Internet Engineering Task Force 105, 25 Jul 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF105-ALTO-20190725-1000

Description

ALTO meeting session at IETF105
2019/07/25 1000

https://datatracker.ietf.org/meeting/105/proceedings/

A

I'm gonna go down control down, I should've done okay. We can just do it like that I'm just looking at the agenda, if we have all the presenters here so yeah Danny is here, sorry yeah saw you so I think we're good to go. So all the presenters are here. It's 10:00 a.m. um welcome to the alto working group and yeah. This is my co-chair. My co-chair BJ givanni and we are very happy. This has been a long time that we been able to chair this together, and so this is a I.

A

Don't know when the last time was we shared this together. I think.

B

It was physically Georgia.

A

Or something many years many years ago, many many years ago, but yeah, so we're happy that we can both be here and yeah, let's, let's start and if you so, we do have a summary monitor on Java right. Yes, thank you just in case. If somebody speaks up just go to the mic, please, and we do have note takers who are the note takers, Jensen Jensen, oh yeah, we have a note taker right, so I think we're all good to go. Just a reminder.

A

I think everybody in the room knows this, and this is an ITF. This is the ITF policy. Basically, so everything here is public domain and there are certain IPR rules. So please take a look. If you don't know about this, and but we just yeah, we want to inform you that these routes apply here and there certain RFC's that state these rules, and so what is the status of our working group and this slide is actually not much different than from last time.

A

We are focusing on finishing our milestones and we chairs have been sticking to that point for quite some time. Now and again we have put up the agenda today accordingly, so we will have all the working group items that address milestones first and then in the end, we have two more presentations on. You know additional stuff that we can discuss how to how to handle it. But most of the agenda is for the milestones and so in in order to make progress. Vijay and I have agreed that we the remaining draft after today's meeting.

A

We hope that we can move some out of the working group even today and but the remaining ones, and we we'll split amongst ourselves, and we will individually paying the authors to make progress by the November ITF. So we can really progress. So we will do that after this session and then basically you will, if you still have, if you're an author of a working group item that is not done, expect frequent mails from the chairs: hey. What's the status hey, what's the status yeah.

C

Yeah sure so I think we have seven drafts that we are working on out of those seven three should be going out to iesg, X tome cost calendar and there is just s SSE, so that leaves four of them and Jana and I will take two each and start hounding you guys to finish it.

C

A

And another thing, and it's good that meals arriving because that's something and we can.

D

A

Discuss with her currently, there is no meeting planned for singer pool because Vijay and I both cannot make it, and what I think we've discussed with me already is that we can again do a virtual interim, but during this week we had, and now that we have another option came up. If some of you guys- and we talked to Sabine already and if somebody else wants to temporarily chair the Working Group just for the Singapore meeting, because lots of people who will be will be there and that's something we can certainly discuss.

A

We don't have to discuss it now. We can take it offline with Mary I, don't know how you feel about it. It was just you know. Both of us cannot be there, but if all the other folks are going to be there, maybe you know somebody can share the working dress for the Singapore meeting. You have any opinions. Maybe I said that's fine right. We can do that like.

E

F

Have you remotely participating? It is yes.

G

F

Participate remotely, of course, you know we.

C

Will do all the agenda, collection and uploading everything it's just for somebody to physically be in the room.

A

And we don't have to decide this today, we can take this offline on the mailing list, and just so you know it's. Basically, both those are the options, so we definitely got to have a meeting either a virtual interim or a Singapore meeting where somebody is chairing instead of us and we will be logging in remotely and.

A

Yeah, so the agenda M has been posted and, as I said, the agenda is pretty straightforward because we're going to discuss the my stones first and then we have the last two presentations are on on my stone stuff. So we put that at the end and I hope we will have time for that. And does anybody want to comment on the agenda? Any anything? No okay, I think we posted a long time ago. I mean almost a week or something so should be good and yeah progress.

A

We had a meeting in Park M, where basically we made some progress. So just so you know the general status we have I, think seven working group items like feature set and the first we you see here on the sub bullets are almost done, so X term is already 9sg processing. They received some comments. I had a lunch with Sebastian last weekend in Stuttgart and basically they address these comments and just issued a new version. So now the the ball is in the is G chord again. So sometimes this happens.

