Internet Engineering Task Force Interim Meetings, 21 Apr 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: ALTO WG Interim Meeting, 2020-04-21

Description

ALTO WG Interim Meeting, 2020-04-21

A

Yeah right, yeah, I think thanks for the reminder. Martin um I would have forgotten so and again welcome to the auto working group meeting here in a virtual interim meeting and as I was saying, this is nothing new for us. We've done virtual interviews before I'm, also with Cisco WebEx, and we get a little behind schedule now, because the Troublesome child at the screen sharing working when everything is working so welcome next slide. Bj.

A

Please, yes, this is the nowhere and I think I, don't have to say much about this and this. These are the general ITF routes and this awesome plan for virtual interim meetings. So you are all am obliged to them to to note any IPR. You might know related to work we discuss here. Everything is public everything and this you know available in the public domain and so forth and BJ I cannot see. You decides anymore.

B

C

C

All right, let me share it again,.

A

So yeah I think everybody's familiar with this, but just tuned so that you know this. The the ITF rules apply and I'm. Also, please open the Chet Martin already was our new idea also and and posted a link for the blue sheet, so we have in a blue sheet. So please sign the blue sheets here next slide. Please.

A

So yeah, what's our status from a general perspective, not much. We made some progress, but the general status is the same as in the past meetings, we are focusing on finishing the charter items, the working group items and for those have set up the agenda, so we're going to discuss those first and we want to finish the nice as soon as possible and we are getting there, and so we will discuss it in a minute and yeah thanks to our new ad martin for updating the milestones.

A

So the deadlines of the milestones, they were a little bit outdated, and so we updated them. You can have a look at the on the website.

A

A

As I was saying, the agenda has been set up and we discussed the Charter or working group arms first and then we have a little bit time left for other new interesting work and which is not covered by the current Charter. Any comments on the agenda any any.

A

Change the agenda: I! Guess that's not the case. Oh yeah I see College Martinez asked about taking minutes. We should have note takers, so Richard said he could take notes and while.

D

A

Yeah I have to pronounce the name right. Keo is not.

D

A

Richard is presenting and also Sabine said she would take notes and mr. bean on the car not see her so yeah. We have no takers, but thanks Martin for asking I don't see on the call, though I have to check. Oh, he means a ping yeah.

A

E

C

She's not on the call so far yeah.

A

So, but we have one and we have so we have one note-taker for now and I hope she will join soon. Okay, maybe somebody can send her an email in the back.

A

A

Yeah, so what's our progress, so the last meeting was in Singapore and November. We had two interim chairs there, because Vijay and I blow both couldn't make it so thanks again and for that and our general status is we made some progress. So we have the new RFC I, see 686, which is the cross domain server discovery document that finally made it. We have to humans. Finally, in the IFC edit in Cuba cross scanner and SSE.

A

So they are, you know, out of our hands so to speak and I think it's just editing now and they will be soon published at our FCS, our C's and we have the i2c deny request routing document that has passed a working blast call, and so we have a presentation. This document was further, but I think this is pretty much done because it passed the working blast call and it also got feedback from the CDA and I working groups. One of the individual reviewers during this working class call us from Syria and I.

A

Yes, I'm a co-author of this, so speaking as optimal. So we have addressed all the comments, but we can discuss after Jensen presents this documents so.

F

A

This document, so basically, we have three remaining milestones.

C

Yourself, if you're not talking.

A

So, as I was saying so we have three remaining milestones: they are an effective performance, metrics, unified properties. You can see it on the slide. Each of them is very much sure for each of them we have a new version and for each of them we have a presentation today. So this is our main focus discuss in these documents and in detail today, what's missing to get it out of the working group other and major issues, other minor issues, what's the sense, so that's our focus and next slide. Please.

A

Everybody's information, we we have an interesting sick home workshop this year and which is very related to what we're doing in also, of course, sacrum is a research conference, but a lot of people. The people here work also in research as well as in standardization. So there's a workshop on network application, integration and co.design the NAI workshop, and I took it here from the website.

A

The scope is contributions to the design, principles and real implementations of systems that enable network application code design, so I think I'll choose even specifically mentioned also in the call for papers, and if you want to know more, you can Richard can say something later on and about it. Maybe MSE is a TPC Cherico TPC chair and the you can find a link here and the submission deadline is very soon April 30. So if you interested please consider submitting to this conference because it's very related to all right.

E

So I'm kind of just like give one me okay, so the deadline is April, 36 pages and we're I think were also causing possibility of accepting two-page abstract and it's going to be very interesting, really motivated by out of work and, for example, we have two very interesting keynote one by Albert Greenberg, who is from Microsoft area team and Bruce Maggs from used to be Akamai and now at Duke. So we have a quite equal, interesting kind of had a released workshop.

E

So we really encourage people to submit to this conference a Twitter's workshop and they're likely by the way, if you worry about traveling, most likely its income even low is on August 14th in New, York City, but most likely will be virtual server for might be. Make attending easier. Okay back to you, yeah sorry about that.

A

Thanks to this were actually good, good points and so yeah. Please keep this in mind and next slide yeah. So this is. This is what we try and I've prepared, and we also want and thank Martin how our area director we had a chat with them. Also in preparation where we discussed the status of the working group and remember, we really want to wrap up all existing working work group items and as soon as possible, that's our focus and any new work will be. You know a secondary thought for us right now.

A

Of course, we allocated agenda time to this if there's space, but we won't really want to focus on the Charter items, any other business.

E

A

Has a something like for God or we forgot or wants to mention something.

C

All right, please sign the blue sheets. Martin has put out a URL on the chat, so please make sure you windows.

A

Okay, if there's no, if there are no other comments, I'm also looking at the Chet here can.

B

We should we talk about the key mechanics say again, cue mechanics. So if you want to ask a question or comment on a presentation in the chapter but plus Q, if you want to leave the chat, put a minus, Q or Q doesn't really matter what and that's how we can kind of maintain a line. You wish to comment yeah.

G

A

Yeah yeah, that's yeah, I wasn't and that's that's, that's very good information. I! Don't think we did it in the past, but for sure that's good. It's.

B

The norm we've been using for 107 meetings. The volumes are higher than usual.

C

I'm gonna be monitoring the the chat. So if anybody has any question and want to queue up to the virtual mic, just indicate with a plus Q and I'll try to keep track of who was there first, okay,.

C

Given that this is obviously the speaker cannot see the queue, so if I feel the question is urgent. I'll interrupt the speaker, you know if, if you can indicate whether you want to interrupt the speaker that might help me. Otherwise we can take the questions at the end of the session.

C

If the speaker is also looking at the chat session- and you know wants to wants to say something so I guess we'll fill it out as we go along. But let's continue.

C

All right, I'm going to queue up, CD ni, that's the next one I.

A

See Jason so ready to.

D

C

Jensen the floors yours I.

E

Think it doesn't.

H

Have microphone.

E

E

From what I see it look like with a microphone, so he.

I

Can use my microphone we're in the same apartment? Let me.

J

Yes, sorry, maybe my macro has some issues so.

J

Today, I just freaked over the updates property as to like a meeting what so, what changes and it's about this in a document the last test.

J

And so, after the last time meeting way, we have put a human to label at car and give some comments from the outer on this yeah working group so have some have three revisions iterated after that and addressed all the comments and major the promoting atyminius and so update the examples to pick some grammar issues in the sample and update the settings say the background section and this critic considerations I see the document is already already stable and I repeat some minor issues, many issues, actually it it because it went on the Becker case and we should have some updates after G if I prompted yes tables.

J

So I will talk about the fourfold for the major update for for the tedious of static I can look at the document. The nest, as I will quickly few some money. Some examples undoing the remaining shoe Tom's a interpretive her nest, one okay, so in the new examples way. So this example suit a actually shoot a server on to simply deliver clothes. Obviously, first AB CIA I don't have an object from the HTTP 1.1 to the TPS 1.1 and creamers example is around costly, try to use the Jason Martinez, but the previous the FCI resource.

J

The response include an article object, so many objects but be removed if a usually it is murder past. So updating example to use a full ribbon is life, so, let's artist, the first artist and next one another. A disease injected example for the FCI research using the auto key PID footprint and the example to cut the FCI resource depend on never vibe, so the lateral magnet they also changed.

J

So it's in the example to super close, a FCI recently dependent to the never my freezer, never map up today, so the service can accept again the simply after, depending on her birthday, Ashley I resource. Okay. So that's right.

J

Issues become a second six update can to actually document use the property map to for one thing, from the footprint to three.

