready Goom was shepherding the segment routing extensions. Apparently this is. He was waiting on one final author to do. The IPR clearance I said that if he should ask all the author's to get a hold of that author and if not, then we've got them moving to the contributor section, but anyway we should that's processed food that we should have cleared up real soon. Now.

H

Yeah and then we have a one in last, call.

H

Is that 7810 is that the restart.

J

So Alaskans were okay. Ten is not a new RLC yeah. It's the business trying.

E

J

Move forward and I believe we're past the t mark we passed last call. No, that's.

H

Almost completed yeah well I mean until it gets the button gets pushed right. It's insane.

J

I'm, not quibbling, I, just I would.

H

J

H

It's not it's not waiting on anything yeah.

J

Yeah I would really like to get that completed as fast as possible because it's just a tiny but important correction, yep.

H

Yeah that one actually showed up in working group status as a new document kids went from so I think we did pretty good on the speed. I went from Newt last call and one meeting okay.

H

So we have some working group documents. We have a new, the LSR flex. Algorithm was published. Last time we have update, drives iose that will be talked about later. The blue ones are talked about.

H

Yeah and then we have the desk source routing which justice was about to expire. There was some concern that no one was going to work on this right, and so, if, if you care about this right, I think that you need to get some more talk going on the lists and maybe move it forward. A little bit. I.

G

Don't see is David Lam Parker in the room. I haven't seen him. Okay,.

H

Fred Baker.

H

H

And then this is just the reason to talk about these, because these are all the individual documents they're going to be protected.

E

G

And then this is yeah.

G

Okay, I'm just gonna, go through these real quickly. How come it's not well hi I'm, holding it upside down, okay, okay, so we have two new IFC's. The OSPF beats re-extended LSAs at long last. So all our extensions now should be able to cover both OSPF v2 and v3.

G

We also have the shudras I mean she's, not the only offer, but the OSPF link over load, which also has some identifiers in it for disambiguating parallel links both numbered and unnumbered.

G

On the we still have these on our s, CQ that top one I don't know about it's waiting for IDR tunneling in caps I, don't probably it would have been better just to use our own registry, but it's too late now, I guess, because it really doesn't share much with IDR, except the registry always be a segment routing. There's a couple drafts in spring that are close to being I.

G

Think they both gone through working group last call and are going through. The discussions and beer is in, is being edited. That was a beer document, but it was reviewed. We reviewed it here as well and last called it missus working group. We have these to the maximum segment def, that's Jeff and L and link-local interface, ID, link, local signaling, that's Peter and others.

G

Both of those are we're waiting on Alvaro I. Don't think he's here, I think he's still celebrating from the football game yesterday because he was in he was actually in Russia for the games.

G

These drafts are close, we're gonna work called OSPF young model. Shortly thereafter, shortly after this meeting, we had one last meeting on it. I'll let ye and talk to it. Next, OSPF, III, segment, routing I, think we're gonna have an update today, but I think we're pretty much ready for that. Not an update today, that's wrong, but that's that that is ready to last call as well, and these two are stalled. The first one is stalled. I think we should have working group glass called this one, but I had it.

G

We had a breakdown, a communication with the offer or any offers to this. One here, I know Alberto's in office an offer on this one. One of the the problem is: is it contains an implementation specific part in the normative normative text that I'd like to see removed and have it be reduced to not saying exactly how you use it, because I think it's kind of a very implement like I, said implementation, specific specific, so, hopefully I'll contact the offers and see what they want to do and then the H bid support.

G

Are there any offers here?

G

Okay is here yes, I I'd, like I, where I'm just waiting for them to take my comments prior to working group last call and then, while we take those comments and submit another version, then we're working request call them.

G

Now we will have a we have a couple active working group I mean a few active working group documents. We have te attribute reuse. We have a update on that today from Peter SR yang again, the priority is the regular yang model and this document will probably be able to go forward pretty soon, because the foundation or the draft that it's based on in MPLS is through working group last call and is taking some level of comments. So this I'd expect to working group last call this soon and that's it in exam I'll. Let you.

K

Good morning, everyone, my name's engine I'm here this is an update of body. I said our eyes are related young mothers, and so here is the list of people who have been actively contributing to all these models. So I'd like to take this opportunity for their to thank them for their contributions and also sense for people who have provided us with valuable comments and reviewed. The document.

K

So for OSPF young model since last IDF. The here is the list of changes with us since last IDF and the major work we have been working on the harmonization of between OSPF and the ISS data model. So for this one we are really in the stage of finishing it up, and if you want to provide some comments about this one or some any feedback, this is your opportunity.

K

We've done other addressed all the young doctor review comments and we are going to have far more revision of this document right after this idea and we are going to start the last conversation of this work.

K

As I see our model, so this one we, although this one we haven't, started the yundt out to review yet, but because of the similarity of this model and the OSPF model, we took the comments we got from OSPF. We all talked to review and fix everything here in the SS yamato.

K

So um we also did a lot of editorial changes clean up after this IDF we like to request a young doctor reveal formally and then hopefully it will go through quick, and then we threw the last cup o SP f on diocese sr young model. um We do have a protocol independent segmenting model, that's in spring working group. These are the protocol specific part, um because we do have dependency on that. They face young model. Also, we have dependency on the protocol extensions, so this one will progress after those documents published.

K

That's it any questions so.

H

I don't really talk to any see about this, but I I'd like to progress these together, like so.

D

G

Only one reason this these documents are so large, I think progressing together. I'd say consecutively, maybe good idea, but but if you send them 150 pages of yang and then you send them three well more like 120 and then you send them 240 I. Think 120 is preferable. So the reason I want to send them together.

H

Is because if they, if something is found in one I'd, rather have it fixing them both right like so we don't, we don't, publish one and then later publish one, that's better. So.

L

Jeff I was offering a bit of advice from bf deal and we're trying to advance our yang module things get extremely pedantic at the iesg level, and it's better if you actually have one of them, go through that process and get those resolved before you actually try to no role in the other document. Otherwise, you're gonna get a lot of free. It's called um unusual comments that it's easier to resolve once and just say. Oh you saw this already and then move it neck support. Okay,.

K

So, although we are seems like that sequentially, but we are.

C

Really doing working.

K

On them together, so other young thoughts or comments that we got from OSPF or regular dressing, I, say right.

H

I mean I think my main concern was like the way that PFD didn't get integrated in eye size until like a whole, other meeting happened right, so I didn't want to see the same extra meeting added to the process.

M

E

H

You know who's up next. Is it Peter? Yes,.

D

It's two or three.

G

C

D

H

N

H

Here is this there's one you want to start with. You have a thing there. Yeah I have to make.

G

You know how to work it. What's this, this is our PowerPoint yeah yeah view. Is it's not a Mac noise it? No. This is like a present. Try.

D

That yeah, okay, what's this accomplice, yeah.

H

It's a Chromebook fancy new, alright.

O

So quick update on the link ativy to use draft.

H

Oh, it may be, it doesn't work at the PowerPoint. Okay.

D

H

Me, what is it doing? Okay, all.

O

Right, so a major change that we introduced in the latest update is that we edit the OSP, obesity and coding stories. So we have the OSPF. We see entirety LSA, which is been defined in the old RFC. Now we have the extended router LSA for that has been defined for v3. So, basically, what we are doing is we are putting the Lully app specific link, attributes inside this extended LSA. We use the registry which has been defined for this. To allocate the code points, it's exactly.

O

The same approach will be used for SPF e 2, so we have our original TSA, which is used to propagate the attributes for TE, and then we have the other applications for which we used the new extended router side.

O

One other change that we did. We changed the name of the TLB because we have extended link. Let's say we had extended link TLB and then we had exiting attribute sub theory, which has been a little bit of more too much of the extended stuff. So we call this an application-specific link, attributes sub theory, so we just renamed it and now it's becoming subtly of either the USP v2 extended, link, TLD or OS BBC router link, tle itself.

O

The other changes that we made the maximum link bandwidth as an application independent attribute, and we moved it out of this sloth of TLV. Basically, the rationale is that, because this is application independent, it doesn't make much sense to advertise it as an application specific attribute and you know, introduce all the or advertise with all the bits set. Instead, we basically say we put it. You know the top-level TLB saying you know this is completely independent.

O

So now we can advertise this inside the extended link, TLV USP v2 or in a router LS a router link, TLB noise PVC. If you use the same forward as has been defined earlier, just put it there in this new containers,.

O

We also added the local and remote ipv6 address sub TLV, so you can advertise these two addresses for other purposes than te, so we could put it in Nice PVC and you know extended router, let's say all turning TLV like this, so these has been the changes. I mean the document has been there for quite some time. We have presented it several times.

C

O

Has been no major changes, I mean the OS PVC and coatings are changed, but there's nothing. You know changing what we do. I think it should be ready by my next idea for Alaska.

