Internet Engineering Task Force 98, 27 Mar 2017

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF98-PCE-20170327-1520

Description

PCE meeting session at IETF98
2017/03/27 1520

A

Just and you can select the right.

B

Hello, everyone welcome to pce I'm John. This is Julian we're going to get started now, it's 20 plus free, so you could take your seats.

B

Ok, so we'll start with no well as we always do. Please note well that this is an ICS session by participating in the ITF. You agree to abide by those processes.

B

Anything you say in this meeting is considered an ietf contribution. You agree that the meeting may be recorded, etc, etc. You can read it off.

B

Ok, so this week, at ITF, 9004 are two pte sessions. First wanted right. Now is our usual pc working group meeting, the second one is going to be on friday at 9am. This is the joint session, with jointly with TS I'm mpls and see camp to discuss yang models related to our our general area. So, if you're interested in young, then please don't leave town, please stick around for friday.

B

So please could we have minute takers if you're, if you're able to take minutes that will be excellent.

B

Anyone who's keen on doing that. Please head over to the etherpad, I'm also on the etherpad. So what you need to do is find the PCE a gender on the ITF website and click through to the link for etherpad minute takers from from the sort of the main ITF HTML agenda.

B

Is anyone in the Java room a few people? Okay, guys? If you see questions or activity in jabber, please be prepared to relate to the mic. If you could, that would be great thanks. We are streaming audio and video. Please speak only using the mics. Please state your name before you speak, so that the people listening in can hear you if you're presenting please stay inside the pink box, so that the cameras can pick up your lovely faces and yeah. We have a pact yeah.

A

B

Have a pact agenda? A few drafts fell off the end, so we didn't have time for everyone's requests, but we did prioritize according to what you know had had discussion and you were coming in. So please consider you know if you're going to make an agenda request, she'll cause if there's been some activity on the mailing list prior to that, okay. So there's our first slide and the next slide sour agenda.

B

So this is pretty full. Anyone want to bash the agenda.

B

Nope, okay, so we'll proceed the working group status using the mailing list. That's this there's nothing particularly driving me. It's just the usual mag that I always do. Please use a mailing list actively to discuss all the drafts that we're working on. It is kind unhelpful if private conversations happen and the new drafts new versions address get published because it's kind of hard to see the the process by which we reach consensus.

B

So please, if you have a comment on a draft, please remember to copy the mailing list in that common. Ok, so our document statuses, we have one urs C&RT 8051, the State Board pc applicability. Probably shortly after the last IETF meeting, we have to internet-drafts in the RSC editors q. Both are missing a reference to stateful PCE. So we have a piece of sense where and the interlayer extensions both ready to go just waiting for staple PC to catch them.

B

Up with the iesg we have stateful PCE and staples Inc optimizations stateful PCE has been approved its pending a document update the authors of indicator. Bell update the document by the first of April. I'm assured: that's not an April Fool, so we're hopeful that I will get done in the next week. Staples inc optimizations has just been updated. Following my ESG review and I see it's been submitted and a proof of publication. So thanks about deborah and then finally would be iesg bears pce initiated LSPs. That's currently pending a routing Directorate review.

B

Before we proceed to ITF last call.

B

So within the working group there are a bunch of drafts which are at a fairly advanced stage, so we'll go through these one at a time just give a quick update. First of all, GM POS piece of extensions. Those past last call is very mature, possibly a little overripe and is pending a shepherd's report so as well. Ok, so as soon as that's done, then that will get pushed to the isg piece of s.

B

We promised specifically escape this update on this, so we now have a document shepherd Cyril, who is not here, I, don't think that um Cyril has stepped in to give a shepherd of that he's. Just reviewed the draftees sent some comes to the authors drew nodding his head, so you must have seen those so I guess: Cyril needs to see a new version about draft and he will prepare his Shepherds report in parallel and then we can forward that to the is she as well, so we'll all be very keen to see that published.

B

W.Zahn r-wi extensions draft is stuck waiting for the GM pls extensions. Otherwise it's ready. We already have a shepherd's report, written thanks to Danny le who has been our kind Shepherd for that draft and finally, RSC 6006 biz V point-to-multipoint draft affixed to that RFC as past working group. Last call and I expect to be something: that's bhai. Yes, you very soon to.

B

Have we got draft authors for the inter area is applicability in the house and, if so, did they want to give us an update on that No? Ok, so I mean the status were bad. Was that Bell we passed working group last call I was happening. It and I had a review and had a lot to say about it. So I think we need to see. I'm is mostly editorial stuff, but we need to see a new revision of our tract I think before we can take it for ok, so not on the agenda.

B

This time that we have to, I beez related to pc association, LSP associations, diversity in policy and the yeah, the staple pce or say bandwidth draft is also a new ID, an RC 6006 business and error buses. Well, it is a new ID, but it's already in the office course anyone got an update on any of those. First three drafts fours.

B

Ok, so I guess they're fairly static, and then we have a whole set of kind of middle of life drafts we're sort of looking to see. You know which ones are going to proceed if any of fallen by the wayside. So is anyone in here who has any knowledge of the state of let's pick one at random.

C

Jeff: apologies for not finishing seventh rollicking Alta Nations this week, yoga the earth wealth.

B

Okay, so you know just to remind the authors and I am one of your so I'm, reminding myself segment browsing in the OSP set of type have an early allocation from iono and that's going to expire in August. We need to publish by august or beliefs, we code points. So, let's, let's, let's get it done. Okay,.

D

Drupal I prob this is regarding the pc p2mp. We will wait for RFC 606 this blue get over, and apart from that, we have an implementation and we feel it's let's table. So we can progress that as well after that. Thank you thanks, prefer request, be we are discussing with the association group object because they are bunch of documents dependent on that. So I think we have to prioritize that document as well. Okay,.

E

This is your own regarding Stephanie CJ pillars, so we're going to revive it and then update I think share, left wrong idea so see how can provide, but will so.

B

Someone's going to take the reins from shannon on.

E

We need to keep on name, but the depends on what you want, but she.

B

Said she's having de penser some run.

F

So I think we're done with the working with status. We move to the press presentations depending please.

G

Let's wait for the slide.

G

So I think the first one is about diversity: yeah, yeah, ok, so this presentation is to give you an update on the changes that we don't know if we are yes, Association diversity.