J

The problematic related the aniridia is not stable. Now, so we have some discussing maybe Sabine times this pinwheel Tipton a bit about discussing and the hot are today is part motor planning and whatever it Saturday entity, tab, City, FC, a capabilities is a resource benefit and a colony be asserted associated with the synapse, a resource.

J

List information somewhere in the document, but.

J

What about the hot related part up and down the update, update, interrupted document, and also the problem at also be used to curiously apps, the AgrAbility, the up the pide entities.

J

Submission that the associated resource up the key editor name must be dependent, never mind. Instead of the s today, ab c-- are resolved, so that should be admitted in the revision of the document. So next one.

J

The planning of the state document that were trying to attach a property map relative part after they, in fact property document stabilized and there we can discuss automotive, the last date so.

C

Ism is the idea to move on CD and I and unified properties. Together there annancy.

E

Okay, maybe if I make a mean comment: it's okay, oh okay, yeah I! Think after that would be a question for the group and I've said by the self. It's really really independent, given that we demonstrate to that and the you know eat this: yeah really become a unified property, three or four introduced dependency, but that's probably too much a dramatic change and I think which is cassia stake and one possibility, of course is we don't even depend on unified properties, we'll just remove that part, but that's probably a lot of cut.

E

Then, if we don't do this lot of the cut and then I think I have to really wait until unified property fully stabilized, because otherwise I don't believe CI itself can really move more on people. There's a big part. Well, there's a section depending on example,.

C

So that's not a problem: I mean if it depends on unified properties, then it depends on the inside properties. There's a number of crafts that depend on each other and move to the other as a group, so I don't think we need to low-bottom eyes, CD and I. If it depends on unified property, we just move them together. It's just a process. Question.

E

So I would think another. So I guess maybe the guidance from you is. That's really still keep the dependency right. Yeah.

C

Yeah, we'll just move those together if they are dependent on on, we have cDNA, is dependent on unified properties. So it's more process, question I. Don't think we need to add any work to our cue to take out dependency in CDN I. That's just more work, we'll just move these together.

B

Well, I would like to clarify so when we say there's dependency, does this mean that be like the actual test of one is going to change this than the other, or just that there's a normative reference that, where that you can come together in, are you.

E

Okay, do I do I answer are waiting on the Kip, oh, maybe Kansan cancer or something okay. Well sure answer is falling. It's a it's more than a reference really there's a second talking about how to use the unified, proper taste even in attack. So therefore, if example, unified properly, a train is unpack, some syntax some way to register, and then the FC I need to change this. So they'll put a slightly stronger cup, then just like a normal reference. Oh you mean you what it's a modern and normal give reference.

E

It's even I could text it.

C

Okay, I think each.

A

It just quickly so just to clarify I'm, also I'm speaking now as chair, but I'm also co-author over this document. So I think what Richard is saying is that we have a like a classic dependency where we want to publish these RFC's together. So what you said was pretty much um the way we should go do this. We should, you know, publish, wait, wait to unified properties is done, and then we can publish them together. Because then, during the pub publication you can directly put the correct RFC number in the sea denial.

A

You can refer to unified properties with a correct and RFC number right.

B

All right, well, I mean that's gonna, be taken care of any editor cue.

B

If there's a if there's a normative reference like a new we, if the only issue is that there's a normative reference and we're getting RFC numbers right, then we can submit everything up to up to the iesg and it's all good, it's already correct itself, but if we actually have to change the underlying, if the actual text has to change based on how you P evolves, then we have to hold this the working group until until all these design details are worked out, I.

A

Understand the difference, and so but I'm not sure how to answer. Maybe Richard you been answered yet, but Richard says it was clear it we think we need more tax changes, also in CD and I as unified properties.

E

So short answer is: there's a risk taxi, so me Amina and depends on how dramatically how drastically we make the changes in unibody property if we don't reach them very, very likely. Well, there's a chance. There's tax changes beyond only the reference numbers, so I would probably have to my personal suggestion. Is we probably with a little bit and hopefully not too long, because the unified property coming up I? Think.

I

C

Ahead because of time, I think there's a good reason to believe that we need to move these together. There might be some issues for the note takers. There might be some issues on more than just an RFC number dependency, but we can sort those out on the list, but for right now it seems that we want to move these together.

C

Okay, so Sabine I know we are running late, so all the people that are coming now, if you could, you know you to the time that will be awesome. I mean I, see your online, so please go ahead. I.

C

K

Sorry I was muted, so I said: I will try to rush the presentation so next night. Please.

K

So yeah in a nutshell, for those not so familiar so as you could hear, a unified property is quite a important piece in this working group. So in a nutshell, this extends ipv4 and ipv6 and points to entities such as ipv4 and six entities, but also pids and abstracted network elements, as described in the path vector draft and on these entities.

K

Properties are conveyed. The unified properties also introduces a new information resource and media type, which is a property map, and it is filter or not. It is used as you see, by several other services. It's can be used as a standalone service on for endpoint entity properties. It is also a part of the multi-part path, vector cost map, server response that conveys properties on the ionise that are are returned. It is also used by cDNA FCI, as you just saw so the entity.

K

One thing is the entity property service, handles information, resource dependant entities and properties, meaning the ID and values of entities and property depend on given information resources. This requires a careful design for resource dependent entities and properties. This requires an unambiguous mapping of properties to applicable entities. This also requires an unambiguous mapping of entities properties to applicable information resource, meaning you can't query any entity or any property for any resource.

K

So for these reasons, this work has been a long-lasting design and specification exercise, but we are, we are progressively approaching to the convergence so next slide. Please.

K

So version that you can see on the Status pages is version number 11 that was submitted in March, so in this version so the basically the the changes that are on this version will be also revised and enriched. So I won't spend much time on the changes on version 11, but rather focus on version that are ongoing. The version 12 in preparation that is in progress and will be submitted in version 11. You have changes to simplify learning, so actually the non-normative text is of key importance because it introduces so many new.

K

There has been a huge addition work to clarify the terminology in these sections, so in version 11 that was already so that started already in version 11 on section 1, 2 3 that are non normative in section 6 5, some text was added on inheritance rules for properties in the ipv4 and 6 entity domain. Since, though, there is a section 6 that was trying to explain from which entity, actually you can get given entity, domains and given properties, and this will be a heavily revised as well and in version 11.

K

You have a new text relating to draft IETF, also part vector, as you know, this works with as a bundle with unified properties. So in so basically the text is about introducing an example and see how it works in the IRD and in how properties are conveyed within multi-part responses of for path, vector Coast. So there is no changes in the design and.

K

Explanation and cleanup and clarification are needed before this draft is submitted for working group last call, so I will go now on the new revision that is in progress next slide.

K

So what is done? What is done in this new version in progress is further explanation on section 2 and section 3 that introduce entity, domains, entity, properties and also resolves dependent entity domain and resource entity properties, because that seems quite complex. But we need to do this because the result, the information resource dependency, that we need to take care of, so in section 6.

K

Well, as section 6, was about how what entities and and domain properties you can request from which resource section, 5, 4, 5, 8, 9 our normative, and they will be further clarified and harmonized. Section 10 on path related to path vector will have more generic text for the example and section 12 on Ayana also needs a revision. So next slide. Please.

K

So, by the way, how much time do I have left.

K

Much okay, okay, so in version 12, so typically a so I'll just take a bunch of example. I mean.

C

K

C

Minutes left, okay,.

K

So that's fine! Okay, typically in in these non normative sections that introduce the concepts, we ended up with a clear definition of what is an entity domain type which its generic and goes to Ayanna registration and what is an entity domain name which is defining the auto service scope and has to has to be different from the type when the entity domain depends on a given information resource.

K

This is pretty clear now and the rest of the text should follow and be much easier to understand and read so same for the entity, property type, which is generic and goes to Ayanna and entity domain entity, property name, which has to be different because for a given property type like PID, for example.

K

This the value of this property depends on the information resource on which this property was defined, and it also depends on the information resource in which the entity on which you query this property has been, has been defined. So next slide. Please so.

K

In the section 3, there has been further addition, so with a text clarifying text explaining why why this design seems so complex actually, and so um so, there has been text saying that in an entity must be owned by exactly one entity domain name and an entity. Identifier must point exactly to one entity. So this explains all the resource, dependent design of entity, domain names and property names so on resource specific entity, domain name.

K

There are further explanation and illustration, and especially when an entity is defined with respect to an information resource, so the and explain how this entity domain name must associated, associate the relative resource, ID and entity domain type, and we given an example for network map to, for example, a PID that is only exists within a network map, a given network map, and if you take the same name for a PID another network map, it can I, it can point to a completely different set of n points. So next slide, please.