O

O

Okay, so this is an update on the Flex algorithms.

O

So what happened from the last IDF? This has been accepted as a working group document, we also merge. The OSPF and ice is like salgo of stuff into this single draft, so it covers both OSPF OSTP to a nice PVC, and there has been some early on allocations done for OSP, routing information, TLV and also for the eye size, article ability TLB to advertise d, the flexible definition. Okay, next slide, please so.

O

As I said now, we are covering all three protocols. The other important change has been that we originally defined this to be really only used by a sorry, but there has been a feedback from the group on the last IDF to make this and usable for other applications. So now we basically allowed this flex algo to be used by any application by, but we need to be careful here because three, the handling of each of the application or how the application is using this needs to be defined for each of the application independently.

O

What we are defining in this draft is how we are using this flex. I'll go inside the SR for other applications. They have to define this elsewhere. We are not defining it here. There has been a lot of editorial improvement, there's a lot of good comments coming from Eric, Rosen and Tony. Thanks for that, so we improved this significantly. That has been restructured. So it's much easier to read at the moment next slide, please!

O

So we changed some of the names of the fields, the TLB itself didn't change, but we renamed the algorithm field to be in Flex algorithm field. Basically, this is the value from hundred twenty eight to 255, which basically is a numeric representation of the of the calculation type metric type and a set of constraint.

O

So this is all being encoded as representing these three things in a single value and the calculation type itself, the algo type has been really renamed to calculation type, because this is what we want and the value comes from the zero 227. It's a standardized value from the IGP algorithms types, but we say if the algorithm time defines anything else than the calculation type. We only take the calculation type, because the algorithm type can specify some other stuff than just the calculation time.

O

So we could I mean that was the intention originally, but we call if I didn't text.

O

There is one other thing which has been missed in the last version is that basically we specified that if the Flex alga definition has something in it which the router doesn't support. Basically, the router has to remove the participation from the algorithm because, basically it can't compute, so this has been specified in the text.

O

And we also specified on on or put more details about the node participation. So what we say is that the participation in the Flex algum must be algorithm specific, because now we want to use it on for different applications and the reason why we need that, because if certain applications want to use it, it needs to, you know, install the forwarding entries so for segment routing. We are basically advertising the the participation based on the segment routing extensions which has been defined for in the icpsr extensions.

O

It is a topology independent advertisement, so basically it means if I'm participating in a certain algorithm. It applies to all the topologies I participate for the other applications. Again, it needs to be specified for their applications. In some other graphs. We are not specifying it here. It can be signaled on a pack. Topology can be topology independent, it's up to the application, how the application no signals. The participation.

O

The calculation itself again, originally, we only were talking about how to calculate for a segment routing here. Basically, we say: what do we do for the segment routing? Also, the other applications have to specify you know. How are they going to use? I can calculate the pass. One thing which needs to be considered is: what do we do with the nodes which are not participating in for segment routing, what we say, the nodes which are not participating in a certain algorithm. They are not going to be used in the computation.

O

The other application may decide to do the same or may decide to do something else again, it's up to them to define their behavior.

O

And the last thing, obviously, the forwarding we specified that how do we use the segment? The Flex nagas forwarding in the context of the segment routing vary. We used the Flex alga specific seats.

C

O

Mpls forwarding we specify that if there is.

C

O

Flexin alcoci, we don't install any forwarding entry and if there is no forwarding entry, the traffic which is supposed to be for what it must be dropped. This is how we want to use it in the context of the SR there are, there are other applications again they can specify what they want to do and if they want to do any kind of a fallback to a default algo, it's up to them again. It's not in this draft.

O

So we are waiting for more feedback. We would like to continue to evolve the job we have some implementations available. So there is work going on this a.

G

Synonym systems Peter, if somebody were couldn't somebody I mean it- might be unwieldy, a bit like not via. If you remember that, couldn't you define a separate loop back for each service, and this would work with that. If somebody wanted to define all the details or is that something that'd be hard to do, I.

O

Lubic, you mean specific address: yes, okay, yeah, yeah, okay, but.

G

The intention here is, you, don't have to exactly right, but that that would be one thing that could, if somebody really wanted to do that on base side, sure, yes, and that would be described in a separate draft.

G

Probably yes, okay,.

G

Any other questions, feedback speaking is working real chair, I wasn't asking for this strap, but just hypothetical.

O

Okay, so the next one is, the ice is a sorry, six extension for ice ice there's a lot of contributors on top of the outers on the graph. So we put the names here. Okay, next slide: please, and there isn't, there is an OSP of e3 equivalent. It hasn't been updated yet based on what we did in the ice ice. But that's next thing to do so. You have a nice view of history document which is going to follow the same thing that we didn't I, sighs.

O

So the major changes from the last version, so we used to have the notes it TLB. We have removed that and we replace it with the SR v6.

O

What we call a sorry, six locator T le the end seats are now being advertised as a sub theory of this sorry, six locator Tinelli, all the MSD sub t always has been or the seat depth is now they are being advertised as using the MS D sub T of e, which is defined in a mention draft, and we added the support for topologies and algorithms in the English version. So this is high level and then I'm gonna go over the details.

O

So in the previous version, I mean that there wasn't any really association with the algorithm for a seat. So basically, if we look at what the CDC is just a 28 bit value, it has certain structure and basically locator and a function function code. So the locators are the entities that are algorithm and topology specific.

O

So, as are the seats in the MPLS, each locator, basically covering old and all the seats which are advertised from the node- and this basically means that the locators are the entities which are installed in the forwarding of the transit routers. The seats are not so. Basically the routing entities are the locators, and these locators will be topology and algorithm specific next likely.

O

So here's an example: you have the zero basic topology based apology based algorithm, you define the locator, which is a slash 64, and then you define your seats, various functionalities as 128, and then you have another empty id0 algorithm on a 28. You do in a different locator and then you define your seats, so these seats are basically covered by the locators itself. Those.

C

O

Are advertised and installing the forwarding.

O

So the e36 located here Lee is the TLD that we use to advertise this I'll talk of topology and algorithm specific locators. They are ignored by the legacy routers. Obviously we do install the forwarding entries based on these advertisements, the locators- if the note obviously is supporting si v6 for that specific topology, an algorithm has to install the forwarding entry based on these advertisements, the locators.

C

O

Could also be the locators could also be advertised as a prefixed reachability T in inside the prefix h ability T. Always that means that the legacy routers could actually install them and and release them in the forwarding as well. So basically, this is used for a backward compatibility. Obviously it is optional. It's up to the user to decide what he wants to do.

G

Acm Cisco Systems, so in that case a router that did support this, wouldn't wouldn't ignore that and install it would use the other advertisement I, guess it wouldn't matter, because it would results in the same forwarding plane for a transit router anyway. It.

O

Will I mean the router which understands will see this advertisement coming is in the two different Elise? Yes, and we say that we prefer the prefix reach ability advertisement from TS r, v6 locator consonant.

O

C

O

As I show up very.

C

O

Honda on the Honda fixture, so this is how it looks like I seen you can only can look at it. I mean it has the it has the the locator itself which size it has the algorithm encoded in it. We have Flags, there's one flag which we define it's to indicate that it was leaked, because this is something that we can leak between the between the areas and there are sub D, always optional, video haven't defined any, but just for the future we defined a sub tle space. There.

O

So the end seats up the earth. Actually we did the DN C sub theory is a sub theory of this, so the N side is the equivalent of the notes in ultra fixing in the MPLS kind of it inherits the topology and algorithm from the locator advertisement. It is not associated with any neighbor, it must be a subset or sub subnet of the locator TLV. Obviously any of these seats itself, they are not being put into the forwarding they.

O

These are basically leaked together with the top-level TLB, and these are some of the SRB six endpoint functions that are supported in the next slide. So again, this is this is the the format. So you have the the seats. You have the flags no flex so far you have the endpoint function and then the seed itself- and it was optional, sub T, always.

O

So the asari six and exits these are the equivalent of the adjacency, see it. So these are advertised inside the the existing tlvs as a part of the neighbor. Each ability advertisement so 22 and you know he related TL- is it in here it's the topology from the neighbor advertisement, so we don't need a topology to be there, but the algorithm itself must be specified, so that is part of the heisman.

O

Obviously it should fall into the or must fall into the locator space, the that is going to be used for for forwarding. None of this is going to be installed in in the forwarding itself. Again, we are relying on the locators which are going to cover this and we have the end, X and and DX 6. You know, as defined in the necessary 6 network programming draft, and we have two versions. One is for point to point in average for LAN.

O

So this is how the TLB looks like again: there's an algorithm field there, which is an important piece and then the city itself, a bunch of flags and sub theories. Can you have a next one is going to be the LAN yeah? This is the de Lyonne version of it with the system ID additional system ID next one.