G

So just to remember so, the overall goal of this extension is to provide the ability for PCC to tell the pc is at a particular LSPs belonging to a particular disjoint group. We're a little LS particular group could be so. This is very global ID.

G

So we are reusing. The base Association, who draft and the ID, is to use a particular Association type Tara. This is a diversity that we want to worship so compared to the previous version. So in the first version we used multiple Association types, so one for each disjoin s-type. What about sodium disjoint? I saw SLG join s based on the comments that we received on the list, so we moved to a different type of encoding.

G

So now we have a single Association type, which is past diversity and we are including via diversity, flavor in some additional GM. So this is the tlv that we are using. So it was already detailed in the previous version, but we are changed slightly the flags, so it's called vida Jonas information tlv and we are using an encoding which is a line to what has been done for ya as wecker object. So we have our flags so for SLG, not an integer ness and we are keep using the shortest path.

G

Bitterness trick nesbitt that were already detailed in the faster in the first version.

G

Based also on the discussion and I think it was a comment from from Adrian about our weekend couple this with the object function, so we give now the ability to add an objective function, tlv within the association group. So what we are proposing that we can discuss this option, we propose to use via of' least yearly, which is already existing, but we are. We want to limited to just a single entry. Maybe we can find a more suitable option, but reusing via of' object was not really good.

G

So this was the options that we are taking for now and we are adding the earth functions. We are reusing the one that have been proposed in via draft from who about our functions for for diversity.

G

There was also some data requested, also by Adrian I, think about what is exact meaning of his shortest path bit. So we added the dedicated paragraph hobbies, so the global ideas to tell for one or more particular LSP in the group I want absolutely to use what we call the shortest paths with the shortest path is in fact, the path that is taking into account all the constraints of LSP except we join us.

G

So this means that this particular ISP with the pivot set will be will take into account all the constraint except the disjoint ness and all the other LSPs will be disjoint on this one. So this is the global meaning, and this is a clearly you need to provide some one path which is very optimal of some customer. Otherwise, if you don't control the disjoint edge, you may have to complete a non optimal path.

G

There was also some requests to deal with the case where the pc is not able to find a path or also you can provide a path, but it can relax the design s constraint. Who are it's also? It's possible, for example, when we are setting the SB 20, so the designers constraint is not straight, so the pc is able to relax it. So what we are proposing is a 15. The next slide, so in case the pc is not able to find the disjoint paths.

G

So we are reusing, no path, vector tlv and we are defining two new bits will be clv, so one to indicate that it's not possible to find a disjoint paths and one which is it cannot compute the path because it doesn't. It does not have the requested algorithm. So, for example, we are requesting an SL g join s and it is completely not able to do it from an advert legal point of view and regarding via constraint.

G

Relaxing so when the disjoin s cannot be found and the pc is providing a path which is not the joint. So we need to inform the PCC that this path is not complete, fulfill fully fulfilling sorry the constraints. So we are proposing a new relaxed constraint theory which is more generic, so it's not only tied with julius kahn. Strange may be used in overall use case where the particular constraint is authorized to beer to be relaxed. I think we need some discussion on the encouraging obvious.

G

So what we are proposing today, but clearly require some discussion, is a just to put the object that has been a so via objector, responding to the constraint that has been relaxed. So this is a proposal better I think we need some discussion and visa on cuddling proposal.

G

So about the next step, so we think that we addressed as a comments that we receive with the Scalia need for some description and enhancement about so as a new proposal so feel free to to give a your feedback and will be a welcome from the document.

F

Thank you, Stefan any feedback command question.

F

No everything was clear, so in that case I proposed to move to your other presentation, yeah.

G

So the next one about state synchronization between RPC, so the god of document, is first to define the procedures for stateful PCE communication between two pc nodes, so because a state full communication has been defined from a pc to pc point of view, but we don't have the exact process. Your follow a pc to pc communication. So this is the ball, and why doing this? One of the driver is to bring more resiliency in the design. So in case I have some pc to pc session failure.

G

I may be able to send some state using this interview, see communication. So this is one of the primary goal and we over topic in this document is to deal with a path, computation, optimality and also some computation loop scenario where you are falling into some split brain and you have one pc computing, one LSP 1pc computing and over LSP, and the two pc and the to LSPs are linked to each other. So, for example, is a path of one LSPs.

G

Depending of yoga, you may fall into computation loop and we are proposing some math of slave behavior between the pc and inkay. In this kind of situation, so the big work that has been done from the last time is we merged with some exiting proposal. This was a proposal for hundra about Olivia real time synchronization. So we have something now, which is a liner. We have a single proposal for all the interp see a stateful communication. It's also include part of year Aly article on PC, which will be based also on this. On this extension.

G

We do we done some slight changes in VR deposit, also globally, in terms of capability. We still have a bit to signal that this is a particular interface instead for communication, when the session is set up, both pcs will behave as a PC on a PC. At the same time, so we will have bi-directional synchronization, so each PC will synchronize the realistic. It's a nest, a synchronous process, oh yeah.

G

There is no particular order here and the idea is that to identify the original PCC address, we are relying now on via speaker, identity tlv, which has been already defined. So we already used tlv tua to carry the original position well sooner or pasetta messages.

G

So now, how does it work so globally from a LSD state synchronization? So when a PCC receives a particular states with a piece of report message, whatever the state of the delegation, it will forward it to all the pc using the states in session and the important hole is that we do not authorize a pc to follow up this piece apricot message to another pc. So this is primarily to prevent message groups and it's clearly simplify the design.

G

So when the pc receives on a pc see it can send to a pc, but when a pc is receiving from a pc, it will not send up to another pc.

G

Based on this, a particular LSP state may be received by multiple sources. It may be received directly from the PCC and it may be received also from another pc or even multiple pc.

G

We decided at least we propose to only keep track of a single state, so we just keep in memory the last state that we are because, from a network point of view, there is a single state, which is the state that has been advertised by the pc. So it does not really make sense, at least on our point of view, to keep track of all the states. What is important to track is the sources of the state, so we keep one state, but we keep track of all the sources where we learn the state farm.

G

So this means that in k, if I want to remove a particular LSP state, I need to have this source list to be empty, so I need to receive a removal from all the sources that provide media or this state.

