Internet Engineering Task Force Interim Meetings, 6 Apr 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: NETCONF WG Interim Meeting, 2020-04-06

Description

NETCONF WG Interim Meeting, 2020-04-06

A

A

Name and great and good lady's name guys, so this is the net counts, working group meeting in lieu of the one of seven face-to-face meeting.

A

There's a well aware of north wales by participating in this meeting. You repeat the idea processes which are outlined in a bunch of these hippies at the bottom of the slide, make sure you are aware of them.

A

Administrative trivia, so for this meeting we will be using the ether pad as a virtual blue sheet. They use the ether pad link at the bottom of the slide. There.

A

You login can aim down at the bottom of.

A

Station and also for the folks that one contribute to the minutes, you can put your name against your color and the top right corner that would be M, so this meeting is being recorded. Please mute your lines unless you're speaking and you can indicate that's Q minus Q to be added or removed from the queue when asking a question help us by stating your name, it's hard to keep track of we're speaking on Java in the next slide.

A

We are using double, but it's a month, although idea does log the jabber content, you.

B

A

Link on the slide, I also send it in the email yesterday it can be used for any side discussions you want to have, but if you want to ask any questions, please ask of the WebEx by indicating on the WebEx chat window, not the job or video. If, for some reason, you're not able to do that, and you can ask there, isn't a Java Script I can ask someone who can post questions for you.

A

Of the chargeable co-pilots, the reader, plus it's working process, there is a presentation today, I can turn it push. Capabilities is passed. The second working group last call the yang push notification. Messages is work in progress.

A

The author's agreed to the public till we have an HTTP trust.

A

My auto is breaking up. Okay, I'll break.

A

Cbs notification directors work in progress and there is the agenda for the meeting.

A

Oh version on the impact on this camp and that control.

A

Agenda any question.

C

You're also here just to tell you that Taurus is presenting the multiple stream originators one. So we're.

D

A

C

A

No other questions are common and I guess you are up.

E

Okay, great, thank you, everyone and please, let me know if my audio is coming through clearly.

E

Okay, everyone or get another opportunity to present this suite of drafts. This is the client-server suite of drafts Lucy. What are they say? Nine nine separate drafts that we've been working on for quite some time here next slide. Please.

E

There were a number of updates that were made since 1:06 and typically I don't go over. The updates, I just simply put change logs on the screen, my site, please, so that everyone can be aware of all that had occurred. There were two separate updates to all the drafts since last time.

E

Actually, if you could go to the previous latest quickly, as you can see, the first three drafts the crypto types, trust anchors in key store. These three are the ones that we're trying to bring to the school as soon as possible, in the sense that they have the core drafts we'd like to get them project preceded progress as quickly as possible. That said, I believe from a technical perspective of all the drafts are currently ready to be progressed, there's no more remaining issues except one, so we're gonna discuss this on this call.

E

Okay, can you in two slides, please slides for this presentations? Focus is really two remaining open issues that we've been discussing for last few sessions. The first is how to define a dictionary of algorithm identifier, and the second is how and we enable a server to declare which algorithms it supports, and these are the only two remaining open issues next slide. Please.

E

But really the question is: what strategy should we beat? Should we be taking to solve these issues? And if we last time we were or historically I should say, we've been working on a solution whereby the cryptid types trap to find a dictionary of all possible algorithm identifier, and it was doing that and I think initially using yang identifiers and then later yang, enumerations and then most recently in the current update, still using enumerations.

E

But now those enumerations are being programmatically generated using a script of registries that I Anna is maintaining and but nonetheless that's the approach that's currently been taken. So that's number one option: number one is sort of continued with that approach, which is to define as much as possible ie those algorithm identifiers in crypto types Draft at that layer.

E

Number two would be to actually break things out and try to move protocol specific information into, for instance, the TLS clients or her draft and the SSH client server craft, and then three would be to actually give up. You know effectively fold on trying to support the ability for to ask a server to generate a key private key, so these three options are discussed in the next following slides next slide. Please.

E

Okay, so again, solution is that in the crypto types later there are two are pcs for generating symmetric, key and asymmetric key. You can send a request to the server to have it generate that key and actually that's defined, a crypto types module and then separately in the keystore module it augments. Those are pcs with an additional parameter specifying yet another key in the key store that can be used to encrypt the resulting generated. Key.

E

That's returned to the client, so I'd say Tripta type layer, but it's augmented by the key store later in that way, the new Who I am so there are now in the current crypto types. Draft I knew I am a cryptographic, primitives registry, so something that I Anna would maintain and actually more specifically, actually there's a four point number four.

E

So we took it to that and then so so there's this registry and so, for instance, there's a registry for all the known asymmetric algorithms and those would be, for instance, RSA or the various elliptical curve that can be used to generate a symmetric keys and then likewise various symmetric algorithms like AES or triple des for symmetric keys and then actually also a third registry for hashing algorithms, sha-1 sha-256 things like that, and so those there's those three registries in this cryptographic primitives registry and then the idea is that there would be a script which actually hasn't been.

E

You know it's not it's right. Now, it's it's in the appendix as a TBD, but there'd be a script written that would generate the yang module from those registries and then, lastly, the idea is that there'd be a new column added to some other registries maintained by Hanna, so, for instance, in the TLS there's some there's a registry 40 lat, various TLS pre, another registry for various states parameters.

E

The idea is to augment those registries with an additional column to effectively map which cryptographic primitive is being used to buy those various TLS for a state, specific registry entries, and only this.

E

The idea is that then, the domain experts, the cryptographic folks as and when they're updating the various TLS and sssh algorithms they're being mindful to ensure that if there's any new cryptographic, primitive SP, that's being proposed that that the man that cryptographic primitive registry would be updated in in you know, concert with their update and then I you know, would generate the new yang module and it all happened automatically in the background and then and then separately. So that's that's in the crypt graphics layer and it separately and the SS agent T less layers.

E

There was going to be also a TBD config false list or which specific algorithms a server implements and I. Here you know so there's the universe of the dictionary of all possible algorithms. That's in the cryptographic layer- and you know what is a particular server supports, and so this config falselist would allow an ability for the clients to discover the subset, supported by by a server benefits the pros and cons. So the benefits of this approach their first office.

E

It's consistent with a view that algorithm algorithm primitives are protocol independent, for instance, in RSA key doesn't itself, you know say whether or not used for TLS or SSH.

E

You know you have an RSA key can be used for any protocol. It uses are safe, so it in the sense having the primitives dictionary of all possible primitives factored outside, of how they're being used makes sense.

E

That's consistent, the constants that is inconsistent with existing protocol, oriented registries or in a right, setlist and I me and I say this really in the context of within the IETF and I can't really speak for other standards buddies, but it definitely seems to be the case that most cryptography focuses on the protocols and you know, within the context of their focus on the protocols, they're thinking about the algorithms that those protocols are using and they give those out where those names and so the you know they kind of build up a solution.

E

That's around that and while it does, you know, use there are same underlying primitives. They don't think that's it in that way. So that's what I mean by that and then and then. Secondly, it's it's a little bit on it maybe unclear if the same algorithm, primitive name means the same thing to all protocols, and this is actually something I learned from in discussion with the cells.

E

Who's been actually a member of sector, but also he's been on the list and and someone that I bounce a lot of ideas off of- and you know, I've had a some discussion on it. The last few weeks actually rich. If you are you on the in the call right now, yeah you're me too go ahead. I can hear you, you want to say anything about this.

F

So I don't know if you want to talk now. At the end, I mean.

E

Yeah now, just in terms of how on point is this slide,.

F

Sure so it's basically good this one thing: I wanted to two things: one, the cryptographic primitives is more. You know it's big. It's every curve is its own primitive because you can't say I want to generate an ECC curve. You can say I want to generate a brain pool he whatever or in this p5 21 curve.

F

So it's a big set. You know 100 plus um and then the comment about Amaro sake. He doesn't care how it's being used. The protocol doesn't care, but there out-of-band considerations that relate to security. You want to make sure that the same key isn't being used in multiple security protocols- that's probably out of scope of here outside. If we just sort of copy some of the security considerations, um but there are, when I saw the RSA key occur to me. There are keys that have parameters associated with them. In particular.

F

Was this thing called RSA, PSS and I forget what the PSS stands for, but it says you know you can only use it with this kind of hashing algorithm or only use it, for this kind of signal only generate a signature for it. So there are keys that have parameters that affect how they're used in the protocol, although it's that statement about an RSA key doesn't care if it's used, that's accurate, it's just a little subtlety on there, I'm, not sure if that's probably not worth going into, but there are other cases here which isn't.

F

Why set gear and stuff being concerned about trying to maintain a universal registry. But we can discuss that later on I know, you have other slides thanks, rich next.