O

So this is this- has one bunch of the flags. These are the same flags that we defined for the adjacency seeds for the SR MPLS. So it's the same thing being copied here next slide, and then we have the SIV six capability sub T, which basically indicates that the node is supporting a service six, and there are some optional sub sub theories, but there hasn't been any one. Any of these define there's one bit, which is the old wait, which indicates the support for the OEM, which is defined in the other graft, which is mentioned.

B

O

Next one and the last thing we basically changed the way we advertise various SR v6 maximum segment stuff and for four different functionalities. So this has been advertised in a different way. Now we have moved them into what we specified in the segment routing MSD draft, so there are sub theories of the TLB which is defined there. It gives us the possibility to advertise these not just on a parent node basis, but also on a peddling basis which you didn't have before. So that's the major or it's the advantage of doing it.

O

This way among the others, I guess it.

P

Said I'm from Hawaii and one question about the NZ sub TL. We can go back to the salon and and X an.

Q

P

Yes, I see you added the support for the algorithm in and exceed sub t OE right.

O

P

The algorithm field yeah, but in the SRM purist algorithm, is only applicable to the prefix C.

O

Right but the reason is that, because the the end exceed its global here in this harvey six and although the allocator itself contains the algorithm field right, so you can argue why do we need it here? The reason why we edited the a chest for a consistency, because this comes from in a different TLDs. It may come in a different LS piece. You may change things.

O

You may change the algorithm of the locator you may configure it may miss configure and suddenly what will happen is that you have certain locator, which you want to use for a certain algorithm, but the the end exit would actually fall from the forwarding perspective will be part of a different locator and we don't want that to happen. We want the strictest association, so you, basically by specifying algorithm here you make sure that you are not going to use the wrong locator during the transition. That's basically the reason: okay.

P

So if there's any inconsistency between these sub TLV and the locators algorithm, yes,.

O

P

One will be preferred, but.

O

Basically, what you say is: if I have a locator, which is for a certain algorithm. Yes, I, make sure that this one will pick that one and not the other one, because, for example, if you for some reason remove that locator that this one may fall into a different locator. So it is like you want this it to be algorithm five, but it will fall back from the forwarding perspective. The best man would be something the locator which is of algorithm six. You don't want that to happen.

O

You want to make sure that you pick the right locator. So that's the reason why we have the algorithm field here. Okay, thank you sure.

H

So I think I had less presenting that.

O

H

So you're gonna do the.

N

H

I mean if you want to do the other presentation.

O

H

Let's go yeah I mean I, just I didn't know. If you got tired of talking.

O

Q

Have you worked with your bgp collaborators on the same stuff to make sure the changes are in both drafts? You.

O

Mean from the BGP LS perspective: yes, okay, you're working on it.

H

Toni dynamic flooding.

H

So, let's, let's keep moving to I, think last year, yeah for the gates. All right- this is probably not gonna, be very long. One.

J

Okay, unless Ginsburg I presented this at last ITF, not much has changed so what kind of zip, through this pretty quickly next slide.

J

um So what are we trying to do here? One of the things that's missing from RFC 5300 zss grace will restart is we did not have the ability to signal to our neighbors in advance that we're we're about to go under undergo a restart, which means that the neighbors are not aware, and they therefore can't if there are other topology changes that occur during the restart. They can't take that into account and understand that the router, which is restarting its Forney claim, may no longer be valid. So.

J

Okay, I've covered that, so we added two more bits into the TLV so that we can indicate that a restart is planned and so that the neighbor can acknowledge that, so that the router, which is about to go, undergo, restart and delay a little bit and understand that its neighbors are now ready prepared for the router to go ahead and restart.

J

What does the neighbor do when he receives the plant restart? He just marks the adjacency state as being in restart. He updates the whole time based upon what's been sent to him and he sends an acknowledgement and then this the state will get cleared when the actual restart occurs and he receives hello with the the restart request. It was normal.

J

This allows again the neighbor if there are topology changes that occur during the period of the restart. The neighbors can take that into account and understand. Even though they've been maintaining the adjacency in the upstate, maybe it's no longer safe to do so. So the only change that's been made since v-0 is. There was a suggestion from uma. I believe that we summarized the changes, so we've done that in the appendix and what I'd like to do now at this point, there's been.

J

You know, no objections to this draft since the last idea, so I'd like to go ahead with a working group. Adoption call questions.

H

Does any okay? So, let's see, does anybody object to working your adoption of this? No and two people support it. Raise your hands, yes, see some hands ask on the list, but that's good.

H

Okay, also I've been peeking into the jabber room and there's a common comment that which is no slide numbers.

H

So it's helpful: if the presentations have slide numbers I, guess they.

J

Were heard that comment before now? Yes,.

H

H

So is aiming doing it.

H

L

R

Good morning, I'm Nanshan and I have present the dist before in Singapore. Yes, and this is just a brief update.

R

So basically, this proposal is to support a large number entries in for routing in the DC fabric, and mainly, it started with trying to support the overlay routing with this Isis as underneath, and we supports the auto tier discovery, and this is for the zero touch gtp and.

R

The main purpose of the proposal obviously is to minimize the flooding information.

S

R

To minimize the flooding during the link down and know down those negative events in.

D

E

R

This proposal works with the other mechanisms, such as the open fabric. What dynamic flooding proposal recently introduced so.

R

The for the last year, those are the main changes we added the tier information. So if somebody wants to do the auto discovery and without explicit provisioning and as long as the tier zero is no- and this can be discovered dynamically and recent in the most recent revision, we added a bit in the link attribute of the of the ice ice subtly. Nineteen of the theory 22 to say out of this link. My connection is a live note.

R

So if you are doing a dynamic, flooding and algorithm, you don't even need to consider those leaving notes, so it reduces the scaling of the algorithm and make it simpler for the network and the. Lastly, there some comments are the question arm. Yeah.

R

The info requester sub theory is a ask of force, basically information being linked or no doubt. But how do you ask for withdraw those information when the link comes back, so this is just following the normal ice ice LSP mechanisms, so there's no new signaling is needed. So we added some more text in the draft to explain this portion and yeah, obviously request form in further review and the comments and I at least know of one implementation is underway. Based on this proposal, and the authors would like to request working group adopts questions.

F

G

So we have a lot of flooding, optimizations and Chris and I have been having discussions on that, and so we're gonna ask the working group for input as to how to move forward and as far as people that want to deploy them.

G

I'll be honest: we don't completely agree on the various tracks, but so we'd like to give a lot of working group input on this. You want to add about the requirements. No.

H

But yeah I mean I I, don't think we're to disagreement. um I just we've got a lot of different solutions to sort of a common problem and I'm just wondering do we just you know. Does it just pick everybody's favorite solution? Might let the market decide yeah III, don't know, but I would like some. Maybe some input from the group on how we think the best way to move forward is with with all these different ideas. Jeff.

L

Has a I'm not following this work very well, so it's not that it's not clear I, just don't pay attention else. Our group heavily one of the my observations here is that this seems be pitched for a standard. No clothes advert type topology, which is very popular right now, but one of the things we've seen in the presentations, including you know, like the open fabric presentation, is that other topologies are being explored for things. My question slash comment to you all is this is optimized for that situation.

L

If you end up with a different topology other than Clos, how does this feature potentially interact with those? So, as you consider do extensions, you know consider how it might actually interact. If you try to deploy it. A different topology I, actually.

G

This is just a ceiling. This is observation, I think this one is simpler, that it's just between tier 0 and tier 1. So it's not. We limited to Clos right.

L

And that's going to be probably the the single most common thing when you look at the other topologies know is when you get no an extra layer to up that, it becomes more interesting. But my suspicion, based on how this is seems to be presented, is that the generalization of here's how it apply for the next levels. Are you actually doing Auto provisioning, though for topologies?

L

No, like all, you have basically the easy mode for clove and therefore you're going to have no, this is your 1 1 2 and that they auto discover each other and can tell that they're. Miss wired, for example, I I.

G

Guess yeah for the zero touch provisioning, it is targeted. It close right and.

T

Chris Martin arista I was just gonna say that if you look at the agenda, we still have I think two or three more presentations that cover areas like this and that doesn't include what's happening, for example, and rift. So going back to your point, AC or, and both Chris and AC, in terms of how the market looks at this, I would say that there's really what I mean from my view, and just Jeff just mentioned this as well.

T

There are at least four different approaches that are really tied to Clos and then there's, maybe one or two that are in terms of generic topology. So maybe we look at them as they apply to them, maybe the bigger use case, which is which is leave spine or Clos, and then you know the generic use case, which is arbitrary through ologies and then let let's decide there because there obviously is a big need for addressing this problem and I.

T

Just going back, I mean having these discussions over the past couple years, we're trying to solve a flooding problem, that's kind of limited link, state protocols being used in leaf spine or in close apologies. If that's what we're trying to do, then that should be the fundamental requirement. You know a lot of the work that's coming out now over the past five years has been to try to fix the flooding problem, so we have BGP being used to fix flooding. When now it looks like everybody said well, maybe we should just fix the flooding problem.