A

You kick it to the ist. They have some discusses and you know you need to reiterate I don't know, maybe if you had the chance to I think we wanted to talk with one of the IDS right. Yes,.

F

Sir one discuss was already resolved in this. Another discussed ways probably just needs some button clicking again, but I also need to look at it, so its menu and me now actually, okay,.

A

F

The main point is, is it's making.

D

A

Right so it's out of the working group anyway and and on course, calendar. We finished yet another second working group last call and we have a presentation from the sabine on the remaining issues, but I discussed with her yesterday they're very few, so we are very hopeful that this is basically done now. I think there are just some typos and small issues that Sabine's gonna fix, and then we can kick the new version out of the working group.

A

So that's done and actually I'm almost the same status as the SSD drop, Richard uploaded a new version, and we will know I have no presentation on this today because basically, last time we already greet it's pretty much done, and so just for from a process perspective once yeah as soon as possible issue issue another working class because Richard posted in your version. So people have two weeks to comment on it, but I don't expect anything there and we certainly don't need any reviewers for that.

A

Second working class- we've done all of that, and so that means for other working room items are remaining and we have presentations on each of these and the focus is basically what's missing, for working plus call and I have just a short summary, but this is basically what has reset already cross-domain is on rise. G evaluation. Cost calendar is pretty much done. I just said that MSS is the same. We will not have a presentation, but basically it's we're. Gonna have a second working class call um CD on FCI.

A

We had a meeting yesterday, also with all the authors and I personally, think being a co-author with my chair had off this is very mature and if you look at it, I think it's already 40 pages lots of code examples Jason exams. It should be good. One thing that I think came up to my mind yesterday. Is it went once we do a working blast call on this document. We should also inform the CDN I working group and because remember this was kicked from the CDN working, a co-working CNI working group to our working group.

A

In theory, we are just doing the transport protocol and the SI denies semantics have not changed right, so the JSON objects specified. We do specifically discuss it with the co-authors. Yesterday there is RFC 8, zero, zero, eight, which I'm a co-author, and that has been finished and that defines the JSON format and we're not touching that so we're just transporting that format.

A

So in theory the CDN I working group should have not have a problem with it, but just to be fair, we're gonna, kick it inform them and- and you know I'm saying always saying in theory- you never know. Maybe they have some comments and we need to address them. We will have to see and write, and this is the remaining documents we have performance metrics where I am yeah. We have to see today doing Richards presentation how close it is to working plus call and I talked to you Richard yesterday, unified properties and path vector.

A

Remember these go as a bundle because their dependencies and I think Pat vector is more mature than unified properties. But Richard will present on that. But let's see again, our question as chairs is how close to working class call um yeah that so that's that's our quick summary on on the status. Are there any other comments, the other business?

A

C

Not Vijay, you want to show you a chi you're, the first speaker, so please get yourself situated on Medeco and for those of you that came in late and having signed the blue sheets. Please do so.

C

That's that I think we can start with highs presentation. Yes,.

A

So that's the me Tyco laptop over there I.

C

Think the actors started here.

B

E

A

Yes, all I know is me deco is aware and there's been an extension. You should be sure to request presentation: okay, hi, if you're listening, can you maybe look on your meet echo and requests permission to start presentation.

C

Yeah, so that's what I'm trying to find out where's! It's like! Oh.

A

C

A

Patman, oh right, no, it's a.

H

Patrick Turkish bath.

I

All right chi can.

C

C

Lorenzo, can you coordinate to see if I can start presenting? Otherwise we want to the next one and come back to kinda end.

E

A

Yeah just watching.

J

B

Chi, can you click on the mic, but yeah.

K

I think I'm, like that's.

B

Good yeah, okay, so we don't see you, but we hear you I, think that's good enough! Yeah and.

A

So we can hear you and just go ahead and say next slide, and then we will move to the next slide. Yeah.

K

Sure, thanks so good morning, everyone today I'm going to give some updates of the past factor extension. So next type is.