K

Yeah, so this is also a summary of the updates on resource specific entity properties, and we explained that actually we take the same approach as in the base protocol, where the property name is associated with the interim with the idea of the relative information resource so the base protocol, you define, for example, the PID property that you query on an endpoint address.

K

You associate the name of his property to the name of the information resource where it was defined so, and there is also updates that are being done on that key feature that well. This is the the this section explains that the client needs to know why property it can query on a given entity domain, and these section points to those sections that are documenting the authorized combination. For example, you can query you can query a property on an endpoint, an ipv4 entity that is defined in a network map.

K

You can so you can associate network map an entity domain ipv4, but you cannot associate network map with entity domain I&E because it doesn't make sense, and the client needs to know this so that, for example, if a server exposes irrelevant combination such as network map and IA, an if the client is able to know that it just should ignore this.

K

So there is no complicated error handling, but all the authorized associations between information, resource and entity on one side and information resource on property on the other side have to be documented in the document where these entity domain types are defined and when this, where these property types are defined. So initially we thought about making an I on a registry for that. It's too complex.

C

K

Yeah I'll be done, and so in the so we decided that it has to be document along with the text that introduces this entity domain types and this entity domain names. So next slide. Please.

K

Yeah, so this is a this is about also the entity or our key and property in your inheritance rules. That applies in some cases where entity domains can be assimilated to a set and defined by the identifier. In of these sets, this is the case for address blocks and instead of defining an individual entity properties, you just define a property on this set, and if these sets of based some strict rules of order, relation or inclusion, you can even define some inheritance rules and you this allows you to obtain substantial savings in payload.

K

So I will not go over the rest, but so we can go to the next slide. So this is. This is what these updates are. We're really fundamental, because this explains the rest of the design so I we can skip the other slide. Maybe the next slide so yeah, the the other important change will be about the ionic consideration, and so we will add a new information item telling explaining why we need to other.

K

The client has to be cognisance of the authorized association between information resources and entities and properties, and also, we will add, a new entry for the property type registry and in the next slide, I think this is. This will be the end. So thank you. So we.

C

Have time for a quick question, if anybody has also please do sign the blue sheets, there is the link to the ether pad on the chat. Only on 9 of 25 people have signed up right now. Questions on Sabine's presentation.

C

So speaking as chair, I think given Sabine's summary of the impacts to to the previous draft I think it seems reasonable to move these together, as a group I think there's enough dependencies outside of just an RFC number. At least that's what I think right now.

C

No all right if there is no other questions, then I will invite the authors of performance metric to please queue up while I put the slide on.

E

So medium here, I'll present and those will help perfect or.

D

Resourced: okay.

E

And I will talk about how to performance metrics, and so you can see that here and talk about Washington, which has not been updated and uploaded. The old version.

E

106 was version 8 and then right by the idea deadline, which is mark 20, something I know about abortion 9. In the meantime, we also made a punker change. It made several changes in version 10, but we thought at the time here to upload a new version.

F

E

Much for the mutes, ok, and so here they all kind of my car case I'm, going to essentially give the major updates from version to version 10 and the main thing really is systematic refinement of acquisition context or someone.

E

We call cause context I know will give a little bit PPOs and we find the details according to Linda performance metrics, and then we did a lot of return attacks edit, which I'll not go to D theta I will not go to details, and then we do have one or two remaining issues which we feel that were mostly ready and but we do one have get last one more round feedback from the working group next slide.

E

Please, the main change is about acquisition context and two beautiful people who are not familiar with what it is, and that was a major major milestone in terms really makes a document already and was basically I. Think that's ITF, 105, I, forgot and submitted decisions. People from auto working group, I, ppm and so on is auto, provide guidance and not Mary framework. That's really helpful. Even we're talking about performance metrics and it's not marrying a framework and also the major insight last time was also provides multiple types of guidance.

E

If you look over here- and you can see things, for example, observe values, that's your Central American results. You can see things like a target value in the middle and lower of the screen. That's a really committed value or SLA, and if you look at the lower part because they highlight a month numeric network average, which means the guidance can be somehow, for example, using, for example like weekly or monthly, they have sort of a lot of diversity in science. Isn't we should capsule them next last piece.

E

This one and I think Alameda success or I used to me. I'm pretty happy so far now is we will define something called essentially called cost contacts. I think something wants to eventually change it to maybe acquisition context, because there's a separate draft, which is also Union out of working group, has some concept called cost contact. So therefore, there's a conflict we'll find out the name, Allah called acquisition context, or so far in the document in version 10 and it's called cost context. So what cost contact is there to give guidance?

E

Remember guidance, different types of guidance, details, guidance. How do we really provide the guidance and essentially use cost contacts? Cos account? Contacts has two members, one source cos Wars one is prime meters. The source essential is very simple: high-level value and talk about the type and prime meters is to give it low-level details which of, if you look at over here, lower like your screen and the cost itself is essentially optional and then on right-hand side. You can see that they're, two members and the parameter really is essential. We think that's really a best compromise.

E

Okay, Maddie kills. At the same time, you don't really slow down the progress and the Kimo flexibility for different deployment. Settings next slide. Please.

E

Okay, and so basically- and eventually is we finalize with discussion with Louise with all people we've finalized four types are Co sources and they are.

E

If you look at the attacks in version 10 nominal SLA, an import and estimation, we made very clear that it's all preacher of also server who chooses the category because which is with guidance, you're, providing and output a mix decision, we're not gonna, enforce it, and we do say that if you don't specify what it is and four categories application mass assume is estimation, because I grabbed the most generic most big one and and and the compounds that would not be used as a key.

E

So basically stew is performance and cosmetic and mode to serve as the key. So, therefore, if you really have a multiple metric, for example, latency- and you know sing in the same resources- then gonna have crime zorro. For that case, you might separate into different information resources. You know information directory, so we have no ambiguity issue next slide piece.

E

So here, essentially, is we add in text to really give description and about four types of sources: nominal, basically, a static, a complication value- if you read it very quickly, SLA and essentially is the right from commitment which talking refers to a service level agreement, but we do mention in text that people use outer terms, for example, target committed. We say from all the other ones and so on, and here we do also give a little hint about a for SLA may be recommended.

E

Ad prime meters should give a link to the essay of a definition, an import and why important I think this one. We had ways with several other people, and my Lois is for example, what if I know explicitly my guidance is derived from my 8570 one is, then you should we import, makes it clear. So, whenever possible, we want to give an option for people to give the specific information same time about people to be a high-level, attractive web sort of an import and recommendation.

E

It's all the prime meteors, please gable, ETS, which systems you tier IV raw data emission is the most generic version. Essentially, it's just the guidance that I computed from my protocols for my measurement data, but I don't know want to really give you all the details. You could let the way I come here. The privately is really out of scope because there's not a purpose of out of zero for estimation. We clarify a little bit next slide. Please.

E

So given is one and now we feel very happy now because okay left hand side and is a definition for every perform symmetric and we for used to be version 8 we have struggle talk about. It was Mary what Miriam in the consolation. How do we do a measurement? That's so complex, so diverse, so many primate hers and now we've simplified to the right hand, side, which is caused contact, specification considerations, so we essentially keeps guidance about if it's a nominal, what you should specify and it's s8 well just specify it import.

E

We give a couple examples and where you can specify, why are you important? You imported your number, but it's really for phonetic specific, for example, and therefore a report- and you can talk about there's normal value for latency, you private bought estimation. We give a little bit details to talk about how we compute it and so on. So for this one is down for every single profile metric we're pretty happy about that next slide. Please.

E

Have a lot of other small changes, for example, use, restructure, metric name and so on. I think it's gesturing family reveal was: please follow our system to 70 to 85 format. So, therefore we're changing from metric name. We call identifiers and we have the descriptions now we're changing to a value of representation and okay, the details about them and so on, and of course we did a testament of normal references and, for example, we really want give a reference about where each Papa merit might come from.

E

There are quite a lot of sources here from ATF and accredit different documents define different metrics. We give essentially potential where they come from now they are no longer really normative. They become informative, informative. So therefore we simplified our reference structure a little bit next, please so basic clay is only remaining issue, is how much detail do with game in terms of guidance are specifying parameters and four different categories and the car recommendation is typically for nominal is no accep probity.

E

False report is gave the configuration parameter and SLA and recommendation is please give a link to the assessor is specified for important. Please specify which particle you really imply and, for example, a PDP is or AI ppm, and so on and estimation it has about give a link and link will give a description attack. Why I described this one will be essential outside of scope, the optional but their most informative. But we do want to talk keratitis about how much which information we want to specify for each category next slide.