T

So if that's the fundamental requirement, I don't know if Chris is what you're suggesting that we go to a requirements document first to try to solve flooding problems in iGPS. Does that say if that I mean I, there Dave cats at a presentation at Manoa in 2000, and it was this comparative anatomy presentation and basically there's a slide. That literally says you know, the problem with scale is reduces really to flooding. This is almost 20 years ago and there's been no attempt made.

T

So, if that's really, where we're trying to do them, why we just focus on that.

H

Yeah, so that's all really good I, don't I, don't know whether we go to a document or not or if it's enough to just talk on the list but I think especially when you get competing solutions. If the focus is to try to pick one, then then sometimes a requirements document helps focus that but I don't know if we're there yet so.

U

I must wait LinkedIn, so my take on this is is I'm much preferential to having unified signaling better than a unified solution at this moment, because there are different scales and different kinds of topologies like Chris said and asya said that you know we need to look at and I would just prefer for the moment just to get on the path of having unified singling and let's worry about the the the which solution is best depending on what happens in the marketplace and where things go in the future, because I'm not convinced that a requirements document requirements pretty simple and I'm, not certain that a requirements document gonna be able to capture like operational realities in a way, that's gonna have to be helpful in the long term.

U

Just a thought.

N

Just in sort of North death rooms, so few 80s ago we started routing in DC effort. If you remember, and as a result, we put two new working groups, rift and the other dike.

N

So I don't think the end goal is to reduce flooding and goal is to be able to address the body a particular size, and this what should be so? There was a document requirements for this routing and number one or number two required was really ability to address large topologist without hiking around so I. Don't should be focused on reducing flooding, who should be focusing on addressing particular topologies particular size and see how, if this is a new protocol fit into this picture.

C

Hi, this is Dina. I haven't been following this work, but 4-leaf spine topologies, which will be a huge market in enterprise. If you want to solve the you know, a lot of people have been talking to me about this and I said I thought Dave can't solve this with mesh groups a long time ago like which was referenced, but if you want to fix the flooding problem and simple, even though large-scale leave spine topology is just you static routing. Have you considered using static routing.

C

Has the group considered using static routing for leaf spine topologies.

H

I've thought it but I yes, yeah I, think you're the first person to say to them. Mike I, think I.

G

Think I'll you.

H

C

In static routing I'll, take you.

G

A long way, a lot of these dynamic ah controller based solutions, basically due to static routing. It's just coming it's just more dynamic than in the past yeah.

H

That's what those are just just son right, use: yeah! No! That's I! Just written up my AC.

C

It's kind of funny: you brought that up, because what is it Sdn controller? Do it's a management plane that configures static routes right? We did this in the 90s, with IP route commands using what was the name of that rancid was the script kiddie right, but.

R

This proposal is there similar to step karate. You actually installed to stand karate or instead across.

O

Tony P juniper so links to fail so.

H

As met observe, you withdraw the route when the link fails.

S

S

O

Right so as met observation, I think the problem neatly split along whether you have a direct topology, the north-south sense or whether you have a generic topology, and if you work on that stuff, I think you have to keep it apart. I think in the north south, the director topology, is actually quite tractable and there's a couple of different proposal of different properties. The generic problem has been walked for twenty years right with by multiple people with poor results and I.

O

Do not think that the results will improve because nothing new, nothing new in graph theory and the properties you will find hot links and so on and so on are still in place. Whether the north-south is more immediate and I. Think it's much more tractable. So it's kind of meta input, yeah.

R

Agreed it's more into like the first here fully mesh that Canova simiiariy.

G

Okay, well, that was useful yeah, let's start a thread on this on the list. In any case, we're not gonna commit to asking for adoption to any of these today.

H

Okay, let's switch back to the lead, I Namek flooding. I don't know who wants to present that peter.

E

H

Be here I think we got that loaded.

E

H

E

H

It's almost like.

E

H

Trying to set up a conference up right and here here I got it. Does things about there's a meeting minutes.

H

G

Obviously the Chromebook was what was targeted at different generation yeah.

V

V

Q

That's Argus.

H

E

Did you move it up here? Oh good.

O

All right, okay, so this is an update on the dynamic flooding draft.

O

B

L

O

Changes that basically, there are two major changes. First, we have the encoding for all the three protocols. Now it was originally defined just for ice, yes, and the other thing is that originally this was defined to work only in the centralized mode. We also introduced a new mode which is a distributed one, because we thought maybe some people would like to do it that way, so the concept of the area leader, but when it was defined. Originally it was the one who basically computes the flooding topology, so those who are not familiar with this.

O

Basically, this is one of those tasks that try to solve the flooding problem and the way we are trying to solve this flooding problem means to reduce or calculate the flooding topology on top of the existing topology and only flat over those flooding topologies.

O

So originally, this was defined to work in a centralized mode where an area leader computed the topology flooded, the flooding, topology and and everybody would flood. Based on what this area leader calculated and announced. Now we introduced the distributed mode where, basically, we leave everybody to compute the flooding topology using as some you know, algorithm, which we agree on and basically it doesn't.

O

It removes the requirement for the for the flooding of the flooding topology itself, so the area leader now has either can in a centralized mode, is the one responsible for calculating and propagating the flooding topology in a distributed mode? It only advertises what is the algorithm which is going to be used by everybody else to compute it?

O

Obviously, we only need a single area leader in the area. There's no need for everybody to participate in the area leader election. The obviously would become a few more ways to do so for the redundancy.

O

So, as I said in the centralized mode, HD area leader who does the calculation and distribution in a distributed mode, it's all the participating routers who calculate the plotting topology itself. Obviously we need them to agree on something and that is being advertised by the area leader.

O

We can either use a standardized algorithm which will tell us how to calculate this, because we need all the routers to do the same calculation. Obviously to have the consistency in the flooding. It is also possible to use the private algorithms, but we need to obviously make sure that everybody else you know agrees on what that private algorithm is.

O

So there's been a change in the area: leadership TLB, where we edit the algorithm. If it is zero, then we assume that this is a centralized mode, and then we added a space for the standardized algorithm in a space for a private algorithms.

O

The area system added sub TLB, we removed the ending index because it was redundant. We have the TLB length, we can calculate what is the last index in the ple and there has been quite a lot of editorial changes. So what is the flooding behavior? If we receive an update on the link which is part of the flooding topology? We flooded on all the remaining links in the flooding topology.

O

If you receive an update on link which is not part of the flooding topology, we flooded over all of the links in the flooding topology and when there is a change in the following topology itself for a short amount of time. We need to flood over just the the union of both of these new one and old one, just to make sure that you know everybody else converges to the new clotting topology and we added the OSP of extension for this.

O

So we have area leader sub TLV, which is part of the routing information LSA, which is both v2v see. We have the dynamic flooding epochal si being defined for this to be used by V, 2, OS BBC organic body. Let's say to be used by OS BBC. Both of these are only required in the centralised mode. They are not needed in the distributed mode, and then we have the equivalent of the existing eye size up.

O

Tlvs used to announce the be the flooding topology itself, so I guess I'm not going to even ask for the working group adoption, but we would like to continue to evoke this draft. We need to define some standardized algorithm, how to compute the flooding topology and if you guys have any good ideas, yeah all done welcome and we would like to start to work on the implementation of this.

G

Comment, so is there an implementation they may? We talked about that is underway with this.

O

Yeah we are looking at that yeah.

G

Okay, so so for both the last two or these funding reductions, there are at least people looking in implementations.

O

So again, this is a generic solution. It's not tied to any specific topology can be using any arbitrary technology.

H

This is a naive comment. I have really thought about, but it isn't like a pin or I mean. Doesn't multicast distribution? Give you server tree. Could that be an algorithm? But it's not reliable. Well.

O

You may have want to have some more here right. I mean you can come up with the very simple spanning tree and it would give you, but you don't even more yeah.

C

O

Want, at least by connect the topology, maybe even some more redundancy, so any failure would not basically leave anyone without the data yeah.

H

O

It's a bit more interesting problem to solve yeah.

O

Thanks, thank you.

H

Go I gotta get out of that mode, yeah, and then this is our and I usually could control them. Make it open.

H

We're kind of cruising along here this.

D

H

Am I here I'm up doesn't even have a chair.

H

Where did they go.

H

W

Yep, hello, everybody, my name is Uma I'm, going to talk about preferred path. Routing last IETF I talked about this concept, but you know, thanks to us, Jeff and Lewis has given a lot of feedback, so we are updating with this title. Next. Next slide. Sorry, okay! So what? Let's look at a simple scenario? Just before going to the problems? Simple s, our network here, a 1 to a a a 7, is a shortest path.