E

E

Ok, so the second option is to, instead of trying to define a universal dictionary of all possible algorithm identifiers that are shared by all protocols. We instead factor it out or those dictionaries out into the various protocol layers, so in particular in the crypto types layer, we would replace the algorithm type from what its current currently being an enumeration really back to it being an identifier and having identity ref to base identities. Those base identities defined in the crypto types layer, but then the derived identities would would only exist in the protocol specific layers.

E

So then, in the second bullet point there is a and the S section. Tls layers extend those base. Identities with critical specific identities, also define our pcs for generating the symmetric and asymmetric keys and really those our pcs would be mimicking. Almost you know, parameter for by parameter the command-line utilities that open SSL and OpenSSH. You know provide so I open, SSL, gem key open SSL, the EC params. You know for creating RSA and and understood sage-infused gen RSA is what I'm gonna say for generating our say and elliptical keys and an openness to say.

E

Ssh is called SSH underscore keygen. Those command-line parameters could basically map them almost one-to-one into our pcs. That would be living inside the S section, two layers and then lastly, defined again that DVD config false list of the particular algorithm, supported by the by the server same as from the previous slide. So pros and cons of this approach pros is that then, now we have protocol specific algorithm names. If you recall on the previous slide, there was a concern or that potentially, while the name is the same it actually it it's.

E

It's used in a different way and there's known possible implications to that, and so this actually would resolve that particular problem, because now we're using the names of the actual protocol uses, and so that part at least we would have to worry about doing a technical analysis as to whether or not there's a concern or not it. Now we know what would at least know there's no concern. It's been isolated. The cons would be that it's either. This approach is kind of inconsistent with the view that algorithm primitives are protocol independent.

E

So it's just the reverse of Pro. From the previous slides, so with the the strength of the previous approach is sort of the weakness of this approach. um I'm gonna stop there on this slide and ask again: if rich has any thing, he would like to add on.

F

Media yeah, no I, think the con is huge, but this is all correct. Okay, good next.

E

Slide please, okay, so three is to really fold on this entire approach of trying to enable servers to generate keys so kind of more specifically, this solution would be in the crypto types layer to eliminate the to generate.

E

He are pcs, so the generate asymmetric, key and generated symmetric ERP just get rid of those two are pcs entirely and likewise eliminate the algorithm I field. That's in the various key groupings, so just below that you can see, there's a kind of an eye chart, but there's a grouping for a symmetric, key and another grouping for an asymmetric key and both of those groupings.

E

The very first node is called algorithm and you can see it's an identity rough to one of these actually I'm, sorry, son, I, didn't say it's an enumeration to one of these dictionaries that we're talking about so we just remove it entirely and surprisingly, I think this would work, because if you recall we moved a what was it last time to we added this key format. Node, it's a separate code called key for maps.

E

We have for symmetric key is just called key for map and then for an asymmetric, key pair there's, actually a separately there's a public key format and a private key format and those key formats are, you know saying whether or not you know if this, for instance, until us in a SN dot one a subject, public key infrastructure, or is it a an RSA private key because of actually the names, if you could see I'm doing air quotes?

E

These are the names of those structures and and inside of those structures you know sometimes actually most the times they actually have themselves and I algorithm identifiers. So it actually everything continues to work. We can still I mean it could still be the case that clients can generate the keys externally, using their local, SSH or SSL libraries and and in in doing so they're constructing. You know these key formats, which would be consistent with this format field and then passed.

E

Those resulting values to server and, and so the ability to you know, store keys in for essence. Key store would continue to be there. We could still support the ability of having hidden keys, as you can see in on screen that you know notion of a server being shipped for manufacturing, with with a pre generated key, for instance, in his TPM would still be supported and also it still be supported to have encrypted keys, I mean so in the key store you could still store encrypted keys, but now the encryption of those keys.

E

It's not clear theme before we had it that the encryption would occur by asking the server to generate key and then passing in another key to do they encrypted with it would have to figure out how to support that. But I think we could still even continue to support the notion of having encrypted keys stored in key store. Okay, so pros and cons of this approach, the pros is fastest closure.

E

We can, you know, take this to really last call as soon as the editorial edits serve made to described above the cons is that there we know ability we, we would have no ability to ask a server to generate ease, at least not until someone else picks up this work.

E

Right so I mean on that point, I'd like to note that it is security, best practice, long standing, security, best practice to have for a private key to never leave the cryptographic boundary that is intended to be used, at least not in its uncrowded state and by shouting I mean like if, when we talked about encrypting the key, that's shouting the key. So then it's safe to to export the key in its sharted form outside the server. But you know if you're having a client generate the key and then passing it in that's.

E

You know the key. The private key actually existed outside the server initially, so it goes against that best practice and that's that's the one thing losing. We may also be losing some usability right. There's some convenience to just being able to ask the server to generate the key as opposed to client even mean meetings. Do it themselves and then, as far as someone else, picking up the work I mean I, you know I think potentially it could be in theory, you could be months, but it's more like years until we because work.

E

The other con I've mentioned here, is that while it's true that the key format can really save the day, you know and allow the whole entire solution to continue to function. The wetware, the one place where it kind of fails, is that there is one key format for symmetric keys, called an octet string. Really it's just an array of bytes. So there's absolutely knows a sn1 structure to it. There's no there's no I! It's not clear! You know. If it's you know those bytes are being used for an aes or a Triple DES.

E

It's not you. You know, there's all sorts of things there unknown about it. So if we wanted to use symmetric keys, we may have to get rid of the octet bytes symmetric, key format and instead only use the one symmetric key format.

E

So one symmetric key is really it's a it's part of the CMS structure, a Content message. Syntax, that's a structure, a structure so it, but then it so it can be used to support the symmetric key it previously. On lists, we discussed how the one symmetric key- and it's corollary, the one asymmetric key- are not really as friendly as the you know, native key formats they're, not typically with the tools generate like, for instance, if you're using open SSL isn't not going to generate one symmetric key for you or even on one asymmetric key.

E

You would have to code it using. You know your own libraries I. Am that one type library, so it's not very user friendly or even developer friendly, but it does so. It would allow us to continue to support symmetric keys, the good news actually while it's while mentioning here, is complex. It's also something that would likely not be used very often, symmetric keys are not often used.

E

In fact, the only reason we introduced metric keys- it's not it's not for TLS or SSH right, because they always you know they're, always using asymmetric keys, and so the only reason for us having introduced them as symmetric keys in the first place, was to enable the encrypting of private keys so that we could do the shrouding and enable you know RMA type activities.

E

Those are the pros and cons of option. 3 next slide, please so.

G

Again before we move a couple of questions, do you want to take them now are at.

E

The end well, the next slide is in fact to take questions, so just go to the next slide, and so now we're really just back to the previous slide and the three options are on screen and opening up for people on with our thoughts, and you see Robert on me, I think I renewed you, okay, go ahead. Okay,.

D

Hear me on the previous slide. Actually I just wanted some clarification on the yield, so you're saying that if you remove it, you don't really lose very much. Am I right and understanding, therefore, that the Alcala field is used both to be able to when you're generating a key tell it? What parameters you want to use well gyum, also in this case of the octet string key format. Is that the two cases where the out type field is actually required? It's not remind in other cases, yes,.

E

So or I'd say in the ninety-five percent use case of the algorithm, identifier was so that was for those two RPC is to generate a symmetric key and generate a sweater key so that you could tell the server what kind of key you're trying to trying to generate it. So that was the primary use case for the algorithm I defer. That said, it was always a good idea to sort of just have it.

E

You know presented, um you know, usability to make it visible and then, as he say, for the octet string or for the symmetric key okay.

D

Another question on your previous slide to the option: two I think that's they're, just moving our our PCs to protocols does that to the SSH, so the offices become under this sh t LS lace. Does that mean you take the our pcs out of the crypto types yeah, because I? It feels very strange for me for a types module to define our pcs.

D

E

Well, indeed, it would be to remove them out of the crypto types in module and and move them in so so actually, I'd probably continue. There probably continue to be called the same thing in you know called generate asymmetric key, but it'd be in the TLS namespace versus the SSH namespace. So that's you know so so you don't. We wouldn't have to necessarily call the RPC itself generate and TLS key, for instance, but yes, we did. The idea is to move them over into those safes and tailor specific traps. Okay,.

D

So my third I think I'm here is, is this seems to me like a nice approach. I will either defer the whole issue in solving later or I. Prefer this option I thinked option one, but then rich made a comment that this corn is a pretty big con. Is it possible for you or rich to expand on why that is the case? Cuz that's made me wonder.

D

E

This quickly, I'll say something and then I'll let rich respond as well, but if so the the list of asymmetric key primitives, primarily you know RSA is you know the the 800-pound gorilla in the corner, but then there's curves, elliptical curves and if you ever did in open SSL. You know you asked it to list curves you'll get a dump of around 88, curves I.

E

Think and it probably it depends on how options, but there's a there's, a number of curves and- and so you know in total, there's you know around about a hundred and that we're doing with now know in my mind that actually the number of primitives that there are doesn't really matter, because it's all you know, programmatically you know processed and who cares, puts 100 or even 500. I, don't know if that really matter, I'm a usability perspective.