E

Please and the of course, all the naming by the we come. The wall may change, but we've got one review and said that we do. You want to be consistent with other people's name. For example, we call one way delay. Every single thing here is most is a one way, and but arts eighty-five said the one cause, unity reaction or, for example, a unique direction, a packet loss and so on. We do not really do that. So.

E

Therefore, our question is, we want, can name their short, simple and you need to reconcile, looks pretty long, but we do want to get a feedback farm group and the second question number two is essential: I believe that's really nice reveal from you know, I believe for marking and is, for example, through the we focus on have various tables at matrix.

E

For example, a minimal latency would open, hands can be stable and so on and our latency, for example, yeah, listen, save meaning more basically its propagation delay and take a lot, probably mostly because not really like a to kill a typical, minimal, really most likely coming from, for example, essentially fiber from, for example, some like a software failures and so on, they're stable. So therefore it centrally so one way to do is how to handle this type of things.

E

And one way is we introduce more matrix, we call the metric the mean and Max, and the person tell and so on, coming stable. Another one is, we include your sister them sister, make it more poke and you centrally have a metric and your really specify. What's that to make it more stable, exactly fit to give me basic you'll, remove away the hakuna teal a most likely, because really typical doesn't happen. Will you have me and sort of for right now?

E

If we go to basically one we delay and we asked one mediate, asked me and one we, for example, last and me, and so on, server for to make a slightly more stable that we don't really make a structure change, but that's the feedback we wanna get from what you do and I believe. That's all I'm done well. We feel that were most already and maybe little single, like largest change, maybe is question two and we might need to add one or two.

E

For example, you lace and probably but we'll say: okay I'm done questions suggestions. Please.

C

Any feedback on the because we need to move it ahead. Martin please go ahead.

B

Thanks for incorporating my my feedback, um just to be clear, I would recommend minimax instead of something like one-way delay. Instead of trying to report a mean which is likely not very stable, so I don't think that's way too many um I don't want to just have a million parameters, it's probably a mistake, but that was the actual intent of mine feedback, although if he choose to do it another way, I suppose that's okay do is do we expect RTT.

B

While I understand that in real the real world, reverse tiles are often different, do we expect RTT to be different and then something other than twice the one way delay in the context of the way we generate these metrics? Okay,.

E

That's an excellent question, and so, if you look at all the numbers one way to la hub packet, delay variation and packet loss, everything is report, it all one way and you're right. So basically one way, we always think that, for example, people. So if you want the other directions in Reverse, your very end points, and then you switch the ignition and you will get the sum of delay, but the feedback you got from or such a common query that you should really keep RTP.

E

For example, I think that the question was number one by the way would be answer: question number won't be y RT became a different problem. We deal a is pocket, I'm source to destiny, I do use a line of a line line and, and the reply might come from satellite can come from and we're logging I be doing. Some was talking about 30% of asymmetry.

E

So therefore- and it's simple saving right so question answer, number one is RTP: come in essentially can be different from two times along with delay of one direction and is really sad and lurking. Precise number from Arvind and many years ago, several years ago was a 30% and the second one we introduced. Our key key was the optimization, because fantasy piece report depends on essentially is proportioned art and instead of one with you like so far, that's a central shortcut.

C

So Richard I think I am going to cut off the discussion here because we need to move ahead.

E

I mean I tried him. So what so actions to cape RTP I'll clarify why we have our TT I.

B

Mean I was actually mean if I can see the case for it. I was just um I was tempted to try to prune this list as much as we can and if, while RTT is often not twice the one way delay if, in fact, our measurement techniques in our SLA is.

D

B

Than if I even might have not been worthwhile to have both but I think it's fine to keep it in there. Okay.

E

B

Because Holloway.

E

It's on a text about that either it's fine, oh okay, good yeah. Otherwise what we wanted! Okay, great! Thank you! I'm.

C

Richard could just push for this on the list and close any issues. You have I mean your question I. Think having you know a metric. That seems reasonable. Lots of you know. Moments are captured in CSV files using using the same stuff. So if you could, just you know, push to finish these open questions. That'll be great on the list. Okay,.

E

Cautious, so, basically, summarize actually item is a top hat, because I really don't want to see. I won't see some kind of structure, delay data mean, of course, I think probably I'm, Marcus or I really can't fight. So I am suggesting and ether the column in delay caught a well propagation delay. Is that right? Well,.

C

You have metric your template. Here's metric that to me what that is implying is that you have a certain metric, you're interested in, and the stats are the moments of that metric. So you know you might have metric min metric that max metric that mean metric party, etc, etc. Good.

E

So I I allow a central. This one could be defined systematically. Basically, you use a of course. Probably only do you find a few but uses an umbrella template design to define of your own right and I'm.

C

Kind of going through that, based on, if you look at lots of these data, sets that are release for you know.

E

C

Learning training purposes now many of their headers are similarly named right, metric move moment of some sort. So, let's, let's just go with that, unless somebody has specific objections.

E

To it, I need section, if not I'm gonna, because this is slightly larger change it in the on asleep or change larger in terms of content. Again.

C

We don't have to decide that here we can take it to the lift.

C

There's no other questions, then I will queue up the next one which is going to be path vector so Richard. It's you itself again.

E

C

It's me yep, so I really no problem. Let me cue it up.

L

Okay, so hi everyone today, I'm Kai and today I'm going to present the progress on the path vector drafts. So, as you may see that we listed two versions of the draft, so the version chain is already submitted to the to the hotel tracker and we also made some changes in and should be published after the the meeting and next page space.

L

So here is the summarize of these to submit submissions. So the most significant change is in the early sections, basically in the abstract, introduction and motivation parts, and we want to revise the early parts to clarify that, basically to ease to the first issue that the extension provides information, basically to clarify what information we're trying to provide with its intention and also according to some comments from the minis, that we want to defend that.

L

This extension is not limited to give an application, but can be applied to multiple use cases, and we also have some minor modifications. According to some issues raised in the early discussions- and so here is a summary and next next page please so. The first issue we want to clarify is the goal of extension and inversion. Oh sorry, there's a spinning error, so in Russian lying we use the term pass correlations as information conveyed in pass factor which limits the scope of the extension and gives the wrong impression.

L

That extension is only for a specific time applications and starting from version 10. We actually use a more general term of abstract, immediate Network parts traversed by a pass between a source and distinction pair and I. Think we think this definition better careful, clarifies the scope of the extension and also makes a concept of any more intuitive and, in version 11 we actually use component instead of parts. So to be more consistent with terms and next page. Please and the second issue we fixed.

L

It is the motivation of the document and in version 9 we use three use cases, but do not explicitly summarize requirements so a lot. A lot of feedback says that this makes the makes it difficult to grasp the main idea of this revision part. So in version 10 we occurring to the suggestion of Sabine. We bring back the floor. Scheduling example, but but but the currents in the current document, the requirements are still too specific to the ex to the flow scheduling problem.

L

So in the next version we are going to use more general or additional requirements which is derived from the first gathering example, but can be a lot more direct generic and it can be applied to other applications.

L

So legs, pastries and you know, are some minor changes according to to the technical part of this document, and the first issue is to hand handle cost constraint and in the earlier versions, mostly in version 9 emerging trend. We had a paragraph saying that how to handle cost constraints in pass factor document and as we notice that only as the same problem is raised through the main list on the course calendar extension. So we make.

L

We also make this explicit in the past vector document saying that, at least for the moment we do in the document. We do not support constraint on the past vector response, and this is going to be included in the next revision next page. Please and the second issue is we start to change the name of any identifier to any name, and this is for two reasons. The first reason is a name actually can be small.

L

The information there, any identifier so when we use a identifier is mostly right from because it is also an entity domain type. So any identifier is also the entity.

L

Identifier of the any- and this is more gives gives more hints that a is an entity to me type and also the meaning of any is is an iteration of network components, so it is actually very similar to the PID so and because PRD uses PID name, and we think that any name would give a better link to the extra meaning of the any and also in the text, supremacy we're using some explicit definitions of basically how the any identify should be specified and in the latex document we actually refer to the because a name must become compatible, always the entity ID.

L

So we actually directly use the format of the entity, identifier to define in the unified property document, and next page please.

L

So, let's issue is also. We change a so we used to have a property called. We used to have a field called any properties which indicates the properties to be curried, pestpac should request, and we also change. We change the term to any property name, because this because property name is also a consistent ways. The entity property name in the unified property documents so like specialist and and also we noticed some park in the specification of the body, part response. So privileged.

L

We, according to the RFC hello, hi removal, remember in the number, but according to the multi-part response document, there is a parameter called start with any case. The first, which indicates the root root object of the multi multi-part response, but in in a specification of the two parts we actually use the first and the second, which implies the order of the parts in the past vector response. So in the latest revision we actually use passive extra part and if you live on a property map part to refer to a different part.