G

Now the last point on this draft is what we call a delegation, so this is a master slave behavior that I introduced in the beginning of a presentation so globally to solve this path, computation optimality.

G

Sometimes I really need to have a single pc computing, a set of particular of LSPs that are working together if multiple pcs are trying to compute path for the set of LSP, let's say to see one computes path for LSP, one pc to compute the path for LSP tools that are really working together. We may fall into computation loop or some perfect mates. So in order to service, we want to ensure that, for this particular group of LSP, we want that one pc will perform the computation.

G

So what is proposed that, when a PCC is delegating an LSP to a PC, OPC can decide to sub delegate the computation to another pc. So we are just defining in this document with sub delegation procedure, but this election of master slave is completely out of scope. It could be a manual configuration telling that this particular LSP must be primarily computed by these species. That has this IP address. It can be done by maybe a protocol extension, but this is out of stuff what we are defining the ears with just master slave behavior.

G

Well, when a PC receives the delegation, it can decide to serve delegate it to another pc.

G

And when this pc sub delegate v LSP to another one, it lose the control on reality. So by itself it cannot compute the path anymore. It cannot update the LSP anymore. It will be the whole of the mass Topsy.

G

So how does we have date work so globally we master pc will be responsible of the computation for ver LSP of the set of LSP. It is responsible for it will update all the pcs with the difference that forms the pc that perform the sub delegation. It will set the delegation flag, one and four viewers. It will side with the aviation flag to job.

G

Why are we sending an update towards your PC is just to speed up the sink the state synchronization, because if we don't do this, we will need to wait for the PCC to receive your date and then propagate it back to VDC. It can work, but if we are updating all the pcs from the start, it will be faster. That's just the god and after it's a question of cascading a PCO date.

G

So when the slave pc is receiving the pc update, it will see in the speaker, identity tlv, that this is an update for a particular of PCC and it will relay the pc update to the original pieces.

G

So this extension is also based for a miracle picea stateful work. So if you want to do the article stateful PCE we'll have to advertise interpreting capability and in addition you will add the article pc, kappa dt- and this is detailed in the artists estate, fula hpc right.

G

When we built this architecture, we thought about first message: groups that we try to avoid best, buy this kind of spittle reason mechanism between a pc, but we also thought that it may be optionally interesting to record some information when the messages are propagated through multiple apps. So we are proposing budgets pure an optional behavior.

G

That each piece is each piece of speaker may add in a new tlv, some precepts picture information so, for example, its IP address, or if we want to I, don't know we got some particular information on yes, this is something that we can.

G

So just a summary, so this is clearly a need for us to have this interpersonal communication to an ends via resiliency or the a path computation, optimality, ax and so on. I think we are now on the good track with this stone that interview see procedure, whatever I saw the article flat or whatever. So yes, it's clearly more generic now, and we would like to request the working group adoption for this document and if you have any feedback, good or bad, feel free to give us.

F

Any comments in the room I see someone in stunning.

H

Taking about you slides up to the time when you're talking about how we select the single stake, but yet.

B

Maybe you use this way, yeah.

H

We say we keep a single state, the last one level, the last one level is necessarily there curing state. So why why you are stabilized? Where.

G

It could not be the different propagation. We are different parts cooking. Yes, mmm so.

H

Pc she sends word masterpiece. Assistance was like a message to the slave. Somehow goes fast. Yes,.

G

But the state, or whatever the order of a message, the final stage that you will have will be the state of the PCC, because the pc are not changing me. Ok, I'll.

I

Do you mean inkay.

G

Up I, if you have a happy yes,.

I

H

One case of clapping: yes in case of flapping, but if you start fluffy this make it out.

G

Yes, we thought about this and one of the ideas that we add east for the PCC to add a kind of time yeah. This was an idea that we have in mind too high to solve this. It's got in the document yet, but if you also think that it's an interest.

J

And, like yeah point, we can edit.

G

F

Thank you any other comments.

K

Okay, I love, improv avi since I think that is actually interesting. The solutions, but here, from the my point of view, I have some.

A

Days about about.

K

This one because I think that in either of the occasion scenario once this, the connection between the pc and the GCC is broken for the pcc.

K

In fact, little harder is harder to determine you. The pce one is the active or usually start. You can kind of check more funny, because maybe if the study, that is not you the broken of the link, so I'm not sure because only the pce maybe walk about the impact, our the pc and pcc. A lot of other links in.

A

K

Some, the PDP on net half and some the data needed to synchronization, but only the PCE walker I'm, not for the effect. The effect that is ok for.

G

K

You know all the future.

G

K

G

Sure to understand your point, but if, for example, here the pcs completely failing it's not an issue, if it was a messed up, you see this one will take over and it will compute the path. Oh.

K

Yeah, that's that yes, but but my in fact my comments to you that the in fact that the scenario- maybe the maybe more complex than we assumed sensory in the different application scenario and also the pc and a cutie, see what.

G

Is sure is that you will never be able to solve all the kind of failure, because you can find some multiple of failure of scenario where you are losing this session this session, and maybe this one- and in this case yes, your computation- will not be fine but yeah. You can try to enhance by bringing more PC more piece obsession, but at a certain point you will never be able to cuddle. Also fail your case. Ok.

K

I will check acabe car. We want middle east of his car shop.

F

Could we have a show friends to be from people read the document? Please.

F

Not so many, please read the document here because when we exclude the authors this I want some money. So I think doesn't make much sense to build the room further today, but there is some interesting work in there. It seems to address missing feature from the pset toolset. So please read the document, share your feedback using the mailing list and think use the van problem representation.

F

Next presenter is a sentire guess.

B

What will win? The next presenter could I remind anyone asking questions, it's a good idea to go to the eighth apart and make sure that we capture your questionable responds correctly. Otherwise, we may miss some subtle tea. Thanks.

L

It's I tinea yeah. This one talks about the pic ability for a pc to request the control of an LSP from the TCC. We go a few use cases and then the proposal on the table so I just kept the slide. We talked about the use case in the next slide, so most more operators are moving towards the centralized 30stm control and also looking for new enhancement and one such enhancement or was awesome and one such enhancement is for the PC to take control of an LSP for global optimization.