E

So with that else, I'll see. If rich, has anything that yeah.

F

Yeah sure, just to clarify my point, I mean when I said it was a big big. Con has other meanings, but when I said the kangaskhan thing was a big one, I mean it's actually from the security of your point and I said. A key identifier like an RSA keys. Kent said earlier is an RSA key is an RSA key, but how it's being used that it's not is not. Protocol independent I mean what TLS does with a key, an RSA key to sign a hash and prove its identity.

F

That way is different from what how SSH uses keys to generate to verify identity, and so it's not. You know using the using an RSA key in TLS and an RSA key in s mine. We are signing or verifying arbitrary content as opposed to content. It comes from a mix of two parties. That's you know. The security folks would rather see those things is independent and put in the protocol layer as opposed to the you know. Oh okay, crypto layer, RSA is RSA, is RSA only in an academic sense. I.

E

Hope that helps well rich if I can just follow up on that point. So there's this notion, for instance, with x.509 certificates for key usage and the key can only be used for signing or for encryption, and you know digital signature, etc and I think it was Russ Housley, who suggested also, we could add to our key types in the crypto type layer, this notion of attributes so kind of mimicking.

E

That idea, where you know the key was known to only be used for signing versus you know, signature versus encryption with that result, the statement that you just made now or even something additional integration.

F

That would probably result in it. I know that I mean yes, that's kind of what you're getting by putting them in the protocol layer. Is you say? Oh, this is an RSA key used in an SSH and therefore that implies all of what the SSH uses are. Okay, as opposed to this, is an RSA PSS, key or RSA key in TLS, which implies a particular yeah. It's either part of the static key exchange in 1.2 or it's part of the signature as part of the I believe the server identification in 1 3, so um yeah.

F

You could do that, but I would think the security folks would be happier with having it and I hate I'm, not representing the entire security community.

F

Right there, like economics, economics, 9 people will have 10 opinions, um but yeah it's the same thing, but you have to think about well now, yes, thing about what the registries are, and you have to get keep subject matter, expertise from, say what SSH does and what TLS does would say with an RSA key depends on what version it is, whether it's it was a big change in TLS 1/3, where we don't. We use only diffie-hellman, whereas before you could have used RSA to do the key change to so so.

G

I actually need to cut the conversation here, be running all the time, Jason very quickly, because I really want to move to the next presentation.

H

Okay, I've got two parts quickly, one just a comment: I'm also and I'm, not sure how to evaluate option: 3 verses, 1, & 2, but out of 1 & 2 I'm I'm in favor of 2 as well. I think you know, maybe, theoretically, a global registration of all methods is kind of nice, but I think the way most buts say meet users or maybe have somewhat versed in security issues, but not security.

H

Experts I think they're gonna, rather be focusing on their protocol, what's relevant to the protocol and kind of adding this extra annotation to say all this is how it's used in SSH. Those using TLS again may be theoretically nice, but it just seems like 2 is a is nicer in general, the users of the of the module. That's comment: my other question is in option 1. Why are we going with an enum there instead of an extendable identity rock? Is it it's because we actually specifically want to prevent this kind of extension of it.

E

The discussion of that goes back to a few months back, but to be honest with you, I don't think it was the meeting tonight at the moment.

E

H

E

Has he's in here to make a comment.

D

Yes, my question really is: is how do we get this to conclusion, because this these draft have been here for a long time and I know the other model to waiting on these. So it's question to Kent as an author and as a and it mashes chairs. What's the best way to sort of resolve this, it's working lost papers. It's.

E

Gonna have that so, as as chair I think we should. We can take this to the list, but you know we need we're in real way to a hum. You know on our it sounds like one is out it's between two and three. We could do a hum and try to get closure on it that way, so I I think we should do the virtual equivalent of it after this call.

G

Yes, that too, but what about and does this affect all the drafts or details there are some drives. You can start work proceeding to last call this.

E

Is encrypted types and and also somewhat in key store, so the you know those three drafts that we're trying to bring to work your glass call immediately or the you know two of them are affected by this decision. We need to kind of do this: first, okay, okay, so that's in the end of the presentation and back into the next presentation is by Mahesh the working group draft or HTTP note of notifications.

G

Okay, are you guys able to see the slides? Yes, okay, all right so I'll be talking about the HTTP based notification, so the update since a one eyes eye in the draft. We added the ability to learn receiver capabilities they put in an example.

G

Example essentially demonstrates that if the capability, where we can either do JSON or XML, it was JSON being the mandatory encoding and that this is how we could use and learn the receiver capabilities, and we have left the option open for, of course, for future and codings if you want to define and to learn the receiver capabilities.

G

So that's as far as the protocol change is concerned. In the model itself, there were two changes that were requested by martin. One was to rename the receiver to http receiver, to give it a more a better name and then the second recommendation he had was to change the top-level container to actually augment subscribe notifications. So both these changes were made in Oh and the examples were also updated, related to band.

G

So talking about abilities. Subscribe notification has an encoding leave that is defined, but it is. If the leaf is not said, then the publisher is must discover the receiver capabilities. That is, the encoding set need to be supported in the example that we demonstrate in the draft to discover we, the publisher sensor, get requests and sets accept types that they would accept in this again.

G

Json is the default encoding in the particular example in the draft, the publisher, so if they receive it, does support XML and therefore responds in an XML format with it the list of supported capabilities.

G

The example we give here is to say that it's boots, binary encoding now I believe subscribe. Notification has an encoding defined for C bore, so if the binary encoding is C bore, actually that's what would be indicated in the response.

G

Finally, the issues being tracked in the on github is the fact that we need to have the binary encoding capability registered with Ayana and actually an issue that I need to talk about, which is, should encoding be mandated. I'll talk about that in the slide, so the question too King groupers, is it I need that should make mandatory to implement encoding subscribe notification draft is actually silent on it. It defers, I, believe and Rick will correct me if I'm wrong.

G

It differs to the transport document to define what's what's mandatory if I need to implement it's not dictate quest to the working group s. Should this document pick one encoding that both the publisher and the receiver should implement or the other option is we require that the publisher support both XML and JSON, or the receiver support both XML and JSON, such that at least there is one communication channel, that's open, so any opinions on this working group perspective I.

G

Got a confirm, Eric saying that it is corrected. The transfer document is the one to select mandatory encoding and we have question from Robert. So.

D

So don't know the answer this, do you have you expressed a preference for this? Did you have an opinion and.

G

Would you say that have a burger.

E

No I do not have a preference.

D

And if you're going to mandate, one which would you mandates because I'm a bit worried about on dating, say XML written, he is with with the nuit. You have protocols and things that XML is not to invoke well.

E

And and similarly with RFC 80/40- and we did not mandate JSON versus XML, so kind of follows that mandating.

D

Anything is rational, be my view too.

E

So if you're not mandating ivory, then the yes, sir, then we founded our class so we'll so for interoperability reasons we either at the publisher must support both or the receiver must support.

D

Would say the publisher chooses what it wants to publish in and receiver has the choice to support one or both depending on how much it's Rob they want.

D

E

D

You let the market decide on what is what I'm saying.

G

All right so you're saying the publishers, xml and json, and the receiver can choose which one they want to support. I'm.

D

Saying that the publisher could choose to support one or both of them they decide and and the receivers obviously, depending on what what publishers they need to interoperate with will either support one or both of them.

E

Guaranteed an interoperability, but the market will figure it out.

D

That's my impression.

E

Their comments.

A

E

Have a comment from chin, America, sorry Erica just.

I

One note is that you have a subscription started message at the beginning of this with different encodings. You can always not okay, the subscription started, and that is a way to say whether you're, supporting and encoding or not as the you're supposed to be able to okay. The initial subscription started message prior to sending and that's to stop you from spamming things. If your, if your receiver doesn't accept it, it shouldn't be sent if it doesn't accept the encoding.

E

So Kent as a contributor or actually said co-author, this is not a dynamic description. This is a configured description and we don't really have the protocol there. There is no description started message to okay, the.

I

Subscription started on the configured subscription initial notification.

E

Well, it's not in the document it's not a currently. Maybe we should circle around again.

I

Yeah I think it might be worth looking at the the state change notifications that start the document to ensure that you, okay, the subscription started.

E

Next comment is from Kim.

J

J

So I'm just thinking actually, why don't just make a publisher to advertise that ability vocoding it is supported and then receiver can selector, but it may be only apply to the dynamic of subscription you want, because in your cases, do you focus on a configured separation.

J

Is publisher can advertise the the encoding supporter and the receiver can select one of the encoding did, but maybe it is only a part I to the dynamic on subscription it is. Is.

G

Essentially same as what properties saying right right: well, not really, but you're saying it's support, the publisher can decide between XML and JSON, and then it is. The receiver then has to pick one that encoring, like yeah.

J

We have a relevant job that you talk about. Telemetry data export capability.