L

So the the the order of these two parts can be interchanged and the root, and we also specify that the start parameter must either be omitted, which indicates that the past vector part is the first part in the multi-part resource response and also explicitly specified. It must point to the Past vector part. So next page, please.

L

We also have fixed some comparable issues: ways the unified property documents so mostly, for example, the anti registry for probably until the property types, and previously we have the prefix of any and in the document argued about properties. This traffic's has been removed, so we also removed this prefix in the latest revision.

L

So next page, please yep, that's all for the update of prospector and next stops. I think a specimen is moving along with different property as a bundle, so we're also waiting for some final updates on the unified properties and after that, we'll I think we can talk about moving to the next stage. That's hope.

C

We have a minute questions.

C

Okay, so this also in a sense, has a dependency on unified properties all right, but beyond that you know the author feels that it's baked as much as we can right now in the working group for us to issue working group last call.

L

Yep I think I think the most dependent part on the past, you know probably would be the iron-on reissues I think when that is finalized in the unified property. I think the perspective document is mostly ready. Okay,.

C

C

I'm, just taking some notes down alright great thanks. If there's no questions, then we can go to the rest of the talks which are on non chartered items, starting with Danny's talk so Danny. If you want to queue yourself up, I'll put your slides up in a second.

E

So lady Kenna can I just give you a quick comment about working out, unified properties and passed vector and CDI is a bunch of us we're having weekly meetings so I do want to advertise I think we should post on a meeting even that you know CDI, even type property and password they're covered in such a way that I think early on with essentially a central extracting out the unified property as a module. It's very very helpful, but the terms are slowed down things because central need to really become a Christian.

E

So I do one say if other people are interested and have expertise, designing things and coming drawn in the meeting, every Wednesday I believe is from 9:30 to 10:30. Us is cost time, I think be very helpful, and then we probably can move things where things faster. If we can get I think we have a run like around eight to nine people per week right. So if a lot of people can join, I can be really helpful with all the essentially cultures and people with insights and and interesting. These three documents yeah.

C

I think we need to really move these work pieces ahead and if unified property is the common factor among all of them- and you guys discussed this in your Wednesday meetings, then for the rest of the folks on the working list. You know please do participate because, at least from the chairs perspective, we'd like to have the authors of unified property, close down any open issues and move it ahead. Since we have at least two drafts right now, depending on it.

A

Correct I can second that and one other thing, since we are lacking behind time and I would ask the authors that are presenting now for each maybe only use like 80% of their time, presentations.

C

All right, Danny the floor is yours.

C

And as Leon said, please see if you can skip time get us some time here and there.

E

Yeah I think that.

C

E

C

E

Looks like it's muted anything.

M

C

Okay, then, in that case, I'll go ahead and go to movie and Danny can maybe come back. So give me a second. Let me load movie until then.

M

C

Hold on is this Danny, or is this author for movie how.

M

The we came okay.

C

Let me load your slides, then one second.

C

All right go for it sucks.

M

Family engine dies: oh yeah next night.

M

There's a lot of new application for the faculty area of example. Now in in the this virus disease, we use a lot of cloud-based application. A cloud office lot education for a certain anonymity. Now we are using now this Fok a decay meeting and the VRC ever emerged.

M

We all know this for this application, who the quality of experience very important, therefore our for the customer and for the user and the one of the key ponder for a good girl. Quality of experience is get the bit rates or bandwidth of one network. In such case, we needed his application to get a note that network status and thereby never with next.

M

But a lot of the kitchen. Now as soon the network has a black box and a continent ism used some private or special moment to detect the network characteristics and adapt, he would attended a parameter. For example, decoding the decoding now also changed a logic function because I'm some timing to even close the video parts to only support this, the voice parts of the application.

M

This is these sang, ro packs. It's not always program, precise I, don't know well return.

M

You experience and to solve this problem. Substandard an organization, for example, is GPP standard it even forty sweetie forty. They have proposed a lot of neutral lot of technology to help the application to for what is this information to the application somebody used the defender year? It is different idea, but they also have been affected in CPP. They hope the Hamas application to the neighbor to change it a bit rate. Also the faculty.

M

You can see in the restaurant the heaven if an network is probably a function and a platform also in a forty system to sell the different services, probably suppose your function and a platform, so they also have defined in the faculty, define the clearest notification troll. It means that if something for the Cure's control them provide, the quality rated, the veloute for education and the Oscar are became to some changes in a faculty. They provided.

M

The autonomous control means that if the current across Canada issue, the provider Gangrel occurs, parameter to the application, then the application capital spirit, along with attorney q s-parameters also the faculty season, I, provide a kiss stamina. Mobility ask a support, information or application. How long this curse her media and attend after the time they can change the context or the motor function.

M

So in our people we try to use the movie.

M

Information in puia to help the application to some changes we now pista with will also to a lot of testing. We believe based on a faulty via network in our labs, and we use the sea level information the usual away permission when you use the AI technology to change this information to the network and view the based on net. Will we take it as an action to change the reading of interesting in the developer trait. In such case, we can provide a good quality and there almost any network as seconds we use the through technologies.

M

Once the reading of interesting means that we adapted encoding on the user, focused area based on the network bandwidth.

M

Also, we use the cloud gaming hours commonly to he invested his sort of reading of interest use the adaptive bitrate a census whether they used eternity in current video and came gaming fabrication.

M

We use the AR tool and the level of our ability and elaborately attended. The pity rates for the application in Circassian followed the available they're Walla Pandavas last night.

M

With for the reading of interesting, one of the key point is to find out for real written of them of the video attention. Then we were coding. The interest in the region will help. Iterate then make the other area with the NOPD rate in.

J

M

Case the whole PD rates of the video is reduced, but the acuity is done not behind. If you show on the website well, the figure were identified, the reading of them character out gaming and given more including to these characters.

M

We in such case we needed to detect the definite region and including a selective, Andy : /. This turn Oliver introduced some latency. We needed the we needed to a tablet. We change these schemes based on the network so that status in circuits we provoke Authority and done not until choose the.

M

Repowering last night.

M

We do apparent impairment with this Natoma composure. We used the full how I thought in three different networks. The first one is a really no video. The second phase is the user very quick leniency. Detection encoding, even just 10 millisecond delay, introduced the service case that we use more acute cities and division in a higher bitrate encoding.

M

In some cases, we were introduced 40 to 70 our delays. In the most cases, we use the our education with the million health. Instead occasion we use adaptive sitting detection encoding.

M

Now you see these four technologies in a suite, if and network wise, the new voice, with some no pity rates, then some high fluctuate. The second is rivers, even network will high pity rates and then no flat rate and number three is resident.

M

Literally now we have some testing on the table in the weather. Besides this table, we found in a quorum for in in the row for that we have from our artificial move. No info sure can help where it could with us. We have a good balance between the deduction delay, not discussing accuracy.

C

Wondering if you can start wrapping it up if he can get a couple minutes or so from from the slides from from the talks, that'll help us finish on time. Okay,.

M

So I am bi, I, think a lot of people and noticed her knowledge is whether to use them. Also, we use AI technology to change it a bit rate than Liberty next.

M

This adaptive piddly changed. We use the information for the user for the user level and the sea level. We also use some text in here that we only use the MCS and PRP and can pour and poetically recess.

C

Martin go ahead from the queue I.

B

Was happy to go that way to the end? Okay,.

C

M

We propose that movie architecture for the network application aware application in this. We will use these in this land workers application in the so sorry because in current times in this Rafi, besides the network, the world, let work only provide the information to the server side. Now they are not efficient a little for whether this information to the client site next.

C

Okay, so if you can wrap it up in a minute, that'll be awesome. Okay,.

M

We propose the agenda at all in the with the movie next.

M

We needed to also need a something we need to how to select a little equation, how to pander this permission with the path of the complication. How do we do the networking encoding to support sSAE and based on the person and how to support us carpet in English I think this all.

C

Pink Martin go ahead. Please.

B

Yeah thanks for submitting this draft, um it's interesting work and I actually learned a lot about a lot of things that said, um I feel like it's a little.

B

The draft is a little focused in the presentation is a little bit over, emphasizing the application and what it does with the data for what Alto is interested in what they do.

B

You need and I am curious if, rather than getting into the weeds of what the exact modulation schemes are and so on, if you can just use a lot of the tools already building an altar like min/max bandwidth to to allow the application to to make its decisions I'm a little skeptical of real-time updates to Network parameters going out to the server it seems like a lot of lower than not so server and it seems like, would have more latency than just direct measurement, which is something you dismissed in your first slide.