L

We've got there's some scenarios where the pc has a better view of the network in terms of futuristic demands, surface demands and this draft talks about the ability for a PCE to take control of a specific, LSP optimizer and then maybe written it back to the PCC if needed. So it's it's a very straightforward rap.

L

So, mostly looking at global optimization of LSP, the pc able to control some specific LSPs in the network, we've got a few other use cases also defined in the trash one.

L

Such one is the pc load, balancing cases where you could have a specific pc requests, the subset of LSPs- and you know you- can they can load balance across multiple pcs to be able to optimize some LSPs in the network and also the failure cases well, if I lose of tce, the neighboring pc or the standby pc could always request access to optimize the specific error stream network.

L

So going back to the proposal the straightforward proposal, what we're proposing is a control, LSP control, flag set as part of the srp object. So it's gonna be part of the the update message and pc sets the flat 12 indicator in control of the SP, and this is identified by the LSP object. So it's a very straightforward laughs so asking for a new control request flag and it's up to the PC to delegate and not delegate so you're, not change any other mechanisms in the PC stateful track.

L

So this is just an add on to it to be able to control. But then you get specific elephants. So any questions comes I.

F

See someone coming I taking portunity to just do this, that it would be good to avoid choosing the bit number you allocate. You should leave that to Jana you request for a bit. We had this discussion on the list, but it is still on the draft on the presentation. Don't choose the bit number now all right leave.

I

It for later thank.

F

H

Related to that so before we allocate the bits, so you started your presentation saying more of operators are deploying this, which can you be a little bit more specific YouTube. We only have as many flags as we had and I.

A

H

To make sure that before we start allocating flights, we actually solving real problems that operators are seeing versus technology heretical problems. If you can, if you can maybe add some comments on you jab from statement more operators to Bill to Bill in the case for extra bit, yeah all.

L

Right so at least from where I stand, we've seen this time and again where we try to optimize something less pieces, a network but yeah we could always. We can probably flag by some of this on the trap and- and maybe we can make for your comments.

F

Okay, so I believe you have a couple of minor items to address, to justify a new version, use that invasion to try to gather some support on the mailing list, and then we can see how we can progress. May I just have show friends on people read. That document drew a.

F

Dozen okay, among those who read, do you believe this is a useful feature.

F

Almost the same yeah, maybe couple less but almost the same. Okay, that's a useful feedback. Thank you.

F

Next presenter is drove, I guess I'm doing the head, but yeah.

D

Hi, this is truth. This is a code point, a location, I think to ITF ago we presented that and we got the feedback from the mailing list as well as during the meeting. So we have incorporated that in this version. So let's talk about that. The main idea here is that we have our piece of registry. As we all know, the usual allocation policy that we have followed is ITF consensus, that all assignments are via RFC's, which are approved by is G.

D

Of course, we have earlier locations which is very useful when we are implementing and we want to get the bits set, but, as you are seeing recently in the pc working group and otherwise the experimentations that are going on with the use of pc in various cases, we also do realize that not having some bits set aside for experimentation can cause problem. So we are hoping with this update if we can have some bits for experimentation, use for pset messages and objects and theories so that we can start using them.

D

The main aim is, as as the experimental usage describes is to facilitate.

D

Environments and main aim is that these values will not collide with any of the other I na locations that may happen in future, as well as the current once and especially now. If Ben, we are seeing peace app also in the open source. We want to make sure that when we have features which are not standardized or who are not at the level of I na location, the kind of bits that we use belongs to the experimental so that it will not create any confusion.

D

So that's a main aim with this document and when we we saw the problem in the open source that this was happening. So we that's why we are writing this job to solve this as quickly as possible. So what were the feedback? And what's the update, we were trying to get more feedback and trying to use that what other? What are the registries that we want to have for experimental? The kind of feedback that we got was focused only on the essential? Don't try to do it for everything. So that's what we are doing.

D

Only three things: p, sep messages. He said objects and a generate p sub tlv. Those are the three things in which we would like a small experimental range set aside. We had an appendix which was talking about what other issues can happen, and what do we do with object? Type notification, so we have removed all that detail from the drafts. Now we were also thinking of whether there is a role of pc working group wiki there.

D

So again, the feedback was not to do anything like that, so we have removed that completely and we have aligned to 36 RFC 36 92, which states that that this range is only for experimentation and the device needs to consent between each other, that what is the meaning of those bits and other implementations may use the same numbers for different uses. So that's why there is no need for us to maintain a wiki and that's the policy that we are following with this update.

D

What's the update looks like expected for experimental allocation for messages? Are these four objects around 31 bits for messages? Eight and forty Elway's allocation policy that we would like to set is experimental as per 5226.

D

So basically, we behave what the feedback that we got from the last time we have incorporated- and I hope the working group is happy with that and with that we can hopefully adopt it and move it so that we can avoid any potential issues in this area in future. As a part of ongoing effort in owners, we have already started using these values for the experimental extensions so that we can avoid issues I think that's it. Thank you.

B

Ready any comments or truth.

H

This was one in.

A

Nine place and.

H

Rude Organon, okay- and this is great work very.

A

Important very needed a.

H

A

H

As in any protocol, we want to add expect that behavior, because you may have an experimental things added a cup of the light network. So you want to describe the behavior. What happens if the receiving end do not know how to interpret X an experiment, our bits no set to some yeah.

D

Luckily, I think the RFC 5440 handles all those cases quite well like unknown messages in case it's from the experimental range or unknown. Tl vs unknown objects, so since the range is already from the not allocated not use. So if somebody who is not following this experiment even follow RFC pie, 440 behavior.

I

Quickly, scan for this.

H

I

Drive not mentioned that in the Dahlia I.

H

Don't think he's big mention he's, probably just for yeah. It.

I

H

No yeah understood no, you know just clarification even.

F

If she's already covered, you make just hide and sometimes 2.22 that to make sure everyone knows, we are reusing existing mechanisms just for gratification, so.

B

She's reasonable.

B

Solid, let's know who's read the draft.

B

Okay, I guess about eight people have owned up to it and and those people who would like to see this draft adopted in peace EE. That's the same people, okay, um yeah I- think that we badly need this. It's really important, so we're going to take this to the mainland list after the meeting thanks truth.