W

It's in blue line and it's an MPLS network, Empire MPLS network global indices indices are shown for each node and hi Jason, see if you have an SR path from a to a 8 B 2 B 2 X, a 9, a 10 X, a 7 shown in the purple line. It uses both nodes later and a distance. It's it's! It's a total eight six rates, its stack, so this number can go up based on the MST capabilities. You have all kind of nodes.

W

It can be there in deployments right from Broadcom chips too high and chips, juniper, Huawei Cisco, whatever so various MST rld capabilities. You need insert the Lal pair at the appropriate transport stack, so the stack.size a signal can go higher and if you use ipv6, you have to use ipv6 and cap header RS 8200, basically, and you have to use as a red setter with all 8 6 with ipv6 addresses, basically 16 ipv6 8 16 a good success. So what's the deal with this, we have couple of issues with this.

W

In the deployments like say deployments, you have hardware capabilities. Various nodes can support various adepts. So so we have a solution for this MST maximum seat depth capability. So with this, PC can compute alternate paths based on the constraint of MST, but it may so that you may not have the paths available. So that's number one and second one is line rate issues we have. We have test run on the ipv6 as our v6 header, with multiple Sates. It's a huge significant performance down line rate down. Jeff has some data.

C

W

That with various chips, so that's one second ratio, 30 she's, empty issue with the high end, Packers close to 1500. We have, if we put, if we add bunch of sites with ipv6 SRH header, we have empty empty on fragmentation issues. This is serious, issuance fragmentation issue you have to. We have to empty issue. You need to change the network nodes. A couple of configurations need to be changed. It's a serious issue, and apart from that, we have Phi G header tags.

W

There are some of the Phi Z slices, like mio T URL C is defined by small small packets packet sizes ranges from 4200 bytes, and this is also common in fixer scenarios. Where interactive- and you know online gaming kind of things away. The short packets are prevalent with this 4200 fights packet, you have to introduce 200 to 300 bytes of net attacks, so it's a huge tax has set its max up to losses of bandwidth.

W

So what's the solution, we are introducing a TLV which describes the path, so routing happens based on the path. You have four components here you have reachable TTL. We have that prefix. For example, 135 to 35 to 37 to that RMT capable is stl ways. Whatever is reachable tell we put the prefix there put a PPR ID, that's creeper PPR that identifier. If you have MPLS data plan itself, it's a node set are as global adjacency said or whatever.

W

If it is ipv6 data plants ipv6 address, it is v4 data points the v4 address and bunch of TLB is which describes the path that is path, description element and third last one is PPR attributes which specifies the attributes, for example, for traffic monitoring traffic usage cases. You can just specify that these these traffic parameters need to be accounted for for this path. This can be described with GPR attributes. There are some other T parameters that can be added.

O

W

So how do we do computation? A simple example here, r1 to r4 in the Purple Line? This is a strict PPR. That means all the nodes and their distances are listed with this. So let's say, for example, on the our line. If, when the TL is received to reach to the SPF on suspect on petition is done or to reach to our for the next, stop is r11 so because it's shortest path so, but based on the PDE list, PD says path.

W

Description says the next next next next element is our 10, so the next stop. We changed to our 10 so for your SPT computation, your tax trust, nothing need to changes during the round download just dip into the pantry. Get your get your last PD element and change the next stop. That's all you need to do and same thing happens upstream all the nodes and for the loose PPR. That means all the nodes and edges senses are not listed.

W

For example, in the yellow lane, R 1 2, R, 4, R, 6 2 r 7 is not listed. It takes the shortest path based on the Dijkstra's. So for that, but you know various data planes require way this capitalist for this also described in the draft. You need two levels here. In summary, it's a simple change from next stop from the action like stop. Spica connect stop to the next stop as described in the PD list. That's all you need to do and you have to.

W

Of course you need to program the data plane with the corresponding data plane with the next. So, with this all FR solutions, your IP l fa, a--, motty's, RL fa, k, LF, whatever is there today computed l FS can be directly can be directly used. You just have to take the next step of the computed. New next stops LFA's local productions. It can be applied, but there is also a lot more can be done with this. That will be discussed later.

W

So what is the relation combining said? Some people ask, there's no relation to binding sage is earlier. The ASR extensions of this advertisement of PR was are there and with binding sites. As of today talks about advertising.

W

It's a mapping, server functionality here.

D

W

From that traffic accounting, there are multiple solutions for traffic accounting in my ETF right now so, which introduces because of the problem with SRM LS sees when you keep on going from one sequence. Another segment label will be popped, so you don't have a notion of path into the into the packet. So with this there are, there are problems in the sense like you know, how do you account for how many packets are pushed onto the each node and intermediate nodes? So there are solutions way apart. The path labels are introduced.

W

The problem with path labels is, you need to put a special label followed by path level. So it's already, you have a cig depth issue will increase more more labels to this sure. So one of the solutions talked about is why don't you compute at the ingress nodes? You can do that, but what happens is if you do FRR any link, our node failure happens. Those statistics will be off. So what we are saying here is in the path attributes you can advertise. What are the path? The statistics need to be mounted.

W

Those can be stored into the node and those can be pulled through G RPC or whatever kafka part out that can be used.

W

So the original example, the same 8 sets with entropy labels at the right are LD 6 here, so you have to put them to be level at the right spot, Lal label. So with this around 12 labels, transport labels, those service level, is only one here. It can be more yeah.

C

W

Prefix label. We have VPN address peer engineering labels. You have for address, not protection. It could be in any case like two to three label. Service levels will be there, so we're talking about around 50 labels here. So we are with the solution. Talks about only transport label, optimization with transport label. You have to fix shortest paths it. That is the note said we have the PPR ID. Another note said- and you described the path of this notes- and this gets complex to one label same with the service.

W

It's the advantage with a sorry substitute here, so with eight sry, six in and SRH it's a serious problem, so it can be reduced to one the remaining places. You can use the remaining six. You can be used for other programming or whatever VPN labels. You can be used for VPN 6, so the transports is reduced to one here. So the hang data model is described in this stuff meant for PPR.

W

There are couple of questions. Searching is asked last time what about scalability? Yes, there are n nodes, n square paths, total Pass and with K multiple of each node yeah. So, but not all not all paths need to be preferred paths, only small set of paths or preferred paths. We have dimension the network with two hundred nodes, five hundred nodes thousand nodes and two thousand nodes with 15 percent capability I'm there from H nodes, two parts, so the sRGB for MPLS case for a lot of customers ask what is srgb usage.

W

So around 15 percent extra labels are needed for srgb, it's not a big deal but but separate solution. We are trafficking a separate solution to scale this with a street tree structure. The tree is rooted at their destination and multiple input point can be there. So this is a separate work where few scaling has to be done. This will be published soon. Totalus is working on this, so we will come back to this soon. Actually so, but as it is, this stuff independent sustain.

W

So where forget nut and fight use cases only from you, node B to UPF, only three slices will be there, so the total path will be less than 500. Based on the network, though, so so with the tree structure anyway, the N square problem is come to a linear issue, KK star n order of n. So it's a it's a linear scale and another advantage of this is you have hot steering support for native IP data planes.

W

This is very important for legacy networks and Phi G, especially where s at least six works has to upgrade all the boxes, because it's a new header, new half way, half extension header so fast steering- can be done with the PPR ID ipv4 and ipv6 set up ones.

W

So, there's a somewhat going on. Remember. Dino knows this city for response is being framed from 3gpp what they are looking for, optimize get a plane on the new and nine interface on. So there are competing solutions there, a sera t, v6 there are computing capabilities, Lisp and ila. So this solution can help in fully compatible with sry six. It only helps reduce it. The transport stack, which is the important criteria there, one of the criterias there lisp also it can help traffic engineering inside the remand, but least has another capabilities.

W

What fight you cannot solve, it can offer session, serve, I will mobility and other SS, which is orthogonal to this. I am going to talk about how this helps in Phi J architecture tomorrow in the DM over Rho. Please welcome there and have questions we can ask.

W

The concept is present in a fight, wha, IETF 101. We have a lot of feedback thanks to us. It was completely reshape the document and what comments welcome more comments and I have.

G

One question: I read the last one: I didn't get through all this one, ah so the selection of the preferred paths is going to be separate work right. How you know how you know what.

C

G

Of those those N squared paths, I'm going.

W

G

Do preferred passport? That's not.

W

Good at this stuff that is, XE completes it. Yes,.

G

Okay, okay- and this is just this really just covers the data plane and the advertisement of the of the of the ingress nodes for the preferred paths. Exactly okay,.

M

Hi kitten tear out the car Cisco so.

W

Isn't this very similar to basically flooding and LSP in IDs flooding on LSP in a GPS right very similar to how our security parts were set up? The only thing difference that I see is that it's flooded all across the domain, so everybody gets all the LSPs and they just pick where the LSP on which they are part of and programmed their PPE, our ID or label it. So it's really flooding all whatever order. N Square LS B's, all across the domain to every router. No, it's not order n square I mean.