B

So it's a bunch of different questions. Yeah one is mid, Maxie's existing parameters. The second one is how much? How much change is there really or how dynamic do you need these? These reports to be I.

M

Think the movies you can think into at all is some difference, summed even with the traditional at work at work is start to select and the point that to communicate, save the transporter resources.

M

So it's like a selecting the path, our paper. We want to say that after path is slashed, we also need a dedicated information on the path now help the application to do the communication. So so today the ATO is to select the path to power quality of service, but our movies that, after you select a path, you also needed to tell a multi-monitor the path.

M

The quality of the inference is the big difference. Secondaries, as you say that yes is networking information. If who you may be porous some traffic, a have traffic I did not hurt if you get to the server.

M

Is well dependent? How sorry won't it play you wanna.

M

Committee, your network.

C

Okay, I see there's two questions from in bar and Richard I am you know really thinking? We need to close close the line here or after Martin what I I think? What's the best thing to do, to move expeditiously and fairly is to probably a more and Richard. Do you have comments or questions that need answers? I have.

N

A question that I would like to get an answer to.

E

Richard I do want to give a very quick comment and about.

C

So let me let's do this Richard. Please provide your comment and ignore, because this is non chartered work right now. Let's continue going to the other other presentation, and maybe if you could kindly post on the list that that might be the most expeditious way to go about this. Thank you. Richard go ahead, but a load, the well hello, Danny stock.

E

So um market very common I think your question was using mineral boundaries and maximum bandwidth bandwidth and all the other things is a really good enough or not and I think that's an excellent question. Two authors I believe em burr reading from their draft. They are talking about the primaries, for example, what's the current low basic, for example, what's the current number of concurrent users in the karna cell cell phone tower, which clearly it's not only anything like a Wii console as program metrics bill? That's more like a load level on station?

E

So that's a quick comment and I think then I think Marty, you ask question would be there was an occasion and real time and so on. I think probably implication for them. What I read, of course, engine need to cracking me would be. You probably need a very flexible way to explore X for all kinds of core metrics. The current out whole way of defining a lot of metrics probably become slightly not scalable.

E

And, of course, if you talk about pushing information, I, remember you're, pushing me every 30 seconds or even shorter right, and then it definitely what happens scalability issue and so definitely income impact on alphab. Okay, how do we get information to become it's really impossible to push information at a level you mentioned ssee, but possibly the first information like very, very quick, for example. Probably I we don't have experiment. I think I feel proud.

E

Ming about power have much more real-life experience which I will talk about, and can we read a portion information at a level? That's the meanest 10 milliseconds are 100 a seconds and west ii and was what were bound really is probably wouldn't you love work, just I'm trying to even take the comments. I'm marketing to really live like okay,.

C

Alright, thanks so clearly to the authors. There seems to be enough interest if you want to continue finessing around the edges on the working group list. That's that's fair game. Our intent right now is at least at least till the madrid IT have to finish the three open drafts. So please continue discussion on the list and Danny please go ahead. Can.

O

You hear me now: yes, okay, fine, okay, I will try to summarize my my playstation next.

C

O

Minutes, please please do so. Is it next, please so for multi domain support team for Alto, supporting multi domain use cases we need to solve the developing key questions, but information.

D

O

Multi domain application needs, but at the users of battling that information on how to design a whole framework. So basically, this presentation will provide some insights to answer which one's next please.

O

So, let's start with a motivating use case example. Next, please.

C

O

C

Sebastian, do you have a question on Danny's comment or previous one actually.

D

D

For the previous slide, I would add to the question: what is the definition of multi-domain is.

B

D

The original use case of optimizing BitTorrent in the access networks of Comcast and AT&T was so to say, multi-domain yeah.

O

Or Casey go back to this second less light, please so for this multi demand product, basically multi-domain involve multiple networks manager by different than mini-state. If Dominque also indy.

D

Deterrent using a tracker from piratebay or what not so even then we had like multiple domains. So what's new here, yeah.

O

The IDF to go beyond to Alto to support multi domain approach, because according Alto design is going to provide information in a single in my environment, No.

C

So perhaps we let Danny go ahead and maybe distinction will become clear as it talks about it.

O

Okay, so, for example, here we have a collaborative network composite of three members: main application web to reserve banded, for example, for two flows and before the application made resource allocation decision to execute two security flows, a two questions, one the first one: what is the end-to-end cost across multiple domains, for example, because in ten more, for example, resource availability sharing and the second one is the application need to find these workings of the mind and candidate bytes? Okay?

O

Next, please, we can skip this life.

O

Okay, now identify a very simple information that the application mean. Now we need to identify key issues in decoding are to design of core or forgetting multi domain negative information, for example, in multi looming scenarios, a not possible optimist, editor fit with only local available information. That is the server to playing on to communication is not enough. So an alto or an multi-domain Alto serial communication is necessary to allow changing Network information from multiple domains. Next, please.

C

Ingmar has a question in where please good.

N

C

O

So, in order to to find a valuable network source between different source and destination, page Antony, to discover which, to my adding board with for the different traffic flows and also need to discover a set of candidate, but to know how to reach the remote destination, for example, X. Please.

O

Once the multi domain connectivity is covered is performing, an application also needs to be aware about the location of different Alto servers, so in multi domain as an additional to say that will be located in different network. The mains out multi terminal to service discovery mechanism cells are are also necessary. Next, please.

O

Another challenge for its proximal to mine and all information is that each domain may have its own representation of the same network, for example, suppose that the past cost for domain B utilization chairs instead of a valuable bandwidth. So in this case, both properties are not comparable together or.

H

O

If all the men have the same width illustration share property, don't necessarily they have the same full of billing, because each domain is autonomous domain, a nature in dollar, maybe Niro's, etc.

O

Next, please application also needs a query to space all common dress or requirement to the network, for example, for a specific flow and make them provide a set of applications. Requirements such as watch really TV, Direction, symmetry quality of service matrix and so on. So decoding Alto provide some a very simple Alto resource query such as, if it's there, a network map is recognized.

O

However, more flexible query languages necessary next, please add the other is related to the scalability to solving the optimization problems specified by the application requirement that could be expensive, app, time-consuming or elect related to the privacy, because new all tests and shown a required fine granted information. So once this tension, goodra's new security and privacy concerns that please, finally, we have some mechanisms to that could be considered when designing a multi terminal. Tough wrangle. Next, please, for example, Alto require not only server to thank communication, but also communication from difference.

O

Alto server could could we consider it a Jirga color mesh deployment with a central, r2 server or multiple independent servers, gotten information from other connected domains. Next, please.

O

Multi my mechanism combining the Equus computational path, computation, it would be also define it or worse than the recited protocols. Computation protocol may be considering a 12 EP base it on VP VP less on others based on the PCE architecture, takes place.

O

If you can wrap.

C

It up in the next minute that'll be awesome. Then we have a question for sure.

O

Same for the multi multi service comedy darris I. Even recently, Alto are CC for closed Lamine service discovery, but other mechanisms. They say that PC, PC or BGP could be also consider it. Next. Please.

O

Same for a multi-domain composition, it's necessary in order to to get all the net1 information from different altar servers to a single ampere 12. The magnatraction are also we have a flexible and query language to to Alto are able to filter nah. We don't qualify the minds nets. Police orders are related to the scalability.

O

We need to recognition for for for to improve the scalability and performance or we need to mechanisms to provide accurate, good information and just in time for protecting each member domain and next the next player will continue discussion about the distinctions togetic e's, who saw the diploma concept of how to formal to me thanks.

C

For finishing a couple, mr. d, so.

D

Mr you go ahead and.

C

Ask your question and sebastian: if your question was not answered, maybe you can come back and yes.

N

I don't really want to ask a question so much as to make you aware of the fact that we are already working on this, so we have multiple networks that have neighboring alpha servers, currently we're not exchanging information through alpha between them, but we are in the process of setting it up and actually building that system.

N

So if you want to get some more deeper insights or you want to just discuss on how this works in technical terms and what we've already done, we're open for for discussion and for questions regarding that and that's when I say we I mean I'm here in their meeting good.

C

N

C

Be going after the next presentation, so Sebastian are really quick. Please go ahead so.

D

That's really interesting work I, think a subset of what you are trying to solve has already been addressed in RC, eight, six, eight six. So what we call trust of discovery aims at parts of what you are trying to do so just saying I'm interested in maybe discussing what's the gap, what you are trying to achieve and what's missing in between thanks thanks.

C

All right so again, you know some interest from the working group so continue scrolling around the edges of the draft as we get towards Madrid- and you know that said chanson you want to come back for BG pls I mean.