K

J

It yeah right so.

E

This is actually like chance or the draft so that that selection is actually reverse flow of what this draft is doing. It's okay, I'll, put a comment for that next in next, and Q is blush, actually dropped order. The gear- oh thank you next in key word, is Thomas about.

L

Encoding, we have sweet and Seaborg. One of these three encodings need needs to be supported.

L

Say three XML JSON and C poor.

G

So suggesting, although that was something that we were proposing as capability in capability advertisement, this is initially it needs for the set up part. So seaboard is very much thick encoding. We want to support as part of firm, receive a capability that advertisement.

G

Okay, I think next we have Robert.

D

So just one question to clarify is when you said these are configured subscriptions at the point you configure it. Does the client have an opportunity to specify what encoding that they want the serve the publisher to use, or is it entirely up to the publishers, publishers, discretion.

G

Do you want to try to.

G

E

Rob, can you repeat, is.

D

This is this Phil configured subscriptions.

E

Only for configurative descriptions so.

D

Does the configuration allow the the client to choose what encoding is or they just go? Yes,.

E

Actually, this is a very subtle point and I'm glad you brought it up because I don't think it's been highlighted. It is in the draft. It's highlighted a draft, but in the SUBSCRIBE notifications now in RFC 86 39, they think there is a node called encoding, and that note is an identity. Has values such as encode underscore xml encoding, Square JSON? Those are the only two defined identities there and what it, but it's an optional, a specified note.

E

It doesn't have to be specified and according to that RFC it says that if it's not specified, then the encoding must be determined through somehow to ban means so really where what we're talking about here is in that then clause like like what do we do in case it has not been specified perhaps are out here is to just say: you know you need to specify it.

D

Yes, I, don't know in that I yeah I, don't know, I still think. Actually it's just let him serve deciding codings.

I

One possible answer here: yeah I'm, glad.

G

I

One of the things that you can do in the exchange of capabilities is to reply back to a subscription, starting with some information. That would let you choose that this is all back to the original discussion. We had I think an idea for two ago that the receiver or the farside should be able to reply with the capabilities in the response message. So one way to handle this is to allow something to come back as part of the transport message that allows information to be signaled back.

I

G

Right yeah so we're running way over time. So maybe we should take the rest of the questions to the litter mailing list. Unless you have a quick comment, rabbit.

G

To the list Thanks, okay, thanks all right. Next, we have Jason.

G

H

Okay, you guys hear me: okay, mm-hmm, okay, so this is one of a series of drafts that are mostly being discussed in net and net mod I guess I should have changes to Netcom. Sorry, it's the same set of slides, but it's all related to the yang module versioning work, where we're looking at how to describe when yang modules end up with non backwards, compatible changes during the development and how to describe that and and when yang module development, sometimes branches etc.

H

So this this draft is part of that suite and this one has some implications for net comp and rest comps protocol there. So we thought it'd be worthwhile. Talking about this one in in this in this meeting, go to the next slide. I would like to make sure we just contain the scope of discussion that happens in net con flow, so the majority of the issues around versioning is being discussed in net mod, we'll try to keep just the issues related to this draft in in that comp.

H

For now, and one of the big issues we have is exactly: where will this draft live? We're thinking? Some part of it mate should probably live in net comp, but that's one of the issues we'll talk about in a few minutes next slide.

H

So a little overview of what this is about. This is an optional functionality. It's it's by far, you know a more heavy-duty implementation work on a server side, fairly minimal on the client. This allows a server to support and advertise multiple schema versions.

H

As an example, here, a server could support their their modules version 6:01, and they may also have the option of supporting an older, a previous version of the modules 5 to 3, and then there are mechanisms for the clients to select which schema said it wants to use.

H

So if you have a client that implemented did you know you did your OSS integration a few years ago, you upgrade your router, you know, maybe that router will allow you to just keep on using the old schema sets, and you don't have to do some work to take into account some of the non backwards compatible changes that might have happened in the latest version.

H

There's a default schema set if a client doesn't request anything so clients today can perfectly interoperate with this proposal. They would just use the default schema set, which can be configured on the server side as well. If desired and schema sets are comprised of yang packages, and so the concept of packages is being discussed in net mod, it's basically a collection of module versions and next slide.

H

Keep on going so these are changes since the last night after the draft. This is one of the drafts in the set that has actually churned the most in in this set of versioning drafts.

H

One important change is we've simplified by removing the RPC that was designed before we had been kicking around the idea of using a an RPC to select the schema set. We simplified things to basically have it so that net comp clients, the only possibilities select a schema, is at connection time during initial capability exchange. So if a net client needs to use a whole new schema set, they want to back out and go to a different version or upgrade and use two new version.

H

They would disconnect and reconnect and negotiate the new schema set a rest comp. Clients use the schema set name in the request URI for each request.

H

The other use case, that's interesting, that we we kind of highlight now is this may be used not just for selecting different versions of a schema, but it might be interesting and some implementations to allow a client to say: hey, I, want to use open, config modules or hey I only want to use the vendor proprietary models, so you could allow to decide between different organizations and modules. Next slide, we clarified what we're calling the basic unit of selection, the Netcom client connects he's going to select something and that something is called a schema set.

H

Basically, a schema set is illustrated on the right. It's a collection of which packages are supported in each data store, so Kahn describes every data store and what packages are in each of those data stores groups it all together and something called a versions schema set. So in this example you'll see we have our vendor config modules version 3, 1, 2 and 2 5 2 you'll, see those supported and running you'll, see those supported in candidate you'll, see it imported in operational.

H

But then there may be a couple of extra kind of state related packages in your operational and on the bottom left. You may see some something totally different in a dynamic proprietary data store, ok and continue on the next slide.

H

Ok, I'm going to move past this in the interest of time. Let's skip over the slide. Okay, here's something another change, that's kind of relevant protocol wise! We had configuration for restaurant protocol instances, we've changed the proposal to just put the name of the schema set right in the request URI. So you can see an example at the bottom here where you can see this. The client is selecting vendor schema at XYZ.

H

The next slide.

H

Okay, I'll just focus on point six, a little bit more key if, if one net conf client, for example, is selects a particular schema set, there's absolutely no effect on the sessions that are already up between other clients on the server, so the other sessions would continue as they were, with no change and what they'd see in their schemas was very it's bits per session and fully independent for Netcom session for s conference a bit different. It's every request specifies in schema, said next.

H

Okay: okay, go to the first discussion topic next, one: okay, so.

H

One one issue: that's the kind of protocol related here is: we do have the capability in the draft that the the server can support. Maybe the server might support three different schema sets, but the operator can configure ibly kind of remove or limit the set that that operator wants to advertise to clients, so the operator may, for whatever reason decide. You know they only want to advertise two of those three schema sets for use because they may take resources in the server when a schema set is removed through configuration.

H

One issue: is you know? What do we do with sessions net comp sessions that we're using that schema, set? There's a few possible options? One is disconnect any net comp sessions that were using that schema set. Second, allow the sessions to kind of continue, but any new sessions trying to use that schema set would be blocked, and then the third is actually reject that configuration don't allow the operator to remove the schema set of its and use our preference from the from the authors. The X design team is option one to keep things simple.

H

Some three would mean blocking config because of state and stops administrator from being able to remove a schema set. So that's not so much preferred. No there's any other thoughts on this one.

H

If not I'll move to the next slide.

E

Jason- and this is Kent as chair- the the biggest question here is whether or not there's internet cough group and there's some discussion as to you know. If used to stay, you know all in that mod or some of it should move to Netcom. Okay,.

H

So that's slightly dirtied. We can skip there.

H

One more slide: 13.

H

Yeah, so this is, this is the biggest issue for this group. I mean there are definitely some impacts to net comps in that there's a there's, an optional kind of advertisement from the from the server and selection from the client and same.

K

With restaurants,.

H

There's the abilities to put the schema set in the URI, so one of the big questions is: how do we tackle this work between net mod and net comps and whether we take some part of this draft or this entire draft and keep it in this working group? So there are, there are conceptually kind of three parts to the draft. It could could be broken down all the way into three drafts. One is just the very specific net comp stuff.

H

Another is the very specific restaurant stuff and then there's some things in the draft that kind of are not necessarily net kampf or rest comps specific, but are just more about this version selection process. So we could either keep all that together and one draft here or split it into three.

H

So I, don't know what the feelings are, maybe more from from the chairs or others on that. So.

E

This is Kent speaking as chair and I actually have both working groups, knowing the charters of both working groups. Net mod is a charter to work on modules that may you know for any working group, potentially, if there's not a better working group for it to be in a knockoff, of course, can work on yang modules that are related to transports and whatnot. But Metcalf is uniquely in chapter to maintain the protocols, so you know not just modules but not just defining lentils, but actually the protocols themselves.