B

And europe again: yes,.

D

So this is an update to a bunch of documents, mainly stateful HPC, and the related stuff do actn that how we are able to use the PC architecture and the PCF extension so far with the stateful, HPC and Association objects, etc. On behalf of all the co-authors, I am presenting this update, since there are a couple of documents involved here, I wanted to just show what is the relationship between these documents?

D

We have an applicability document, which is an informational document which talks about how do we apply the various piece of extensions in architecture inside a CDM and I will talk about the update to that. But, apart from that, we have the various extensions that come into picture. We have the PCF LS part, the stateful HTC part and the VN association part in this presentation. I am focusing on the yellow, dotted lines, but peace, fls and even the drop which must be which differ and just presented. Thus, the state sink is also related to stateful HPC.

D

So all these drafts how they play a role in the actn, that's the document map of which, in fact, some of the documents. As as as we discuss our kind gendrich, they are not just for. In CT n, you can use stateful htpc just of as a part of your p sub network, and it doesn't have to follow the whole actn architecture and see Emma and all that stuff as well.

D

Similarly, piece fls is just a mechanism to carry link state within the pisa pisa pls is one of the ACCDN is one of the use case for it, but it's an independent work as well. The VN association is something that is kind of specific to a CDN that we have so starting with a stateful HPC.

D

The document is pretty straightforward: it's an informational document which talks about how the HPC and the state for pc can be merged together to form stateful HPC. Basically, it's an. We already have a hierarchy PC. What? If those pcs are stateful pc, then it is a hierarchy of stateful pc. This was discussed, I think in Ghana Cyrus, and we had good feedback. The kind of changes that are needed in protocol stefan is doing it in a generic way, so that can anyway, we used in this document. This document, that's why it's just informational.

D

It doesn't ask for any protocol changes in the document.

D

So what are the oh I'm running through? Yes, this is a so. What are the changes that we made? There were comments related to adding clarification on how the recursive nature works and it can go to multiple levels as for the hpc architecture, so the idea is all piece of messages. Delegations of reports updates can be done in a recursive way, similar to how fun presented in the state draft the only differences that this is a parent pc and a child pc relationship rather than a master and a slave.

D

How do we know? We clarified how the capability advertisement would be done? We already have a HPC capability tlv as a part of pc her. I key extensions and we have the stateful PCE capability tlv, as well as the the bit that is added for, inter pc communication. Together, we can use that for advertising the stateful HPC capability.

D

How would we manage the state synchronization as per the previous presentation, that you saw and are there any considerations with respect to scalability etc? So we have tried to handle that how that can be taken care, what kind of information that needs to be pushed up to the parent pc? So our recommendation is that we need to report only the LSPs that the parent pc could use for its optimizations, which is usually inter domain or the LSPs that he is personally involved in.

D

You don't have to push all the LSPs from all the child, pcs to the parent pc, and apart from that, the pisa protocol allows you to have notifications and the other way to handle scalability as a part of stateful pc draft itself. Those are applicable even in stateful HPC case, and we also took care of the security and the manageability considerations as well. So we have tried to make sure that all the comments that we have received and the documents get a little bit better. So that's the update for stateful HPC.

D

Moving on to the VN Association I. Think the main concern here was that what would be the use of having this kind of association with in a P sub? So, within the actn working group we have talked about what is a Wien and how knowing which set of LSPs together form. A virtual network is quite useful when we are thinking of reorganization sore when we are thinking of how to represent a virtual network for the end user as a selection as a as a list of BN members.

D

So having this association as a part of DC protocol could be very useful.

D

What are the cases we would like to optimize all LSPs belonging to a particular VN, rather than doing this / LSP so having this kind of association, I can have a better optimization functions and better policy to handle that and any VN specific action that I would like to perform. Having this relationship between the LSP is known in my her I key of controllers is useful, so passing this information from parent to child is is needed, so we uses we use the Association object itself.

D

Add a new Association type for VN and rest of the procedure remains the same. We added a charity to why we are doing this. What are the cases that we would need to know this kind Association, whether it is for fresh path, computation or during the optimization, where we would like to not just optimize a single LSD but optimize the whole VN, that is, collection of all LSPs needs to work together to get the best results for the easy TN scenarios.

D

So that's our main use case and that's why we feel passing this information could be is needed. We also added a vendor information. Tlv support within this association object so that any vendor specific things we can pass from parent to child as well.

D

So now this is the document that brings everything else together. It tries to say that how do we apply the various PC extensions of which are in place within the a CDN architecture? This is the actn architecture. It has basically some four functions which the MDS see and or the pnc is capable of. I will go through it in the next slide.

D

Basically, in this document we list with in various ways in which ECF can be used, but we also know that you don't have to use P sub for all the functions it's up to you. But this aim of this document is list all the possibilities, and then people can pick in for one of those functions. They make still continue to use yang or BG pls, but first another functions. They want to use P sub so that flexibility could also be useful.

D

So what are the four functions? The functions were: multidomain coordination, the virtualization and abstraction part, the customer mapping part and the virtual service coordination part stateful HPC, with the initiation functions under per domain. Stitched LSP mechanism is quite useful to achieve the multi domain coordination as well as for the customer mapping and translation. The P sub LS part with the support for virtual networks and the various abstraction methods that we have talked in.

D

The t's working group can also meet the virtualization and the abstraction function of the actn to map between the virtual networks and the LSPs, the VN Association. Our draft help to achieve this. So this is the map basically saying that how the various actn functions can be achieved. Why are these extensions, which are already available in the P sub space?

D

So what? What are the changes that we did as I said that we clarified that PCF is not mandatory? We are not saying with this route that p sub needs to do everything, but we are giving a possibility that if you want to use P sub, these are a set of tools that you should use.

D

We also added a reference to other drafts, like Association policy or the state sing, as well as the various abstraction methods, so how that is applicable within a CDN context. We also gotta come in to specify how does this relate to ECE base central control, so we added a separate section that describes what is the relationship between the two and this also handles those comments. Okay, so final side, we want to know what's the best way, to progress our documents. The main thing is that some parts are independent.

D

They are not actn specific per se, so I think the stateful HPC and the piece fls they have their validity or on their own. But then we have the acct and specific drugs which are the acct and applicability and the VN Association. So we need to figure it out as a working group. What is the order that we are progressing in or how are we handling this?