J

C

A

C

You can finish up a couple minutes early that'll be awesome, so you.

A

Have 16 17 minutes time left and four presentations, so maybe try to make it really quick. Okay,.

J

Sure I see my pay per click, and so that's so far doesn't about we're trying to leverage this BR I's for what blasé, auto services and inconsiderately Auto, sir I think there is some basic others.

J

Map services like the narrower map and the cost map and oblem need some narrowing permissions and out who had to wear this information to the ultimate services and the resources provide to the application and before the neighbor magnet, we needed avert information and it includes a inter-domain coverage and the intro time in the product and also they said the earth dishes. So you can get coverage this information from the conversion provision and cos map.

J

So you need to write information into the performance, metrics information, so steel, pan caddy in turn, girls and the rotten cast and I think well consider about the PTP rights because it at a particular product to have to come out at the information table so discuss how to toil and nested.

J

So the beach actually allows digger to whoever has a link state it without having dinner until paste is connected at EPA, and you can see the benefit is that is it conduct a unified interface to advertise, a ATP to erupt information in the ATP to me and the addition of different metrics and it reuse the existing procession. So you want to be trying to collect all this data source so that I have to stay at four connections, but the limitation that the only one hotel admin can be can be done by the GPRS.

J

So you cannot so these guys, I wrote had no cannot be propagated to the remote routing servers and it's a nice big.

J

So if you're trying to make the alto stop talking about the talk to they re single, your router's Singh is very efficient, so they have some efficiency. Requirements are deployed out using the DVRs so consider. The other server should only communicate with the necessary speakers, and data. Sagara should only enable the PCI dependent are the necessary sessions between the speakers. So we have some minimization, so minimum consideration. So next page.

J

So the speech issue when trying to use if it's Paris to collect the ADP to birds, information so I consider how to do that. To minimize the connections to the speakers to be select when I stacked a minimal set up the years is a tankers and maturity to ejs establish up its PR sessions with at least 100.

J

So we found this anchors and then we may fail to serve the a junk remember. It's received a pedes er attending to the to tell the server so so that the other server can cover see how the ITP permissions from say ho error from the area at widow, mains and.

J

Difficulty in setting to assume the the inter domain authority is read, his people be taught to cast a question. The the other has come from the sink of a SP, so you can control the earlier system to know the huddle that's appearing figuration between the asses.

J

It's about some consideration about the routing information collection and one of the reasons that I've seen today rough from the downstream years. It can't be hurt by ZF, promptly upstream es, so so I decide resect the minimal set up two years. It's a rather standard, so we still try to does a minimal set of them and make sure the ETS is provider of a remote provider of at least of anchors, so that they make you make the each anchor may be written to be read with the auto service and in the auto.

J

Server can O only 50 routing information from the nozzle of the anchors, but it can invert out the routing permissions, so it's necessary to the compute, a customer, so nasty.

J

A quick can read about the the PCR can be used to provide a necessary information but I'm trying to consider how to collect an ester information base with minimal these possessions. A I've got a process more efficient.

C

Okay last question: so amar: why don't you go ahead and ask a question while I load your slides, because you'll be going next? Okay, I'm.

N

Actually, hunts will be presenting I'm just here to annoy people with questions. So have you thought about the all? The considerations of these different implementations that especially a GP, brings when you're trying to calculate cost map and just having the cost map calculated based on Aesop ology does not actually tell you anything about flows or how flows are going to be map times is going to touched on that in the next one long story short.

N

We have been dealing with this problem for about three to five years now and there's a whole lot of can of worms that you're going to open with this one. If you want to discuss this in more detail again, I can only offer two we that we reach out. That said you reach out. We were very open to discussing this, but there is a whole lot of stuff when you're trying to integrate IGP and especially traffic mapping, both mapping traffic flows that you're, probably not aware of yet.

J

Actually so far, building considers a a OSPF and SS in Diaz. If you that's it, there are some, but.

N

Yeah, did you you in the end what you need to do with map traffic, not topology and mapping traffic to topology is going to be hard so.

C

I think we need to take this on the list. I think this is interesting or cuts Jenson's talk and your talk are essentially now trying to skirt. You know trying to figure out how to create these map automatically already.

P

C

Yeah and I know people are doing that, but to what extent they are, you know the efficacy of those the problems of those. um So that said, it's it's it's interesting from the evolution of Aalto, where things are trying to head out now before it used to be that the service provider would create these maps and now there's strats looking into how to create these maps about maybe five six years ago, I did some work in using the broadband information across the u.s. that was being captured by FCC to create maps.

C

So this is, and that's published as a paper as well. I can put the information out, so that said, Hans please continue and again, if you, okay, that's a couple minutes. I.

Q

C

There's lots of universe, but maybe it'll cure itself. As you talk.

Q

Hello, can you hear me.

E

Q

We can hear I can hear you, okay, perfect, okay, yeah. Let's start I, try to my very best to be as quick as possible and probably shorten a few slides next slide, please so yeah. The promise thing why we actually do automatic generation is probably well aware. So we have a big complexity, especially in the network we are connected to. We are thumbs up routers, but I. Think it's not so this is probably a really really end and large-scale network.

Q

So it also is a problem with moronic networks, with probably tens or hundreds of others as well. Then we have a diversity of customers which actually take or continuous Maps. There are a different mapping engines. They have different requirements regarding what to optimize. There are different capabilities, actually how they can consume data and what detail in the last type of automating network maps and customs? Basically, the actuality, if you automate things you can be very fast or much faster than doing it in another way. Next, slightly.

Q

Ok and yeah so, as I said, we basically fully automate of automated processing or pre-processing and regeneration of network and cosmas for net worth. Of course, we need ciders further to know for the individual piece and, of course, grouping criteria, some sort of rules that how to put those fits together and for the cosmic we need some sort of like finish affinity criteria, basically, which means how to get from one Pitt Union to the other. It's symmetric and, and actually the data systematic runs are so amazing me.

Q

What do we need for to correlate and process and ultimately generate all? This is data a lot of data next slide? Please.

Q

It does not scale very well.

Q

So the first thing is that we need yet some sort of the network, so we need to sign up and we need the topologies, the topology, not the technologies. They are situation. I can't have certain a little, but it's not up the scope.

Q

Yeah sources are already mentioned before it's a GP for some sort of signers and topology and BGP, especially for internal and also external files, which are not from within the network, but from public networks and yet also for topology in the previous code was mentioned, the BGP la where we can also or might also be able to obtain a topology formation from so. The problem here is that the data from run road might not be enough, especially if you have multiple peering points. For example, you need the information from it every pre-order better.

Q

If you get the routing payments from all others, you know how to send data to which destination in this network next slide. Please.

Q

So now we have a policy later than we can done it yeah, unfortunately, not the problem here. We only see our own networks, so everything what happens outside of our network is out at all of you and there from an IT perspective. Of course, the most traffic comes from our network. It's me just dreaming. It's gaming with download the problem.

Q

Yes, that routing protocol only tells us the forward path, so we know where, where the traffic flows to, but we cannot actually tell where the thought is when the traffic stream from entering our network, meaning the entry point, cannot be right for the routing protocol, and that is a crucial part. You cannot tell from the routing protocol where it actually comes from. You need some other mechanism to tell that the traffic is not if you don't have full control of all the aspects so that the traffic comes from outside of your network.

Q

You cannot tell for sure we're actually entering your network next slide. Please.

Q

This is where we use English point detection, so in West, Point is basically the point where a traffic flow enters your network. So it's a router and a few more specific interface on this router, where data flows from one to one destination or to all destinations engine enter your network and how we do this, so we process actually flow information together, for example, by Annette Flores for protocol.

Q

We collect this information from all water routers of the network, where we can actually see all the traffic that enters the network from outside and also monitoring from internal routers as well to very close inside of the network, and then we statistically evaluate information and with the statistical evaluation we find the side effects, for example the sauce and the route and the interface we're entering our network. So here we have a small example where death on the right side. It's on my screen. It's pretty small. We have four different screens.

Q

The red and the green screen are basically from one destination to from source to one destination, and then we have two yellow screens, which are basically coming from the same source IP, but depending on the destination entering the network on different router interfaces. Are there for the same quality you can have different in response depending on the destination. This is why you theoretically need to do a destination split on your evaluation. Well, as basically I can say, we haven't observed such behavior.

Q

So usually, all the from one source entering the network is the same yeah, ingress point and yeah and generally because of the large data, especially you need to process. Should you collect the information that you get to? Those English points is a very heavy operation, so we are sticking of bigger bits of data every second just river. This slow record next slightly.