E

If you want to make changes to protocols so, and only not can do that so to the extent that there's actually protocol changes, I think they must be done in Netcom and not in that mud so similar to the nmda work. Where that mod, you know, did the overall overarching solution, but then parts of it were factored into Netcom I think we have to do that again for this work and then as to within the net conf.

E

You know if it's one draft or to drive something like a draft for an account for such a four restaurant. That's something the working group can decide personally or for protocol specific drafts. Let me discuss later.

H

Yes, the only speak.

E

E

G

My second Ken's opinion I think as soon as the protocol work is concerned. That has to be done in that conf and I would say that, yes, in in a way the authors and the working group can decide whether it should be one or two or three documents: okay,.

H

The I'm also the yeah I'm all for the idea of keeping this work in net comp I think it's. My slight preference is to keep it as one document. Only that the I mean conceptually, we could split it in two. It would have to kind of be three but I, think they're gonna be so interrelated. It may get a bit confusing, especially if we move the third one to a different working group.

H

So I mean maybe maybe one part is to progress it as a single document initially and see how that feels as it progresses.

G

All right so final comment.

D

G

D

I agree the chair, say: I think doing this in in conf is the right choice. This draft I agree with you also the taking its first determined telling with split into three, so keep it's one I like really Jason that to start off with, let's keep it one and then I can't we can split.

G

Okay, we'd really have to get going unless Jason, you have anything else. You want to talk about. No.

H

That was the key issue we can move on Craig's next.

E

Slide is or presentations from chin and self expects. Planation data note AgrAbility.

J

J

Yeah I, my name is Shenmue I'm here to discuss the telemetry data tracking capability chapter, the title we gave his self explanation: data node attack capability. This is not a new chapter we already present in next next slides presented last Singapore meeting. Actually one of issue is how it will relate to the notification capability chart and the changes were made in the sense of the you know latest version.

J

Actually, we try to align with the system capability model that define in the notification message complete draft, so we introduced several additional ten meter data tagging parameter in the region stretching also in the models. Next.

J

Why this job? Actually, as we know, ITM already.

A

J

Amount of device model on an enemy model such as BPA, SAR and vendor, also propose a different vendor specific model. All these model can be used as a data source. It will capture all these datum all of these ten immediate data from this device model. Actually it will cause massive data connection and process in the net.

J

Aside from the connector point of view, actually, when you can establish a connection with the device for the next step, what they need to recover work on the net telemetry data, they can fetch how connected and they usually rely on the pre-comp selection, filter or subscription policy, and this subcommittee policy or selection theory you either can be either specified by the LMS or is oh I instruct by the device.

J

So next, so what we like to do is we define the telemetry data tagging capability model uses the model to advertise, so what kind attachment telemetry data you can capture to to have a client to capture the characteristics data from the device model, so the bigger benefit that you can identify the performance mm and the related data for service assurance application. In addition, you can provide a network visibility to there's different category of data like an anyone know that traffic flow capacity, and so also you can provide it like a tendency prediction.

J

Storage code based telemetry anomaly. Detection has many benefits to do this next.

J

So so, what is the metadata tagging? How do we classify is telling budgetary data? Actually we define the four typical ten metadata tag: the first packet. We call the OPM tag, it's a provisional opportunity, I p-- property and a metric and Aziz can happy to capture the character data of the telemetry data and in addition, we provide the fine granularity OPM tagger with AMPRO. When you connect to the performance america value, this value can be the maximize the value or minimize the value. So we can use the operational type tag to indicate this.

J

In addition, we come provide a metric of proceeding or metric a scale for this OPM tag and the second category. We call the service tag actually for it option. We observer can be used for the multiple purpose, so we can use the soviet akka to describe each purpose and one of the example you may have Arabic in service. You have tea turn on service. You can use a service tackle to do that to representative in ER, so this and in a third category we call the data source type.

J

It can be used, identify different data source. Let it relate to the connectivity resource hardware and the last one. We call the parent root King. Actually, this is something really to the my resource, aggregation.

J

Suppose you connect the performance metric. Oh, this is a permanent metric can be connected from different source for the online card or sub interface. So you need you need to figure out which he performs America from which is also can be correlated. So we can use a parallel group in to do to do that next.

J

So how do we use these 10 telemetry data attacking capability? Actually we're? Here? We give the example you can actually to capture the performance, measured, mem and later using the OPM tagger, and you can advertise this kind of ability from the device to the client so Cline and know how how to how to how to specify the dynamic subscription senator subsequent subscription message to the client, and so in this way we can provide a multi-dimensional diamond in the temperature data Alice's next second case. We call the massage clinic telemetry data aggregation.

J

So in these cases you may connect a performance image where from multiple source will AMA from multiple line Carter. So you need. You can use parent group into indicator which performs a magical freedom for areas P average in they didn't say from which inline car need to be group together, and you can otherwise advertise these capillary from the device to the client, so client and know how to you know, specify the message next.

J

So it's a motorway proposed. We actually augment from system capability with a set of the telemetry data, attacking we actually classify them into four categories: for example, OPM tag, service tag and data source type tag and a parent group in tag, and so we can use so we Ottoman you use this model to to advertise the different tag from the server to the client.

J

Packing capability, we can, you know, use these two captures that characterize risk data to correlate the data object from different model provider cooter and then move it a bit here analysis. So we single it's very useful to to have this kind of worker. We would like walking over to consider to at all these kind of walk, Thanks.

J

E

You you any questions.

E

Okay, so to the question of adopting I mean this has been censored a few times and I think that again we probably want to get a sense from the room in terms of adoption. We could take it to the list for a virtual home I see that below Justin Q question.

K

This consists of two parts: one is the idea that some of the counters or measurements telemetry needs additional data to specify what kind of counter it is. The other is the exact set of counter or exact set of properties. I think that will be more difficult to agree if we adopt this example, I'm working also in 3gpp, which has a very different idea of what information we want to know about counters and measurements, and others might also have different ideas. Thanks.

G

E

Thank you, and we are over time and Ken I want to ask you not only as a presenter but also I, believe someone who is involved with pretty much all the drafts that are remaining on the agenda. If you would like to somehow coordinate how much time each or which ones represent and how much time each.

C

E

J

Yeah we we have a as a or a really relevant chapter, and so what are you propose to do? You want to? We.

E

Can continue with the existing agenda but we'll run out of time. We.

J

Try to make it a shot actually yeah also.

C

Here I can make mine short my hash, because you, oh it's Kent speaking sorry, due to your request to to about the way. Instead of the content. Actually I can present my slide one two and then eight and be done with it and then leave the time to do burn. Also, if you want me to make it in five minutes, I can do it in five minutes. No problem sounds.

E

Like we'll continue with the existing agenda order, so next is Peng, presenting Boggs description to Yangon event, notification.

M

This is the pump from China Mobile. Do you hear me clearly, thank you everyone. So it's about a bank subscription to young event, notification that size.

M

The subscript notification draft allows users to subscribe to the event streams that they are interested in and notification message. There is sorry, next, okay and bundles defined in draft notification as ages.

M

Allow bundle multiple, if occasions, into one single port message and the used subscription ID for identifying the target, and but there are also some quests for problems that the subscription protocol real operation doesn't provide specific criteria to classify subscriptions and therefore lacks the ability to explain clearly indicate which specific subscription associated with a notification should be bundled together and give two examples, and the two subscriptions can be bundled based transport and decoding parameters, and they may also be bundled based on the relationship with a set of young with models.

M

So we think these problems may not be solved and present an English draft. We devised a younger model and associated maximum which allows sub subscriber seems to bulk, subscribe to publishers, event stream based on group I didn't so this.

D

M

The publishers to reap out.

M

M

So they are to use case for Bank.

M

One is dynamic subscription and in the dynamic subscription the subscriber may have already had very.

K

M

About correlation between subscriptions like a quick aggregate aid, subscribe data from multiple.

M

With the prior.

D

M

Knowledge, it might send the bundle subscription request to indicate what specific notifications associated with subscription must be, bundled together for.

D

M

User, a want to package subscriptions one two: three together.

M

O users and established subscription, including the notes, powerful bus and device, responds with ID one two three, so the Christ, the.

C

M

Updated advice, so the device can give back reporting and in the reporting, distinct multiple occasions corresponding to the past vertical operation. For those notifications are bundled into one transportable message and using notification message: headers and bounce defined in notification messages, but the next price place.

M

Use keys to is configured subscriptions and so different from the first this case. Sometimes a subscriber may not have very knowledge about relation between subscribers, so an income biggert subscription. You can use the self explanation. Data nodes ability advertisement, which I introduced in the primer, obvious presentation, and it may describe the correlation between different notes in different youngest model from different publishers and can.

D

Be used to further establish correlation.

M

Between subscribers, so in and.

M

We can save from pictures that users want to packages of paper subscriptions that marked as both cater.

M

So there is a telemetry intact hugging capability advertisement also the bow and the banners are in the same Park, so the client infer that they may have a relation, so it can generate a bundle subscription of.