D

You may have seen the some implementation reports we presented in the last ITF as well as well as there were some discussions on the mailing list where people are saying that part of what they have implemented so far. We feel that we have able to handle all the comments that we have received. Now we need to figure it out in which order we could see the work moving. Thank you.

B

Any comments, the truth.

M

Hi, my name is michael zhrov. This relates to.

A

The question that we had.

M

Earlier in the morning or the cmi interface, if you go to it, I think slide number 9. Where you show you know it was the one way, but they are yet that's the third one here and actually I thought that the VN concept is on the cmi interface in actn. Now you extend p sep for the yen concept, but you don't show it on the interface. So this is this point said this morning: it's confusing what exactly the role of CNY is and why?

M

What a VN concept this and I also think it's still unclear why you need that extension in piece of here, but the bleeders can be clarified by this I. Think it's an issue in the actn architecture as a whole.

A

And by the way, I mean the.

M

Mmi interface is not well defined in actn. I.

D

Leave the actn part of that. Maybe we can focus on the piece of yeah. Yes,.

M

D

M

About don't have any concern for the feasible right now, but as tell us that I think there's an unclear T in the AC Chen part of this and I think for the VN part of the presentation. It's unclear how this exactly ties into actn, but this is an easy to issue. It's not a piece of issue.

D

Handled that as a part of easy deal, yeah.

B

I was always going to say that sounds like a general actn issue, so I think I think probably the way we would proceed is by looking at.

B

The applicability as being sort of the primary document for now and and staple hpca is being sort of an independent track and look at those two and the VN I. Guess, um let's get by clarified in the actn architecture, that.

D

Looks uglier elbows, okay,.

B

um So on just just to check quickly has anyone who has read be applicability, statement.

B

We have about 10 people, okay,.

M

B

All those tan who would like to see the applicability statement adopted by pce.

B

It's about almost.

J

B

Same okay, good and how about the stateful hpc II, who is red? The stateful, hpc e, rather more? It's good, okay, maybe 15, and who would like to see that adopted by the pc working group.

B

Okay, it's like about 10, okay! Well, that's a good start! I think we will confirm both those on the mailing list. Thanks truth and next.

J

Up is usually on.

F

Okay, so I'm working here for my colleague you wanna hear you could not attend the IETF meeting this week, I'm going to introduce you a proposal about integra.

A

F

That using the pc, stateful architecture on.

I

F

So I don't want to remind you all the beauty or PC architecture to address the entire domain. Programmed provision to the many recipes just 22 maybe put more.

A

Emphasis on the.

F

Specificities to the interdomain, especially in taro, is very strong restriction and what protocol can run over those interiors boundaries so typically LGBT. Isn't the cause really consider option to enforce NSPS across a s boundaries? So there may be some cases where the PC architecture may help, even if we don't have a DP messages across those boundaries, so today, typically to address this kind of stuff. If you have to macys in your group that won't rather nice be set up end to end. We under under.

J

Them separately.

F

And we may stitch em, possibly using some bgp mechanism there, but as soon as you want to do some key there, there are some missing points.

F

So the idea here is to label the TR stitute to be applicable even to cut all those interests case use cases, so the state will be RPC the proposal that extends the backward recursive procedure that is already defined in RFC 5441, if I remember correctly, which allows to its draw chain of domains using the EP set messages having Purdom npc talking to each other so as to get some Milty domain global pass competition on here, we propose to extend this procedure by having the state for messages, namely the pc, yet message to trigger a new TRS provisioning, SB versioning over a chain of aces.

F

The idea is to make use of this message is to trigger the missing piece between yes to make the blue between those is and get what we call in the draft stitching level so as to be able to agree on the label to use to carry the traffic across the aes boundary between the neighbor Hayes's.

A

F

It relies on the existing on a roll TC message and we propose to reuse the USB type that has been already proposed, as we were talking. You said before, to distinguish between RSVP related to LSPs as our type as piece here. Keeping those two options we propose to have.

F

You had other cut point so as to cover sr4 entire domain cpt for inter domain, honest, don't cut pile to trigger labor location from the ASB are of the downstream the III as we are of the downswing denying so as to get label in the pc you import on senator pc from the blame on them up to the upstream domain, so this is probably more clear what we're using drawing.

F

Supposing we have a chain of three races, each AAS as its own pc. The waiter comes to the first one once the trigger dsb provisioning. We have the visual all piece at Wragby, separate exchange from be able to proceed, but the offense we may end up with a shorter past three and possibly using passkey anchoring to I'd the details with respond over the downstream Oasis and based on that the first PC to change figure.

F

A PC initiate we're using the City PD resulting hero to push some provisioning to the downstream domains on this peace initiative relay to serve them. In here we may end up laying about the main sir, make you safe passage routes procedure on back in each domain. Each autonomous system is able to push its configuration than to its network node using the pc yet to the Anoat.

F

The beauty of the idea that one, a yes may choose to rely on as a pity problem the PSP segments, while the other yet another racing chain, mary, died on second rotating to apply ESP conversion between it's the problems. So based on that, we have a report the edge of years via of the s3 here, as identify the cut points for the entire domain case on in the report. It includes the label to be used on the entire Dominion link between a stew and the s3.

F

Here we have the label which is allocated by the downstream domain. It is reported up to the PC of the domain. Three on Wendell III gets back to the report to its actual rain evil, a still able to use as interior areas, bloom.

A

F

Pc to push the state down to its network nodes, it realized this stitching rabbi labeled onto its nose. That is installing the network elements. So the procedures also learned. We are all available to me uttering the main. We use the same procedure, but in the hatchery initiating the main. We don't need any specific, SP type, let's be typed, because the upstream, the name is a need to allocate a label for another domain, because it's the first remaining two chain. So we end up in a global em to analyse, be relying on me.

F

Your basic mechanisms that are already find any shape to port on the opposite procedure.

A

F

Add TC the type option to trigger then stitching of a location. So the proposal when you lie on this draft, which should should end up in ER, see quite soon because we have already highly location pending on this. Do you think a political boss, rsvp-te and sr a autonomous.