Q

Another way to measure in this point is actually active: measurement which cannot be be home from cannot be performed from inside the network, but must be performed from the content provider. The basic way to do this, for example, using a trade route and just detect the first network of the first router you see within the target network.

Q

This is detectable, for example, of a as IPS is basically from their address space of the ISP, for example, or the whole thing this possibility to provide information before, and it makes it easier if you have a private term, you also can just take the counter part of your own border router. Then you need, if.

C

H

Wrap it up now we're getting close to the top of you. This.

Q

Is a lot crucial part actually, so, after that, I can run pretty fast and yeah.

Q

Okay and then you need to turn into the oven this. Basically, when we compile it's like, we had the idea, why not integrate information and so I? Don't map request it slightly.

Q

The idea was just to integrate information network map request. I know this is a get request at the moment, so we might need to change it to post. We try to insert data and also from our site. It is not so we cannot pre calculate those information anymore and need to be if we ever want to do this more cost-efficient, so because to speed this up.

Q

This is if this is something somebody think this was pursuing or has an idea where it can be integrated in something existent I am happily discuss with anybody was interested in next slide, so yeah we need some additional data in the end. Get that one. If you want everybody tails on it, feel free to ask me and yeah: that's I, don't need to summary. Basically the point of there and that's it all.

C

Right I know we are at 9 o clock right now. We have two more slides, two more talks to go. Yawn. Do you want to maybe extend for five minutes and and have them present their ideas quickly?

C

Okay, so those of you that want to leave. Please do but we'll go about. Maybe five minutes over Richard. You have a quick question. Yeah.

E

And while you're loading myself I begin I'm next one, basically the quick quickly reply to a hands question about the discount I'm, an American and an absolute there's. A very strong interest occur not from my side, okay, I mean by the way number one is really really impressive.

E

Work and number two ask you a question: is there interest and you post question about unki management of map and and a key person actually allow birds and brt z, ly re and light allow and his and here's very I, believe I heard him I heard from him a couple times. It said: okay, the only way eventually to handle these issues is to only man and using it's quite a trigger to measurement I, don't believe his online here. You know I think a funny also from Sprint she's.

E

Not here, I beat him there's a conflict, and but the short answer is there's a very strong interest about it. I think that's really really I think that's very increasing dramatically, which is good direction and for pursue.

C

Q

C

In the work please go ahead and on the list and we will figure out how things are moving bye-bye Madrid, ITF Richard, please, okay,.

E

And I will try to find her as soon as possible, so I talked about increment object using HTTP tuning slides, please, and basically is our very quickly gave an overview of SSE, which is the essential to server push and then why we talk about HTTP too, because actually the motivation came from the ISU next slide, please so alkalizing. He turns out to be relatively useful and provide three ghost functions, which updates instead of pull and push number two is a William doula push I, don't send a whole information to a compact, incremental encoding.

E

Number three is: if you want you want to put in from a push information, you should give a client to tell it'll tell you what information got really into it. So, therefore you should do dynamic stream come show and the current alpha SSC essentially is designed using two services once update service, you get all Albion messages and then second one stream control and which is essentially, you specify what are interesting.

E

So that's architect, excites, please why the SS design as it is is because it is HTTP one point X compared about design and now with that, it's a for architecture to services and what's even number one and it's a lot dynamic addition and removal resources. Basically, we call as it causes some streams to receive data updates, and but if, if you want porn, X woulda lost any only one request a time, and so therefore, if you send and for example, look at a client looking bottom page and a client sent a request.

E

Okay, I'm interested in this update and later, if you want to change and set- and you cannot send in requests anymore from a sink connection. So therefore you had really started a different connection. That's why it increases concept. Cop stream, control server and stream control server, can request and then through a private channel and gives to the abuse server and as a server was sending the additional all will stop all the updates. That's why in that was in architecture. Next slide. Please.

E

The second issue now with this architecture, is you in the multiplex, a multiple logical data streams and because why and when a client and the connects or the server and then essentially have a control update, tell you which stream stop and running and also database about different resources. For example, I subscribe to electron and subscribe to c3 Knight will come apps, and then you need a port reports. Three Network Maps sort of for you central. There are lots of love, logical data streams and they have different media types and some we've.

E

We encoding some with different, incremental encoding and so on. So essentially, it turns out of a complex of how to really do is multiplexing and for HTTP one poor. Next design and the design was using ssee and essentially it exists. Its GDP html5 design, basically ASU, send SS in events. Unusual event will have the immediate type and data I the media time essentially details of what really they type is you can use. For example, you can have a control update and you can also have data update and essentially mean it haven't helped.

E

So what are the encoding is, and you have data ID essentially to do multiply singh to tell what really is your multiplexing is and then we'll call you send all the data around next. Let's please.

C

So, really quick if you can wrap it up and.

E

C

What's the issue here so.

E

Basically, here's a really is single serialization and interest stream, and incus upstream you see rising into everything in a single happiest stream. Next slice, please let me speak the next place.

E

Next slide, and the basically is we can actually do better using HTTP to go back was let's please and the yeah basically really is I. Think the push from I say G is: if we have a treaty to do it end up with the same design. Can you really gain from benefit to and turns out? Itv2 can keep to bad ways. I'm, not with them benefit number one is multiplexing and, as is essential, to use a single serialization of odd streams.

E

So therefore, you see realize we do your own multiplexing, but it's to end up with only one day. Hustler for your suit can have its in freedom of line blocking, and it can allow concurrent updates and second one if you will allow by direction and sort of what you don't really in that with two connections you might be able to only once and of course, more efficiencies, header compressions, flexible controls, pre our case, our bandwidth allocation and so on.

E

Next slide, please so, basically, here's a initial design which I don't need to go to the details, and essentially is we wanted potential 1182 designed to do multiplexing and have single design and to take advantage of HTTP to to give them at a performance and the simpler structure. Next slice piece.

E

Basically, we have not uploaded yet and feedback I'm, lucky that problemi I, probably just post on the I, guess they meaningless and, of course, really question to a group is also. Would it be right now, of course, I kept also evolving from two to three and so on, and do we target to I'll really continue its central chasing on the newer transport layer and if I exam three are not so our puzzles questions online, great.

C

Thank you, I think. A lot of this introspection came from I, believe one of the 80 reviews, or so but again, let's take this onto the line online.

C

This brings us to our last talk, which probably keep it to two and a half three minutes. Please, at the end, I don't have any slides for this.

P

F T inside a source for its upload, the slice so can I hide my screen or it just to talk. You.

C

Can share your screen, but please do it really fast I'm trying to find you, okay,.

C

Could you tell me your name so I can give you the ball.

P

This is Tontine.

H

The age it's ong d.

C

Zho ng ng, okay, somebody is displaying, so please go ahead. Two minutes: okay,.

P

Puncture delivery network: it is our concept of delivering functions to the edge of the network, similar to the CDL method, which is used to deliver contents. We are using fdn to deliver functions to the edge of the network. The architecture is like this: have you go story story really quickly, so we are using a controller based mechanism to delivery functions to the edge of the network and the PI functions with me, the some operational cost or some business logic, or even application, some some application models, because the time is really limited.

P

We escape the technical details, but the function. Delivery network is deploying with the China Mobile's and China telecommunications network in China the country. It is operational in several provinces of China and we are integrating the idea of Auto out with, at the end, to make the delivery in the edge computing area more efficient and more applicable.

P

The for the implementation of the FDA reused, IBM, open risk and also Q bonus. You our disseminate the talkers and containers and make to optimize the network traffic in the system. One.

C

P

So that concludes my slide. Thank you.

I

Closing comments, yes,.

A

I can so I'm. Thank you. Everybody for participating, I think this was actually not a bad group. 25 people I think participated. So thank you and thank you for all the presentations and I can remain. I am so basically, it remains what I said in the beginning, and we want to finish the milestones first and we I think we're in a good path, but I think today's presentations show that we asked there's still some work to do and I'm happy to hear that Richard reported that there's going to be weekly meetings to progress.

A

This we Jane I would like to have this done by summer. Whatever the summer meeting is I, guess it's going to be a virtual ITF meeting again I'm pleased personally, I cannot imagine a meeting happening in Madrid in July or.

D

A

So yeah, so we want to close the milestones by then and and yeah thanks thanks to everybody, and we continue on the mailing list. I mean all the questions we had and all the progress of the documents of questions. All the continue with the mailing list. Great.

C

For those of you that haven't signed the blue sheet, please do the link is on the chat. I'll, stop the recording, but we'll be on to let people sign the blue sheets and see you in Madrid, in some form, virtual or physical, correct.

C

I

L