M

W

Paths, yeah all are all paths, so let us say highway if we have no issue with des our path. Please go ahead. You don't have to advertise. If you have an issue, please advertise and reduce your stack size. Otherwise you don't know how to advertise. Okay, but let's say we have a thousand such thing in a area, then those thousand would be advertised by each of the ingress node and they will be flooded to all the routers correct.

W

So it's it's really an RSVP signal LSP, which has to be programmed hop-by-hop or loosely, and it's being flooded by ITP.

M

Another sip, you don't flood it. This RSVP signals only on those hops, but in this case we are flooding you let let's take this discussion of the list and Deena will be the last question comment. This.

C

Dida, so I have four very quick comments. First, one is I'm overwhelmed by complexity. Number two: your tax overhead was completely understated. Number three I think you found new. You have to build new machinery to reinvent MPLS in a more complicated way. This is just my opinion. You could take it or leave it. It doesn't matter, but you could have solved this problem with straight source routing with not every hop-by-hop as the SID.

C

You just could have found a repair path, input one entry in the source route to get around the failure, and that would have taken you a long way. This is not for only for a.

W

Ferrari, a Ferrari is a side product of it. If you want to.

C

Go so is coming. Thank you. This is for te as well yeah. He said, there's more to come and there's a lot more stuff in the basic stuff is already complicated. So you know there's no complication.

W

Looking forward to the.

C

H

No well I think yeah.

N

H

We could be more talking to you, um okay,.

G

Do you want to start the discussion on this oma on this document on the list.

W

Please we are ready, actually fine.

H

So Dino, maybe you can watch for it.

D

H

H

Okay, Charlotte, oh okay, good.

P

Morning, everyone and children from Hawaii and I'm going to give an introduction about the IGP extensions for the second routing, based enhancement.

E

H

Hard to tell which one is fullscreen in that okay.

P

So here's some background. The enhancer APM framework is defined in this draft, which is to meet the requirements of network slicing and other similar scenarios, and this draft describes the architecture and the candidate technologies for the resource, isolation and the integration piece, as this drop will be presented on Wednesday the test session, and we have another draft, which is a second routing based the enhanced VPN solution, as defined in this draft race.

P

It basically to extend, extend the same routing to identify the partition, the natural resources and created the resource, isolated virtual networks and because this time we didn't get a time slot in the spring. So please read the draft and comment down the main list, and now we have. We need the IGP extensions for this SR pasted solution, and this work belongs to this IRS, our working group.

P

Basically, we need the mechanism to distribute the narrow slice informations to post the controller and the networking nodes. So the current rate in this current draft, the design is to try to reuse the existing, my multi topology, routing or segment outing or is some necessary extensions. So we expect this mechanism. Will be applicable to post the SRM pairs and SR mystics.

P

Okay, so this diagram shows how we want to extend the Sigma routing for the resource identification and we can see on each link and note in this. The topology different, dedicated, adjacency, see it or the gnosis are used to represent the partition resource and each seed will be associated with a virtual network which can listen another slice and we can use different groups of ceased to construct the resource, isolated virtual networks, as shown in the right side.

P

We have three virtual networks, each with different group of seeds and this it can be mapped to the data plane resources isolated from each other.

P

Ok, so here's the overall mechanism I mean not go through the details, so basically, this is a controller based mechanism. We use the controller to maintain the natural States and receive the service request, the controller view to the computation of the logical topology and the resources needed. Then we need the network nodes to allocate the resources and also the associated adjacency and no seed for the logical topology.

P

Then we will need a GP for the distribution of the mapping between the seed resource and the logical topology. This we information will be used, but in ingress and every node to build a same route in forwarding entries for each budget apology and since we want to reuse existing mechanisms. Well, let's take a look at the exiting existing technologies. We have first the map. Mountains are very routing. It was designed for to create a multiple, logical, topologies in the same infrastructure.

P

So far it has limited use cases, maybe for the creating topology for ipv4 ipv6, for the different apology for the unicast or multicast, and there's some limitation in the data plane forwarding with Maree topology is that's not a fully supported sharing of the link or IP address between the different apologies. Then now we have the second routing. We can see how it works with naughty Polly routing. As we know, the multi tier routing is already supported in the IGP extensions for the same routing NP RS.

P

So it allows for the distribution of the topology specific adjacency seed and no seed so that we can have multiple topologies enabled with the same interface or IP address. We can use different adjacent seed or no seeds as a discriminator for different typologies on this same link or note, but the problem is so far: we didn't have the mechanism with MTR or semi routing to support the resource isolation of the identification. As I said before this, this extension, we need to do with same routing and we needed a GP extensions for this also.

P

So here is a proposed mechanism. Basically, we will try to use the MIDI topology where Sigma routing was the resource identification. Then we can have each node or interface participate in multiple logical topologies and the resources on these nodes and links can be partitioned and allocated to each topology. Then we have the adjacency and no seed to represent the partitions of resources allocated for each particular topology.

P

This sees will be used to constrain the traffic to the allocated resource in the data plane. Then we use IGP to advertise magical mapping between the seed and resource in inch topology and with this mechanism we can support both the straight and lose paths as our forwarding, which topologies constraints.

P

Although way you can reuse the existing mechanism, we're proposing some extensions which we see necessary, the first is we need to define new sub shell. We called similar out in bandwidth sub theory.

P

It is used to describe the link van waste associated with a particular adjacency seed for the use in the particular topology, and the second point is: we need some audit apology support for a 36, so in this document, with the proposed a new until I T sub G re under the SRS, it knows it shall we defining the I sighs contentions, and then we notice that the newly updated as a wrist extension draft also introduced as a part of the money topology with our services. So we think we can convert in this point.

P

So the next step is we want to students if it feedback on the proposed the IGP ice ice extensions and the next up. We also want to extend this fall to the OSPF.

O

To speed up shinnok from Cisco, so obviously MTR exists, you can use it, you can create whatever and you can do a lot of stuff with it. My comment would be: you know how scalable this is given that an MTR you have to Velma ties every link, every properties of the link, adjacency prefix, everything on an empty our basis. So how scalable this is I, don't know how many slides is you are thinking of and from the provisional perspective, it's the same thing right.

O

You have to provision it on a per link basis, so I think there are maybe alternative solutions. You can do the same thing or achieve the same thing in a better way. Yeah.

P

I think we can discuss a flying about the alternatives we think the scalability of MTR. It really depends on the number of topology of slice. You want right now, so maybe you can see whether some optimization it's possible hi.

C

This is Dina um a C and Chris I have a general statement. It's not relative to your presentation, but I see a pattern happening every 5 years that I GPS, especially when state I GPS, are being used to move information across the network, and so they, you put him in LSPs and they're stored in the middle of the network that don't use it so you're using all the resources of the middle of the network.

C

Where really the edges need it and the I GPS just used as some sort of transport so I mean we saw this would I be GP a long time ago and put in route reflectors and what we saw time and time again is you need a reliable, multicast messaging protocol that works among the edges, and you know how expensive routers in the course storing stuff has only used on the edges. It's not a good architectural principle, so I'm making just a general statement is that you know.

C

Maybe this group should look at running reliable multicast calls on the edge to have helped transport stuff, so routing works better I know this is a link state routing working group, but you know you only can do routing so many different ways and now, with the advent of segment routing that came to the last couple years now, people are saying: ok, let's go look at source routing techniques or Sdn controller techniques versus distributed in the network techniques which we all know and love.

C

So the problem is, there's only so many ways to do these sorts of things, and now people are just taking every combination of everything and it's not helpful. It's just creating complexity and we're not solving any really new problems. Sorry to take up your time, but I just thought: I wanted to generalize the problem to other things that I've seen as well. Actually, we.

G

Have more more time than I thought? I was thinking back to the 9th 1130 as opposed to the 9:30 to 12:00. So I heard my apologies for hurry in the last presentation along so.

X

You know I sort of agree with what you're saying that we do need. So we would know the last 20 years that we needed a general-purpose data distribution protocol fall in the routing space, but with real problems you do have to move on and the is is seems to be a good platform to move this on in the short term. But in the long term, then we do need to work on a on a better data distribution protocol. Yes, so I think we should separate the two problems. Well,.

H

Just to clarify Dino Dino actually was saying that it's a it's a question of not wanting this far stuff in the center of the network. Right like a lot of times, you're just trying to push information to the edge and, if you're, putting it LSPs, it's putting its. You know, cost a lot up on the internal routers in your network to store that state, because those are expensive.

C

Because you got to change the middle of network to get.

S

C

S

R

S

C

Because you have to change the middle of the network to get edge functionality, you know.

X

Everyone has to know about those she's.