A

F

So system remain autonomous in there t2 inside there on the rain on. We also have a list of improvement to become 0 for further Emily's. If there is interest on working with to to walk into that direction, so issue our feedback, please feel free to share it or send it by email. If you prefer. Thank you.

J

Any feedback visually.

N

Yep anti from powys, so I think actually at this job is very interesting, so I think you are trying to make the piece if to have the essential on functionality of us with here right. It means that you are trying to use hyssop to create end-to-end connections across multiple domains. Yeah.

F

You may describe it like this: the glue between rather at the TC level, instead of at the network level, the network level they are just seeing what label is at the boundary on they don't see. The river downstream or same naval domain has a part of the industry yeah. So.

N

My question my comment is so on campus ship, the price, a sweetie outlink, to create end connections.

F

In that architectural, supposing love segment loading label on all your domains? You may.

N

F

Yeah, you may have it yeah.

N

Okay yeah, so on another uncommon. Even do you think that if pccc can adjust your own requirements, if we have 0.

F

As you promised,.

F

Maybe there are.

N

Some woman anyway I perspective IDs, made.

F

Are some similar ideas there I don't think pccc solve the blue of the stitching level. Let me try to address here, but we may find some common ground between those ideas. I guess so maybe we could look at that together in detail to see there is some commonalities to be. Maybe progress together, similar to look at more detail at that.

N

So I think if there's a centralized pccc and the pccc can stand, you know kiss it to the pound. Angelo's enter stitching to LSPs. Look to thank Union yeah. Thank.

J

You thank you but I. Please, please be brief. Proof.

E

D

I, like the idea I, wanted that if sa label thing is not tightly coupled to staple brtc in the sense staple hpc can may also need that kind of view. So if we can get the label thing and then the question is how do we write it in a general way? I go back to Fatah that maybe the pccc kind of way was much more generous than this, but it's kind of dated other than that I have small currents which we can handle offline.

D

But overall, this thing one one thing which I wanted to also talk about is that initiate message when it is coming from a parent pc to child pc? The relationship is kind of known, but when the initiate message goes on the fear level from one pc who's hearing you to another pc, we may have a little bit more like the security issues and the manageability parts of that relationship. We may not have to evaluate a little bit yeah.

F

Oh, we on the bottom, it's a zero version. So has you know security is left for further seen you, but you write as soon as you touch anything about inter-domain of all securities, even something to be considered even deeper than for intra domain. Sorry, well.

B

We're gonna have to take the rest of it, the west coast with the iron schedule. Now we don't have time sorry, um okay,.

J

B

Some people Julia, we can, you know, get some more on the mailing list and yeah hope that was useful. Thank you, okay. So next it's Rakesh.

B

Rakesh we're a little squeezed for time. So, if you could like shake one minute of each of your presentations, that would be awesome.

O

Good afternoon, everyone, my name, is Jacques candy and I'm, presenting a destruct and associated their bi-directional LSPs with PC on behalf of authors, listed.

O

So we'll take a look at the requirements and scope of this document. There are two association or initiation types defined in the document that there will have a look, we'll review the PCF extensions, proposing a document and discuss the next steps.

O

So, basically associated bidirectionally recipes have been deployed in the packet transport networks in the RSVP world. The recipes can be caught out adorned on carotid, so the scope of the document is the either pc or pcc, initiated bidirectional other space, and it can also be user. It's great that pc, for example, if you want to do a corrupted park, competition so in case of double sided. The initiate initiation, the pc initiates the LSP on both endpoint nodes. There is a new Association type defined for it and again LSP can be corrupted or non corrupted.

O

In this case, and in case of single sided initiation, the stateful PCE initiates the forward and reverse LSP on one end points of the bidirectional SP. There is a new Association type defined for it. There is also a tlv for the association that identifies the forward in reversal sv as well as can specify here. The karate flag.

O

So as men there are two a new Association types defined and the rest of the field in the Association objects are populated using the RSVP RFC 75 751.

O

So there is a deal we define for the Association object, identifies a forward forward, LSP, reversal, SP or if LSP is caught out it which can be used for path computation.

O

So it's fairly as a short and straight forward a document. So we appreciate the review, comments and suggestions. So is this song slightly useful to have in the pce wall it's heavily used in the recipe in the transport networks and if we agree requests for working of adoption.

B

um So thanks for caching max if I keep in my short I think we need to take those questions one at a time. So.

B

Before we get to the working reproduction call I think we need to see some some comments and suggestions on the draft, but we can certainly ask the first question who thinks that this is useful functionality to have.

B

Yeah, like 12 people or so so yes, I mean that sounds like there's. Decent support for working on. This will be good to to see some comments and support on the list.

O

So good afternoon, again, I'm request Guardian presenting this draft am on behalf of the authors. This is for the PC piece of extensions for reporting, the pls p PM parameters when using the stateful pc.

O

So agenda is as following. We'll look at the requirement and scope of the document followed by the overview of the piece of extensions. The PM matrix are delay, loss and bennett utilization. There is also a multiplication message defined and then we'll discuss the next steps, so our first to give a high level picture or the protocol landscape. On the left side, we have RFC's and graphs that deals with those metrics for the links. Those are the igt and bgp extensions.

O

We also have the service aware draft in pc to use those extensions for the csv f park computation on the right side. We had those extensions for te LSPs, so we have RS RFC, 6334 and 7806, for example, to measure the those metrics what's missing is that stateful pc is not aware of those up performance metrics. So this is what the district is addressing.

O

So, basically, requirements off our that there is a need for the stateful PCE to know about recipes, n2n performance matrix like end-to-end delay and loss and vanity realization. So it does do the path computation, but it doesn't have the interregnum metric information, which is measured by the pcc.

O

The bennett utilization can also be used by the state for pc to do the appropriate dynamic reservation. For example, if we want to implement the utter bandwidth at the stateful PC that does the Bradley computation.

O

So the scope of the dark is the PC to request and report and received at the performance metrics parameters for both the PC initiated and PCC initiated LSPs. What's out of scope, is that how this matrix are used by the stateful PCE and at the pc see how this matrix are measured? There are separate, RFC's and drops available for measuring those matrix.

O

So the overview of the PCF extensions proposed there is a measurement capability which is XA exchange in the beginning to make sure that they are capable of p.m. both. The PC npcc, followed by the PCE, sending the measurement attributes request, which is responded by measurement report from the PCC.