A

M

M

We can give a solution overview of it, the augment of Bank subscription and there are the additional subscription Praetorius a. Why is max from the literacy.

A

M

Isn't it before specific young notifications, general radiative, master, aggressive publisher, and this tribute enhance his features and we think it can provide additional subscription under classification criteria? Another is compression. Algorithm is.

D

M

Some subscriptions has the same compression algorithm. We can come together.

M

The subscription ID in the case, which must should be bundled together and the it can also indicates which must not be under together by Musco, mask the subscription ID. So.

M

Twistings the bank subscription you can improve data collection, efficiency and XPath data collection performance and we hope to work together ways message bonding if anyone interested in so thanks and in comments or suggestions.

E

Kent as chair I'm, looking at the agenda, I know that the remaining presentations are all related to descriptions and telemetry data, which is great that there's so much interest from the working group participants in this area and this and then in question and that's the observation. The question is: to what extent has this or and I guess for all the remaining presentations? Are there implementations?

E

Is there any? Are these just ideas, or do you reserve some code that and you work and you're working on the code.

M

Means at the ending work.

E

Right so um have you prototype this idea yet.

J

Can't yeah Kennedy's teaching, actually we work on that actually for some of the idea where proposed has you know some prototype.

G

J

Adaptive subscription and actually we we actually plan to to prototype I.

E

Think that would be good I mean for if we get like an overview, not just this presentation, but all of them and also Eric I think some work. That's going on with your.

C

Buddy about this tachi soft, the dressed up for something, that's my turf, no problem. All.

E

Right with that, the next presentation is also from ping. Let me trade data export capability.

M

Subscription with it.

G

Thank you is fading out.

M

And go to the next ice. Is the vacation capability modified in occupation capabilities allows a client discover, settle capability is supposed by the server like basic system capability and young push rated abilities, and this capability meant that the justice behavior to take advantage of the features exposed to the by the device? The problem is that reconfiguration, possum transport, specific parameters like transport protocol, including format encryption by, is a right, is still in a battle and which may cause and expected video and additional message exchange between right answer.

M

That's next slide, please, so this draft is to use telemetry data. It's both ability to allow the client to the SUBSCRIBE, install those updated with specific transport protocol, including format and so on. We think about providing those abilities like.

M

Configure the subscriptions with adaptive telemetry policy row and they provided notification message bounding if the device in the boss message funding, support per subscription of her datastore and also provide event based elementary. If the desk dog, an old, can threshold value next ice place and also gave three use case for eight unbiased ways, Bank subscriptions and adaptive subscriptions, which will be present in next.

M

The first is negotiation of subscription policies and transport protocol you can see. First, is the telemetry data export capability notification from device or server to the client?

M

Then the client can specify transport protocol and encoding format into a dynamic subscription and venerate the young notification for so updates with specific transport and encoding, and if it's done to have mobility and the Krait need to understand why the previous previous app is a request for specific transport figures relies on error, response to failed policy and the next I place, and the second use case is about message underneath support.

M

D

Can see in first.

M

Capability and first is data portability and if occasion about honoring, some servers on those may support.

M

While others are not so. The client can generate subscriptions dependency policy in the bank subscription to test on update based on the dependency ferry, and it can.

M

Is Donald updates for a set of specific modifications and if it don't have the notification, the client may not understand the boundaries policy and then drop all of the message from server.

C

M

Also have whose case it's about the policies on.

M

Some case we may have to the Justice, they are sure and change the policy.

M

Though, because first, the notification about shots both for specific, specific node, then.

D

M

Client can design even faced the elementary policy, for example, one who is hikers and for if we all figure young notification for a set of specific data than those next right, we all gave configured subscription ways event based elementary policy policy row. Finally, then the device can say be it without a notification cried me feel to configure it based.

M

Policy row is bounced me not able to tell or even based dementia policy configure fused if no new error codes is based fade for threshold, not support.

M

Please turn to page seven.

G

So it's like: did you see seven right.

M

Not his daddy and we are not going to.

M

We can see the data abilities Moody. This is DC, there is augment which includes transport protocol coding, format, secure transport and.

M

Banking, subscriptions and adaptive subscription those proposals among system of abilities, model and young, and it.

D

M

Also provide additional data export tributes also, so so she eighty-three system it is, and young update.

M

So next I space.

M

We sync the key value and providing flexibility to select different transport, specific parameters and detect, whether is a device about message under threshold based the telemetry and therefore it I inspected a failure, and you should know message exchange between client and server. So we called adoption with as a time any comments.

E

Get a comment from Mahesh thanks.

G

Kent, sir peg, the observation that Kent already made is that you know with a lot of these drafts. The exchange of information is in the opposite direction from the receiver to the publisher. My question to you is that in the isn't, the transfer protocol pretty much decided at the time of implementation itself. So, for example, if I'm going to implement the HTTPS note of draft, which is the workgroup document, isn't the transporter and its security already decided by that implementation? Similarly, I think the HTTP is not of draft already supports message.

G

Bundling so isn't it just a question of saying: I want HTTPS notif and then everything else kind of falls in place.

M

G

J

Yeah I try to answer your question actually I I think, because this capability advertisement is used at the implementation stage. Actually so so the device or the server or client and maybe support different transport capability, so they need to negotiate which answer so so we think that some more devices may support with a more UDP transport or some device about a TCP transport. These need to be know by other kind that the client can select a rider transport protocol. So this is just one example.

J

Some other example may be the supportive and compression motor, so the client can select the right of the compare compression mode. Pounder support is another example. So right.

G

Which I agree so I'm saying that say you decide that you want to use the HTTP based notification as a way to transport messages.

G

J

Own son is a complementary to each other. Actually you actually are you assume, you know the young separation. The encoding actually is optional. You may notice that encoding, but the our assumption is the welcome letter publisher to advertise the encoding to the receiver, the receivers to select a writer encoding. So so these specify the encoding in using young you using young pushy separation messages so.

J

This is something you know overlap with. What do you propose the in HTTP notification? I think this is something compromise complimentary, because your proposed the dynamic receiver capability discovery is happen after young pushy submission message that is sent from the client to the server right.

G

Yeah but maybe I should let Robert ask this question. I, don't want to take the time since we're running late. Anything.

D

Just a quick comment, not social distraught, but more generally, that these obviously compete and capabilities yeah model.

D

This was done and there's a few drafts that augmenting those capabilities and I think Germany capabilities, something that needs to be solved, but I have a slightly sent we're adding to that on the fly and whether we as a working, which we tried to look at hip abilities as a more general issue and working out what the long-term strategy should be for the capabilities in yang I just have a slight concern that we are and developing a solution flying here without really known exactly what we're aiming for.

D

So to look at that yang model of one particular example, ease example, York transport protocol as a leaf and coating for us a leaf, but I'd have thought those would need to be a leaflet or something that you could support multiple transport protocols of encodings rather than a single leaf. So so my cause basically I wonder whether there's a bigger underlying issue here that we need to understand. I.

J

Think that that's a good comment actually where you can allow the server to support multiple transport protocol which changes that changes a living to the latest. He says it's a it's a it's a user option we can, we can do. We can improve step by step. What is I.

J

Think, oh, you know, you know notification capital provided generic. You know, capability model you know is allow you know more extension, so for for transport to parameter than need to be negotiated. Without this transporter parameter negotiation you you need, you know to the pre configuration or without the pre conversation you may need to an apple. You know some failure. You may use RPC a reply to indicated that that will cause the additional message. The exchanges, so we think probably we need to you- know- have these kind of transport, the telemetry data transport capability, Rob.

E

As chair the notification capabilities draft is a post working. Your blasts called, but I've been holding on to it, because there is a required instance discussion that respond out of it. I think it's currently being resolved in that lot working group. But to your comment, are you suggesting that perhaps we should maybe see how this work should be incorporated there, or can you restate yours.

D

So I think the district AP with cables drafts of progress as as before I don't want to slow that down. It's got a key, a key need for that. My concern is more generally that the whole issue of capabilities is wider than just these particular points. There's lots of different capabilities of devices have and having a very general solution for capabilities would be useful and I. Don't quite know whether that model is sufficient.

D

Longer-Term I just think we should find a look at all those issues on it, something they are and make sure that the solution that we're building encompasses everything that we need it to do, rather than just solving a few point pain point that.

E

Makes sense it does end up bringing a level up. This conversation I mean again looking at the agenda, there's related to notifications here, I'm thinking that perhaps we should consider having a design team. You know really focus on this. That holistically, you know and then bundle it. All up is I mean I almost feel like I need a presentation for the presentations. At this point it seemed like a bunch said two disjoint efforts and- and that concerns me.

K

Fear that some of these arguments, when I changed the drafts to includes January capabilities, but there are so many capabilities. So if we keep on discussing all of them, that drafts will stay at my draft. The notification capabilities will take another three years. So I am very happy to hear from Rob that that can proceed and although I support that we should have a general discussion about capabilities. But we don't keep this draft on for another three years.