C

Not there out there what I know the one example was: a mapping function here. There is example putting services in the network there with respect to doing traffic engineering and fast reroute you're. Absolutely right. That's for.

X

The inside of the network, engineering and slicing absolutely kind of very close friends, so we do have to be able to do traffic engineering in here and so will you have to be able to steer the traffic to particular NPU instances, for example, and specific cues. So we are dealing with a lot of detail in the middle of the network that we have to advertise. The findings.

C

For this specific, for this are taking.

X

Application yesterday a.

G

Synonym that was I mean I, maybe just glancing over this I mean I, read the draft and everything but I thought for this particular because you're doing resource reservation. This would not be a case of only the edges needing to know I mean there are cases like you're talking about. Do you know what I was thinking it? You would need this state. Yes,.

X

I really do you. We are gonna have to introduce more state to make this version very to this our work. Whatever way we go about doing it, you're gonna, it's all about particular resources being allocated to a particular slice.

N

So we had this presentation route number of times right and.

C

N

Or enough networks, some now I would.

Y

Say we need to build more hierarchical.

N

Structure, it's very flat, so you flood everything to everybody, the capabilities and semantics of your links in reservation, their property over. Not so, if you build more hierarchical structure, very, have not and also attributes inside that no business doing the other, not in the network. You might scale my brother.

H

Isn't this what ideas were.

X

We're more than willing to talk to anyone who's got any ideas as to how I solve this problem in a better way.

G

Okay, I think the good news it here is the base network slicing would really should advanced before this correct, like the SR for networks for feet and hands. Vpn yeah.

P

G

So so, when keep you know, keep a surprise this. If that's adopted, you know anyway, we can hide still. It advised people to read this draft, but I think it's contingent on the base graft being accepted in spring, correct I. Think so. Okay,.

P

Okay, since I joined, didn't gather visa in time. I will present on behalf of him, so this is OSPF extension for the inter area, topology retrieval.

P

So, okay, we have the background and extensions and the process. Here's the background of this draft. So basically the advantage of the SDN controller is it has a global view of the topology and the other stays in the network, and you can calculate the end-to-end path to ensure the QoS and performance requirement for applications. But when the network is divided into several areas and run OSPF as IGP the detail, topology information will be hidden by the ABR routers.

P

So, as the in controller will only gather the summary topology information from the routers run running the P GPRS and adding locator, which is located in the each area and eyesize currently have the solution defined in this RFC. But there is no corresponding solution in the OSPF. So this is why the extension is proposed.

P

So the proposal is the for the fixed format or SPF messages. We need to reuse the rarely used field and the redefined the associated field, to the short backward compatibility and for the TLV format, OSPF messages. We can define the new secure ways to carry the prefix sauce router ID and include this subtree in the corresponding OSPF way to and versions breeze oasis.

P

Then the B GPRS can recall this information use the same object regardless of the IP protocol used. This is the format of the fixed format our says. Basically, it is proposed to you, reuse, the TOS field and the correspond the four consequences.

P

Other fields to carry this. You define information so when the number of the TOS is given, one special value we can use is the other field to carry this stuff. Router ID information.

P

This is the extension for the TLB format, OSPF messages. The new subji obviously are defined to carry the prefix or sprouter ID for ipv4 and ipv6 respectively.

P

This is the process of how to retrieve the inter area of topology information. Basically, you need to collect the topology information. We have P GPRS and then you need the network summary and software outer ID information and gather from this nearly proposal, and you need to combine these two information to identify the links between the area different areas so that you can view the multi area topology at in the end.

P

So this is the I think I dream for the adoption of this draft and it is asking.

G

A see Linda there's some history here we already we rejected a long time ago, using those toss overloading the toss and using further. So so all the the non TLB based I don't think we'd standardized that part. If we adopted I mean I mean there was a I think it was Padma.

G

Had a proposal way way back to reuse the toss, toss bits for other purposes, and we didn't do it then we're certainly not going to do it, and and at that time we did not have a way to expand OSPF, v2 or OSPF 3-3, and now we have a way. So we surely wouldn't do those first, those first, two, okay we'd only do the TLB based.

H

Right I mean if you're, even if you were look at.

C

It you wouldn't.

H

Pick just one solution to use those better yeah.

W

uh I get anthological Cisco, so we shared comments offline with our Anna. You see that at least some of the non backward compatibility changes have been addressed in this one, but as AC mentioned, the proposal to modify any existing format is, is simply not going to work and backward, but I would support the use of source router, ID TLV and that's kind of equivalent to what is there in ISS, so I think that's something that can be used for sure. This.

G

Is AC speaking as working group member again I think that you know you can look at the use case and say it's a little weak, but this is this could be used for other things too. We have in an is, is I, think I think it makes sense to add it with the tlvs. Okay.

O

Thank you, speed option at Cisco, so I agree. We can add sauce caller ID, but I still don't see how that fixes. The problem that you are trying to solve, because what you advertised is prefix and you associate it. The router I need a source. Relative is the prefix, but what you have seems to try to do if I understood it from the layout. Your stripe, you are trying to reconstruct the topology, but there's no topology data. It's just the prefix yeah.

B

O

I, don't know how are you going to use this to reconstruct any topology data from it yeah.

P

I think maybe you think it's questions go I joined house and you can send the question sure.

Y

Chameleon circuit yeah yeah turning here.

P

Is how to use it apology and the preface together to combine the multi area information you're.

H

Just using it as a locator right and then you're gonna go talk.

Z

To that box to.

H

Get the info using some other records with them.

P

Okay, thank you. Thanks.

P

D

Guess it is Russ I think it's Ross.

D

U

F

D

This should work yeah.

H

D

H

Scrolling I, don't know oh yeah, oh so you can't tell which one is falling, which is okay, so you just have to hit it twice on that. Okay,.

U

All right LinkedIn, this is just an update on fabric. This was presented about a year ago, the draft something out there for a long time. Basically just some review, you just four people ever said before, both the simplest possible disturb you like they put a call, no policy in their configuration, we're trying to get to the point where we don't actually have configuration on the box pretty much and trying to keep out feature creep as much as we can changes. Since the last draft.

U

There were many minor changes based on a lot of it from people's input, which is really appreciated by the way we have a lot of people who have given us feedback and ideas, and things like that and the one major change is that the way the fabric location character is calculated, there are many different algorithms possible. What we found is I've been working with Christian net net deaf on this.

U

A lot in working on an implementation is that there are a lot of ways you can do this, but most of them only work in one topology, one form of the clothes fine and leaf. So we've come up with a solution that I think works in all different types of spine and leaf. Then we can think of it. Basically just uses two routers manually marked as T zeros or top of racks, and then you just find the location based on a simplified SPF using hop counts from that and by the way we're using namings advertisement.

U

So this is what I was talking about before the mic is trying to get to a unified signaling we're trying to rely on maimings draft for signaling, rather than building our own signaling as much as possible and working with naming and less on that stuff.

U

There is two ways of optimizing: flooding I'm not going to go into detail again, because this has been presented before and if you want to talk more about it, I'm glad to do so in person or on list, but we're using RFC 7356 to do limited scope, flooding kind of partially there's a forward and a reserved, reverse optimization. I. Think AC had a question on the list about reverse I'll. Talk to you about that later.

U

If you want to AC now, because I think it still works but remove lots of stuff, we don't need to or care about from eius eius, for instance, in this size of a fabric, we really don't need level 1 and level 2 flooding. We only need level 2, so we just actually in the implementation, if taken level, 2, processing or level 1 processing out as far as hellos, in having multiple sets of LSD bees and stuff, like that, there's some optimized neighbor formation stuff about the way you build adjacencies.

U

When you bring up an adjacency rather than going ahead and exchanging LST bees, you check to see where you allow your first formed adjacency to fully form, and then you just do see snps to figure out what you need. Beyond that current state. There's an implementation in free range, routing, I say it's in progress: it's actually an acceptance test, phase and skill test phase right now. We're gonna do some functional test scale testing, and then we intend to push that rebase it and push that into free range, routing open source.

U

So there will be an open source implementation out there. It's actually a compile time switch on the is is code, that's already in that's already in free range, routing, I think we need a use case, draft end or use case appendix we can talk about the requirements document that Chris was talking about earlier. I think we're gonna need some yang. My model modifications and we'd like to see more implementations of this and, since ac said, we're not accepting anything as working group items in this area.

U

Right now, we'll just skip that request for the moment and any questions again on list off list we're available. If you want to chat about it, what we're working on and where we go from here, yeah.

H

I just want to clarify I wasn't requesting a our Miss document. Cuz I think that's a lot of times. It's overkill yeah.

U

That's fine, yeah I mean we're mostly working on this because we feel like it meets some operational needs that we have on our fabric. It's.

G

So AC Linderman, so this is just since this I don't know this almost seems like it might be a step separate track because it diverges enough now from is, is I I, don't know, I I.