O

So there is a common set of measurement attributes defined for delay, loss and manipulate with utilization each having their own tlv types. So the attributes are measurement modes. The measurement interval the threshold for reporting, including the count and the trestle being absolute or percentage, and interval report interval. So these are common properties for all three matrix. That's considered in the document.

O

So the similar to the bandwidth object, that's defined. We also defining delay and loss measurement objects for reporting those matrix.

O

So this is the definition of the delay measurement capability tlv. There are two flags defined for unidirectional and bi-directional measurement capability.

O

So these are the typical object formats for reporting the delay, so we have defined one-way, average min max or delay variation and similarly a two-way values for them. There is an alignment of those parameters with the link attributes the PM matrix, so we are using the same attributes as well as same types.

O

Similar to a delay, there is a capability TL we defined for loss measurement as well. We have unidirectional bidirectional loss measurement as well as we have infirmary more bust measurement.

O

Now, similar to a link for loss, also, the matrix are aligned with the link matrix where there is a packet loss, is as a percentage of the total loss they reported.

O

Just like a delay and loss, there is a capability tlv also for the bandwidth utilization and the bandwidth utilization report contains a variable length. Tlv. There is a new object, I've defined for the bandwidth object and how many samples are there in the in the report depends on the threshold and the interval.

O

So there is Bishop a notification message define in order to help with scale see if that is ever a piece stateful PCE getting overwhelmed. There is a notification to inform PCC to temporarily suspend the notification and the values are defined to enter and clear. The state.

O

So welcome your review, comments and suggestions. Is this use useful functionality to have in the PC wall? And, yes, we request for the working of adoption.

H

They entered Organa nokia, so a the capability is useful. Would I do it that way? No, especially since I have this available today, and there are mechanisms develop design for scale that do not overload protocol like yourself, you can do it with telemetry. You can do with any other things that this type of mechanisms are not scaling.

H

That's why we can have with a little bit more at scale mechanisms and other pc me has enough of the work to do to set up and react to to LSPs that overloading it with collecting performance information and she seems risky. So I.

O

Yeah, the motivation is that pc is already getting a lot of LSP state information, yrp subchannel, so this is just a distant information that it would have everything at one place instead of having two channels or four different purpose. But that's the motivation.

H

Exactly saying you're overloading stuff and you're going to get their performance issues, they are like it. If you want to have this to really be a centralized control thing at scale, having one channel to hand a lot of things, it's not helping and- and we breaking this purposely. So there is this- this is useful functionality. It's available can be done, I see like I, really don't see any anything new other than adding stuff into the protocol.

H

Doing it differently doesn't mean necessarily better.

J

Okay, please please try to keep your comments as we sure as possible with your general de de las one attached speaking. We do now turn to. There are pictured out the relationship.

J

N

Yet this one no is a list.

O

N

One so in this picture you describe the relationship in your draft and current upstage know.

A

N

That you knew lon, don't mention on p CEOs job because I think actually and measurement information part of the link information. So I think peace, EE LS. It had very close relationship between your job and she say I asked yes yeah.

O

So so bgp igp is there. We can also add a the GPL.

N

P

C

I'm saying pretty much same concerns: ng yep breast. So my question: do you plan to only measure performance from adam respect or on behalf of Rinat in the network? Yeah.

O

So this is the PCC I hadn't sent to end performance from the head end, but.

C

There's no association, this particular LSP, so I presume we will be just distributing your debt are closed, EP into pieces and shouting captain pc incorrect, no.

O

But these are initiated by your pc right, so there is some. There is an association not sure.

C

So class again you are going to send on information relevant to LSPs, instantiation particular not right.

O

C

You are missing the picture anywhere further down in the network if.

O

Those are not directly managed by the pc, then yeah they're.

C

Nice but their.

E

Pricing list right so.

O

Pc wouldn't be aware of them and you would be kissing them. Listen. Thank.

B

J

Who has to have a blue sheet.

B

Sergio then he will not sign the blue sheets. Okay,.

Q

Hello, this is some work about the pc, exp CP extension for there's air SP scheduling versus thanks all the authors and the contributors to this work. This is sir. ah This is what we have discussed in previous idea meeting and we have received some major comments. The first is the motivation requirements and the use cases we have troops prepare.

Q

Also, we have just a documenting tease to discuss the the motivation and requirements, and finally, we get some consensus in these groups, as we decided the architecture for this problem and people think this is a good direction to go. So now we come up back to tease our PPC to our to to provide the solution. So the second question about the solution is how the pcc schedule will be synchronized between pc and a piece PCCs.

Q

The answer is, we use the pc report and the request and updated messager to do the synchronization and the last one is. There are two jobs on the same problems, so we work with other authors who provide a merge diversion which is available on the web sites. People can read that much fortune and there are some updates in the new version and the the most important is yeah.

Q

We two jobs has merged into the single solution for the problem we have discussed in case and the extra features for this version is, we add, the terrific Allaire SP scattering from the high most jobs and also we provide a quick graceful period and the alaska time interval for the for the scattering which can make the people can have more choices for the RSP scattering work and we also add a new section for the RSP update.

Q

So what if we have already have a scheduled, a resp, but we want to update the information and some editorial changes. So this is the home page of the PCA extension for there's a resp scattering, and this is the architecture we have decided in the teeth dropped and during this, in this document we just provide some peace, app extension border up to carry on the time information between the PCC and the PCEs.

Q

So next step, so yeah we have make some congruent is for the architecture and the program itself. So now we come up to the pset to provide the extension and provided a solution. So now two jobs being merged into the single one and a single solution for this problem. And then we want to ask the group whether it's ready for the adopt.

F

We may get one comment if someone wants to there's been a date today, right of the document, so I don't stick. Many people have actually read this latest version. So apart from suggesting people to read that one, which is the the first step to the merger of the two initiatives, which is a very good step for one I, don't think we will do morrow today, but thank you I think we are right on time. So thank you very much hope you wanna sign the blue sheets.

F

If you have comments on this graph or other, please use the mailing list, try to use it as much as possible between the meetings and we'll see each other on friday morning for the join young session. Thank you. Bye.

F

F