E

Okay, great, thank you. The next presentation I think we should move to it. Jim presenting adaptive subscription yang education.

J

Hi Ron, my name is Jim I'm here to put an adaptive subscription. It's a new draft and the next.

J

So this standard, this chapter, relatively young, pushes up wishing and a young push, stop reaching allow decline and to subscribe. You can use data store, updated without pooling and the right now they supported to kind of the solution mode. One is political subdivision which has fixed update interval or period after interval. The second is I'm changing sub, switching which it will be triggered by datastore content change or by the protocol operation type of change on the datastore content.

J

This young subtweeting will be used between the client and server, but in some cases a both client and server of connector and a publisher may be completed in some article theoretical interval. Actually, this multiple period interval will be applied to the same set of data. Obviously one of typical example is: what is a scenario in you? May the signal stress may change over the time, Oh sirs tapping why this man sous-sous put up a packet delay, maybe change over the time. So this is a mostly caused by the radio results contention.

J

So, in this case, such cases, what we really like to do is adjust a seminary interval to adapt with a change of the signal, stress or throughput or per packet delay. Next.

J

Here is a use cases. We we call the wireless network performance management, something we made actually is. We will, you know, see the wireless network performance as a good quality when a signals chance or really risk resource strength indicator is greater than the minus 35 DBM and based on my experience and aesthetic is actually when, when you connect a 100 stampin points, usually more than 80 sampling points will have sickness chance that it winters end.

J

Second 5 DB, only less than 27 points actually weighs signal strands at it. Less than minus 65 DB m. So in that case that we expect to connect more sampling points with weaker signal strengths.

J

We may change change the sample in interval into high value or low value when we set as a sample interval into the low value- and we can, you know, get a curve actually better reflected the signal strands, change trend, we don't you can identify the micro post or you can see the the peak point or very point a baton. You know curves and but is that the price you should pay? Actually, you may need to consume large amount of system.

J

I will resource so another shows you set the sample interval into the low low value, a high high high high value. Actually, that means you not enough for seven point: you can connect. Actually you can just get a cause. Isis I am changing curve. Actually, so our proposal is look over the balance between the system, Iowa resource consumption and a temporary period value selection. So we can change some sampling period, a based on certain condition.

J

Solution we propose this is a model we proposed. Actually, in this model, you can say actually welcome and a subscription model with a set of parameter related to the condition, expression and interval that need to be stretch to and in a condition expression. The most important parameter is to tresco the value, and so you need to make sure that they'd, obviously we observe the need support the threshold of handling, so we may use a telemetry data tacking cavity to notify whether its threshold is support.

J

All we can use all the telemetry data export capital to know to to indicate a storage code. Support for specify data object will observe that so we can initiate the adaptive subscription to so so, where you can, you know, define the specific conditions that issue should be met and based on the condition we witnessed. We can change the interval accordingly.

J

J

It's a usage example: we provided it so in this example, we can monitor the voltage or power in the optical device. Also, we have optical box telemetry model so to support attentive adaptable subscription. We can, you know, define the condition expressing an ax and corresponding period, and so you can see we can support it to to interval with each interval has a different condition that should be met next.

J

We sync for adaptive solution. Actually, it's very useful to address a meeting performance bottleneck and identify some macro policy in some database, but it'll be today and it's a anomaly data analysis and a so. We would have like a walking boot to consider to top as looking watch out, Thanks comments.

E

This is Kent as a contributor, interesting idea. I was wondering also, if it's possible to specify percentages. So friends I mean when monitoring, counters and whatnot many times or not counters, but you know things like voltages. It only matters when the when the value has changed by a certain percent. So as long as it's within a window of plus or minus, you know one or two percent: it's not really interesting to communicate to a controller, but if it varies you know plus or minus five percent, then it becomes interesting. You know.

E

Some sort of relative description might be also an interesting I'm, going to consider that yeah.

J

You know condition expression. We have sorry pammi that we called our value of the name and another operator. Actually these are generic. Generic I mean we propose. We actually can use these three familiar to to support a percentage and some other case away. When we come up come up.

E

Okay. Thank you. The next presentation is from Pierre subscription to multiple stream originators.

C

It's the bus, who's presenting it all right.

L

It's subscription to multiple stream or originators. The next slide slide please. So this stuff was first posted in July 2017. The working code adopted it in September 2017 and received his tenth revision by November 2019 in the United, which is a group of operators, vendors and universities. We are currently working on an open source software for the data collection and I've actually want to have at ITF 108 hackathon a very good working solution and test it there. Therefore, we do submit this draft on March 20 20 nice slide.

L

So the main problem statement here is that, as an as an epoch operator, we're seeing that SNMP is not scaling anymore. We see the movement going slowly for towards young poof, streaming Elementary and within yang push. We want to give witty stuff a possibility that matrix can be pushed directly from the line card without involving the terrible processor towards the collector, and that's of that's in the line card.

L

Our processor that only is available within distributed forwarding playing in there, especially lots of metrics, can be export exposed, and we want to make sure that the that high-frequency is actually not having any negative impact on the health processor, because we believe that the process of its main core function should not be forwarding matrix from the line cost to look towards to collect. It. Has all the functions which should their app assist say so take care so next slide. Please.

L

So these are the main software components involved. On the collector side, we have the subscriber and the receiver where the subscriber maintains the subscription at the publisher where the receiver receives the matrix from from the publisher and the publisher, we have two holes, the master and the agent. The agent resides on the line cadre the master decides on the out processor.

L

The agent is responsible for the component capability, where the pub there, the master, is responsible for there, the router, the global capability. So when the master is receiving the subscription from the subscriber, it further puts the subscription down to the component subscription based on the component capability of the agent when the agent receives the component subscription. Based on that configuration, it pushes the matrix down to the receiver next slide.

L

So for the the message generator ID, these would be the extensions on the on the subscriptions needed so to generate that message generate. The idea is basically that a piece of information which maps to which software to which line card which messages belong to so it's a unique identifier. So at the data collection we relate which message is actually relates to which to which software component on the router, so in case of fragmentation, for instance, that the right fragments can be breathed together.

L

That's four configured and dynamic subscription, so next side, so for dynamic subscription. This would be call flow when the subscription is established. The published master basically pushes down the subscription to the publisher agent and when then in the notification message has messages are produced. It contains the subscription ID with map maps to the subscription to the young model to schema where the generator ID is mapping towards the the software process, which is actually generated.

L

The messages and further down below you also see what's what's happening when the subscription is modified according to that modification, and basically that message Shannon write, a generator ID could change next slight ik configured subscription is almost identical. The only difference between than any subscription is that in this case the subscription was established previously out of bankfull through CLI, for instance. Otherwise, it's the same saitec.

L

We would like to receive from the working group feedback and also from the mailing list and would like to call for adoption.

G

Okay, Kent: are you that and.

E

I think it aside too early to call for adoption this time definitely should take less, but there's a second presentation also interest of time. We should move to that sure.

G

Getting quickly I asked my question and then maybe it could be answered as part of the second presentation.

G

So as a chair I noticed that, and there were two drafts with the name subscription to multiple string resonators. So I was wondering why there were two drafts with the same name and why I was the other draft, not reused.

G

If the content is the same, that's one confusion that you might want to clarify there in more important question. Is this draft, as you noted, was discussed in 104 and 105, where both Tim, Kari and Kent had raised a couple of issues around a closed ecosystem that was described at least, and then how does and have questions related to anyone how they discover a master?

G

How would a subscriber know that there are line cards so is that still in the scope of this draft or out of scope- and you can choose to answer now or as part of the second presentation.

L

That would be part of this draft.

L

G

Saying it isn't scope it.

L

Is in scope, yes,.

G

And so you just maybe you want to clarify with the other why there are two drafts at the same name, I.

L

Have to check that with Yuen long, which is the author of the draft and from the previous software-wise, has changed the Alliance for that on the list.

G

So we can move to the next draft.

E

There's only two minutes for the next presentation: I think.

C

The challenge now to respect out of respect, probably for I, take the challenge. I'll, try. Okay, all right! I wrote this oh yeah, so this is, can I start.

E

Well, actually, really good just answer. The question which is: are you presenting a UDP notice notification, channel I'm.

C

Udp-Based book channel Oh am I you a my impatience.

E

Here no I mean just like my mentioned a second ago. There was a previous try for the same name. Draft was adopted, but then we, the working group, unadopted it and, if isn't why it was unadapted, was because it wasn't presenting a notification I'd like a notice channel, we said come back and define a note of channel, though, is with that. Can you please.

C

Presentation sure so I'm I'm a researcher at the National Institute for Applied Science in force. This is joint work with all the ones you know. Additional rotors are myself Thomas from C's from Swiss come and borrow from NTT us next sorry, previously I know you were already on that one yeah, so yeah as you can see from the history of that drug. This looks like a draft on life support, so I understand you're, questioning the. Why, instead of the hull so well, I'll try I'll try to do with quickly.