V

E

G

What III don't want to say experimental because it could be more serious than well a lot of experimental protocol, start out as experimental and then become very serious. So maybe that would be something to do and just continue I don't want to dissuade you from changing things there. Other comment, I said: don't you know you need some discussion of the checksum? Oh.

U

You know Christian, we didn't think about check them. What check them? Yeah.

G

Okay, because because you're gonna change you, you know, routers are gonna change. The flooding scope then other than the during listen.

U

Oh I shouldn't matter it's because the way the sunny trip is done. It shouldn't matter. Okay, yeah, it's that it's summer, 72-56 extensions which are basically.

Y

I think the PTO type is not encompassed by the checksum, so you can actually change it. Okay,.

U

Good, okay, all right! So if anybody has any comments or questions bring them to list or privately to me or Christian era,.

H

Also I was thinking about something earlier with the flooding with all this flooding stuff. It is do we have anybody in you know like the old interrupts labs at universities and stuff I mean it would be really nice if somebody was actually taking these different algorithms and and comparing them and running them through scenarios and getting maybe publishing papers. Yeah.

U

Actually, it would Russ white LinkedIn one thing I'll say about if in fabric that's interesting is this is actually modeled or uses the same algorithms or mechanisms used by the OSPF maeín a work that was done many years ago. So actually this these algorithms are pretty majority deployed and running I.

G

Men have over much more complicated than this one, though.

G

Okay, we got we mon.

G

Also, since this last presentation anybody not signed the blue sheets, you please do especially.

H

If you're standing up and didn't get a chair, you might want to sign twice.

H

AA

Couldn't Molly have for one today I'm going to talk in about a flat e reduction, so basically our folks are updates. So during the last ITF, we quite a few experts with some issues regarding to flatten topology, for example, if for some dingdong some interface time, and in this case the plugin for G may be broken. So in this case, how do you provide or currently reliable, flooding in the case, some link? Bangs am not done so in these updates.

AA

We provide some solutions, so the second update is that in the general version we talked about a different mode or computing flooding modules. For example, we we focus on distribute this river mode, which means that every node in the network will compute fraud, energy and use their own, providing for barges that is to promote.

AA

We also mention the about centralized mode, which means that in one network we select a leader so that a leader will compute etymology and then flag in authority into the network and then usually in the other node we receive that topology and use that party. In addition to that, we'll also talk about statically configured that were not in commercial, so in this case a niche mode was in some way operators can configure the foggy info module and then each node in rebel just use that static convicted technology so that we mentioned this remote.

AA

So in this version we provide support to like operator to select whatever mode the lack. Also so sometimes operator me after select select mode, the may change to another mode. So in this case we also provide a support to allow operators to smoothly switch from one mode to another mode.

AA

So, first we provide some solutions for reliability.

AA

Distributed or flooding the link state module, so here we define a critical in the face, so greatly in the face is a unified in the flood in topology. So when this interface is done, they need the flooding already where programmers meet. So this interface, this kind of invest is a critical interface. So similarly, we can extend this definition to note critical node, which means that when one node in the FATA model is done and then the flat homology is broken or split, so that node is a crater note.

AA

So here we give an example.

C

AA

The flag in creepy interface, for example, here the stick, the stick: nine three! Nine is a critical interface, because if these are in the face is done, you can see that the freighting homology will be split broken.

AA

So, in this case, I think the simple solution is that greatly innovates is done, so we just a user, will return to the mobile flagging, which means that every node in the network just flap to every interface, that's a traditional flattened, so that's the simplest version in cradle, university' or create remote is done so in this case we're currently the link state. We all rely belief flooded. So that's first, simple solution. So in fact this kind of solution is a simple, but we use more resource.

AA

We can go down further to some kind of optimized version, which means that we make ritalin visit our work written all these done. So we can compute by cut, has some kind of backup or flooding path. So those have a backup lighting path will be some way those hambaga passed back up tomorrow. This is a way to go through the.

C

AA

Which connect those are splitted Parvati, so I suppose we get those node. So the simple way is that each of those node on the pike lab staff we just flat to every interface. So in this way we are currently the front of our G at the estate. Well reliably flooded.

AA

If we will go further organized so we took them for those nodes on the back hub property. We don't need to fill out every image we just flood to some of the in the base. So we can see that we have was several level ways to provide County, and this is the floppy simple way.

AA

Is that, as we have quickly innovative free cloth on every not just return to the role model, very simple, so you will have some kind of organized, and then we have her son of note each each of those those were for not to ever know every week. Then that way we have some kind of optimized solution. Further, we can go to even optimized solution as those part of North, just rather to some of the links.

AA

So, secondly, we provides a summer support to allow operator to select select remodel. So here we provide so so. First we define operation so operation is that operator can select, whether we wonder flat reduction so operator can select this one very easily. So over can just configure on any note in that in the network, for example, flat.

AA

So I saw as a operator configure that one on any one load and then we can just a flood this instruction flood reduction. So in some case after we have a flood reduction, so maybe customer want to do back.

AA

So in just a configure, wait just go back to global flooding, so we provide some kind of operations. You may be flooding and then go back.

AA

So if we give a customer configure from a production after that, the customer can select the flood mode. So we will provide us provided us sever mode. So we provided a customer can select centralized mode. So if a customer can also select each human mode and a static mode.

AA

So for you to mode so we divided the roost so, for example, on customer config our simulation mode and then we me that the whole night of needed to select and leader at the backup dealer and then the leader will comes up computer logic, Barbie and the flood is quite entomology to Evernote envelope and then every other ever don't even matter. We've received the flag atop our key from the leader and then use these private parties.

AA

So for his tribute mode is the Evernote computes Fraggle property, and the news is only also our operator can indicate algorithm for computing from added margins, because we make if I'm different. Now several algorithms and the user can select one of them so to currently Evernote will use the same algorithm to compute for our topology.

AA

So this is a detail, significantly use paint, changes or central mode, because the four central centralized involved. We need the Select leader so in what would support this one? So each node in the network be to distribute some information, for example the priority for become a leader and then also after selling leaders is selected or in the process of selection. So we need a device, some kind of POVs to indicate that the leader or backup leader is selected. Also, we need to have a algorithm to select a leader and a backup leaders.

AA

So in summary- and then we address the issues risk during the master idea and also with profiles, support to allow operator to select operation mode and switch from one mode to another mode as smoothly so I see this one is a general for general, flattening reduction for any apologies. I think we would like to wake up comments.

J

My skin's Berg I.

Z

J

Like we've have two presentations that are saying exactly the same thing. You know comparison to the earlier presentation that Peter made so from a working group perspective. It doesn't make any sense to me to progress to documents that are doing the same thing.

J

Yeah, it was striking. Me is very.

H

Similar as well.

G

A CNO speaking as a working group member just intuitively, and it's not because I used to work for Tony and Peter's my friend so I don't take that into into consideration yes yeah so, but intuitively. The first one is better because you're reverting to normal flooding at just the point where critical links are failing when you're going to have more load on the network. So it's it's kind of you know: you're saving flooding when you don't need to, and then lots of changes come in the network or if you revert to normal flooding.

AA

We discussed different solutions, so one solution is of adjusting, so another solution is that we can find a bike rack back half of our apology. That means that part of knows what ink will be used to guarantee when one criticalink of notice that we use those by hardening controls some of link. Currently, we just have awakening that as another solution and then even in between some horrible loads. Just to the most of those of you use the five important different levels for optimization. That depends. We can discuss that in details.

AA

H

So Theresa's point having common signalling right, I mean, but then what les was saying it's we don't need two things to do the same thing. Maybe you could be working on the algorithms we'll talk about the distribute algorithm. Sir I mean they really seem to be doing the same thing.

AA

I think just from beginning so I think economies are not winning me the knife. Well imagine.

H

AA

Then my fortune is distributed right, but also I talk about distributed other side calculations, but my focus is the distributor and the Tony mention is a signal fire. Even.

Z

AA

AA

My first question I talk about a sweet mode and then I think us the issues raised by the expert and also generalized to cover war, molds and operations right.

H

Maybe I mean maybe maybe maybe there's room for working together, I'm, not sure.

AA

The same problem then, maybe combined together, I.

Z

Z

H

Let's do that before we do another example computation again: okay, anything else.

Y

H

Yeah I mean I just struck me again. This is a naive, accommodate gonna. Reread this draft but isn't static. Topology is that mesh groups.

AA

H

Static configuration of top up fighting topology as much groups.

Z

It's the starting company that uses first thing.

AA

They already have it.

Z

Right, it's mesh groups. We.

AA

Doesn't mentioned that one user.

Z

M

C

AA

Are mini virgin at them? We do need a computer with a user. This one here is house as an example for girls.

AA

AA

H

It looks like we're wrapping a little early. Yes,.

D

H

Other comments thanks: everybody.