C

So my own way is quite simple. In September last year, I got a call from Perot asking me if I was interested in evaluating the performance of a warrior water that had been deployed in Thomas lab at Swisscom and which was been swinging and which was blasting streams of UDP stuff from its tankers. So I said, of course, I cannot refuse an offer like that, and so currently the statuses that we are playing with these three days in the lab and if I understood correctly, we'll be using a man really soon.

C

So regarding the wife or the other people, so I quickly made a whole interview of the entire team and lightning.

G

Explanation to make sure.

C

We are all agreeing on this, the operators. Basically we want choices. We want to extract the most from all boxes. Then, of course, would like to have fully capable top-notch premium gears everywhere in their network I'll, to say that Donald Rumsfeld was saying you go to war with the army that you have not with the army. You may want to or wish to have it of in the future time right.

C

So this is the reality of things and then, when you look at it from a technical point of view, while put it to wrap it all up, what they want is IP fix. It is a migration from IP fix to streaming telemetry without a loss of functionality.

C

That seems pretty clear to you now. If we move to the last.

C

Draft the Warwick is an affiliated. Well when you ask them the reason why it's getting pretty simple operators as then for what you find in the abstract of the draft when we proposed something the operator said, go for it and they did it and now we're testing it. That's a simple is that so now, if you want, you can move to slide 8 so.

C

Although I guess this will be the usual stuff on the working group, drug production, so what I would say is that, indeed, before we decide to adopt it or not, we need more batch. We need more input if you want us to write a draft with the names of operators that are telling us to go for it we'll do it no problem. If this is a problem of lacking of customer requests, this is not a problem to find.

C

One question that I have to the working group is that if this does not go through with the operators that are behind this, what do you suggest because we are going to end up with juniper with their own version of it proprietary doing the same thing in a slightly different way? We have one way whodunit and Cisco. I do know. I do not know so on this. If it's get into the working group status and then.

D

C

Stage we decide to pick the solution of another one that gives us the same properties. Fine, we'll kill the draft. We don't care right, but if we don't do that, I'm going to have a coup finding how to proceed with the operators that are wanting this. So if you could work on clarifying those things, that would be nice if there are no other questions. I think I can give the ball to two to blue.

G

So very quickly, almost I think if you could answer the. Why and if you go back to 104 and 105 meeting, notes the questions that were asked. If you could try to answer that on the list. That would help sir.

K

We'll do that no problem.

G

For the last presentation Joe or that Manish who's presenting I can degrees.

E

Mention we are over time, but it's a personal thing and for those who would like to stay on, we can in turn have this last presentation.

B

Yeah, thank you, and this is Manish.

G

Sorry, wrong presentation: okay,.

E

The name of the placeless lip automation, capabilities verdict latest edition.

G

Now the name of the presentation, it's.

B

Not capabilities for closeness.

G

G

All right, Manish thank.

B

You this is a work that we've been doing here, I'm trying to demonstrate you about partnered capabilities and is basically focusing on having observable and providing some optimization related to those observable for operational data collection. So yeah, can you please go next yeah. So if you see the high level right, the servers I mean the telemetry data that is exported out for post-processing analytics loop automation.

B

There is a granular information that is required for this telemetry data right to automate a selection right, so that exam information is what we are trying to explain in this draft. Can you go next slide Mahesh, so one of the example of that information that we're talking about here is for like if interface counter is not updated.

B

What country then 30-second in the hardware any any collector if this is polling, you know at a frequency lesser than that we end up having wrong conclusion at their site. So these South having this information available to the to the client side, which is a collection side, is important here, so that they don't draw the wrong conclusion next slide yeah. So this is just a reminder that there is already these two draft IETF system capabilities, an idea from occupation capabilities. We are adding another draft using IDF system, abilities or augmentation.

B

Can you go to the next slide? Please yeah. So this is a. This is a this is about ITF system, node, specific metadata that we are we are proposing here and what we're doing is augmenting idea of system capabilities to provide the so that the optimum operational data collection and we achieved by the clients we also provide to our pcs to simplify some of the operations. You know that that can fetch the information at the real time for fault for the folder and word telemetry metadata I'll.

B

Take an example in the next slide to demonstrate that yeah. So if you take the yang model data tree, the information that we are augmenting to system capabilities, node, selector power path defined in the data model, one other thing is minimum observable period, which is basically this turn of this information will allow allow the collectors allow the client sites to know what is a minimum period that they can subscribe.

B

The stream for on on a node selector right so be up all right, and then we have the suggested observable period, which is the which is telling from the vendor perspective. What is a suggestive was ever period that this part can support. This value can come from for filing done by the by a by the vendor and at the same time, based on the implementation, if they have a constraint about what is a hardware capability for minimum, but there is a software you know constrained around.

B

So what's a suggested observable period for that node selector, there is a concept called optimized measurement point, so this basically is an empty type, which means a particular node selector when a client is requesting the metadata associated or on the device. If a node is either is having this leaf, node optimized measurement point. That means that particular expertise is on the device is implemented as a optimum, most optimal point where the collection should you take an example in some of the server in the yang data model.

B

Tree bunch of leaves are grouped together where your actual collection is happening, so it's optimal to collect, though, that data in one shot, which is collecting individual leaf inside that group of data. So again, implementation can have a more optimization specific techniques here, but the idea is that to export that information to the client side so that they know what exactly experts inside a particular inside a particular data model are the most optimal measurement point for them to collect the data. So it's basically represent that point. We have this discord also.

B

This is not only a concept, it is also implemented implemented. You know idea, and then we have. This is a bunch of additional informations required, like just telling you know corresponding map ID and and a related note, and this in self-exploration Airy. The relative node basically is for known MDA compliant implementation.

B

What basically is interested interested is when we are exporting the data for let's say for data expire it is. It is interested to the client that what configuration exactly this this particular you know, expert belongs to. This is basically coming as a part of engineering.

B

You know initiative that we have and learnings from that service assurance perspective. It is important to tie the configuration node with the operational node and for known I'm there and in the implementation it is. It is not you know clear how that can be achieved, vital vendors in a standardized manner. So that's the one item that we put together here from our service assurance. No learnings can have next, so you please nice. So then we have the to our pcs get measurement metadata. So this is actually an important aspect.

B

In a previous slide, I explained that we have a optimized measurement points and we have a minimum observable period and we had the suggested observable period, so the static data right so that, yes, that's why the capabilities, but in many time point is in clustering, no at any. At that this moment what is exactly observable period that I must I must use on this path. So that's where this RPC has been defined, that for, for example, you you will pass any expert and the implementation needs to take that X path and figure it out.

B

What is the optimal maximum point for this X part? If a server supports this feature, because this is a feature tag item, if a server support the optimized optimized measurement point feature, then it will compute the observable period that a client can subscribe to at that moment right. So that's. Why computed, because it's done at real time dynamically and written the returned, the information back to the clients, so at that moment, client can subscribe using this computed of zero period.

B

In most scenarios, this will be greater than the minimum observable that we have in the static list explained as a part of augmentation acting measurements. So this is an open item. What, basically this measurement is define, is we have more subscriptions on the router while we are computing, the observable period for clients to subscribe to at real time? There are all. There are already multiple measurements happening on the router, so that information is also provided.

B

It gives the client the flexibility to choose now whether they want to do what the device is recommending them computed observable period at the real time, or they want to do their own optimization. Using the leaf list, which is basically provided as a part of the active measurements which has information about all active measurements happening for that part. At this you know on the device we have an also RPC called get system, node capabilities. So this is just a just to simple operation like we have the augmentation.

B

If someone is interested in the node capabilities, you know you just pass a node and all of its subtree.

B

So this is also important for operationally, because IETF system capabilities model which pH we augmented you know, doesn't have an flexibility to query the subtree. You know on the device if someone is interested only in subset of the capabilities, so this RPC will write them benefit about that. So next slide. Please no open issues that we have from early feedback that we did internally related note like it said it's for non MD implementations. If we can be split it into config northern state node, and then we also have these examples.

B

We took care of any design you as a part of the draft how the RFC's are used with different options, so I expand on active measurement. Use case so that all the open issue, so we have provided some examples that we need to close on. We also.

D

B

Open issue zone rather than nano second, is the right metric to used or or the micro. Second is a one that we should aim for. So I'd like to thank you for giving me time it is. It was after 9:00, I really appreciate, they're, giving me a spot and listening me here.

G

Thanks finish so I think just need to close on this particular chart. I think as Robert mentioned there, there is plenty of interest in the working group for work related to node and node capabilities. I think we probably need to form some team around it to see. What are the problems that we want to try to solve in this working group.

B

I think I'm I think one thing I want to highlight here is this optimized measurement point is also already implemented in some of our implementation. So it's not just an idea, so we have an implementation as well.

G

Any final comments: I'm.

E

G

More all right, that pretty much concludes this Netcom for group session. Thank you.