K

So this page basically gives an overview of the updates we have made in the last two revisions. Basically, since last Society meeting, we have pumped two versions, so in the 0-7 version we've found eyes the specification and also made some changes to the references regarding the specifications.

K

We revert to the cost type design back to the one we used in 0-5 and we add a new article billion request field to negotiated properties, and we also have the new property to support some use cases other than cocoa scheduling and in the latest revision we focus more on the clarification of some concepts and also ways more use cases.

K

So next slide, please so.

K

This page summarize the changes we made to the crosstab design so in 0-5, we'll use array as a cost mode and any path as cost metric in 0-5 is changed slightly different, where path vector used as the Coast mode, and we only include one property for the cosmetic and since they're all seven. We change back to the one we use the zero file. Basically, this cost I better, conforms to the auto cost type design principle where the mode represents the interpretation of the value and the metric means the semantics. So next step is.

K

We also clarify the property negotiation process in the for the past factor in 0-5, público is handled by the unified property back extension, basically in individual curry and in 0-6 we have merged the two responses into a single one.

K

So, but we only left one property to be curried and it is encoded as a cosmetic and since there are seven we have introduced the following mechanisms: first, available properties are lost in the IRD entry capability and the client must submit selecting the properties in the curry and basically, this process mimics the negotiation process of cost types. So in the next page,.

K

We give an example of what the changes look like. So, on the left hand, side in the article abilities we have a new capability, called any properties and it is defined as an array of party names for the any domain ending the and right hand side. We give we add a new field to the.

K

Filter that is submitted by a client which is also named any properties, and it is an array of selective properties by the clients. So, let's page, please.

K

We have also introduced a new property should, in the row 7. Basically it is. It is called persistent entity. It is an array of any identifiers that are persisting the scope of an out of server. So when it is not present, we, it is truly as an empty list. So the motivation for this new product is allows clients to call refer.

K

The information related to energy is discovered by pass factor and in enables application, such as in network cash panning and it, and, in contrast, any which is dynamically generated and specific to the correct procedure, entities or can be used to create related properties in another unified problem map or some other auto resources.

K

So in the next page, please.

K

And in this page we give an example of how the persistent entity works. First, the the united energy must be exported by another unit by problem map. For example, we use the domain called HTTP proxy here. This process is not formally registered, but so we only use it for this example and in the unified property map.

K

We claim that this unifier protein map can provide property is called price for this HT proxy, where the price means, for example, dollars per gigabyte stored on the proxy, and then this resource should be used by a path vector resource to indicate that this entity is exported from this from the used from the property map in here in the users communities and then in the any properties, the server came along persistent entities as available property, so in the next page.

K

D

K

Page, we give an example of what the persistent intense returns, so it returns an entity identifier of the HTTP proxy domain. So here we basically still using the IP address for this entity identifier, but it could be, for example, a domain name awesome or a host name. So basically, this is a new newly introduced used property. So in the next page.

C

K

In the next page, sorry I think the connection is temporarily lost, so so that if this page gives an overview of the clarification text we have made to the draft in the 0-7 we the first range in the three sections. You should actually overview and motivation basically is build around the requirement of the pass vector design and how the extension interesting requirements and also the co flow use case and the since the in the latest revision.

K

We try to clarify the importance of transportation in two aspects, so, first it can be used to support new usage scenarios and also it provides fundamentally different. Information between the correlation of natural path and the base protocol provides preferences of literal past I. Think this better clarifies the necessity of the pass vector extension and, in the overview section, we give a top-down exploration of the design, space and design TCM justifications.

K

So here are three most critical issues. For example, why do we encode the information in a single message? Basically, it is for like server, skillopedia or other customers, and we also give some other details and also we.

C

Cai-You there I.

A

Think we lost him kai.

G

A

Mantra, Kai yeah.

A

So since the SIA I mean he would, he was mostly done. I'm looking at Wawa, not mostly.

J

Yeah he sounds like he can hear us hello. There.

A

Is back right, okay, but kaya. I also would like to remind you to speed up a little because we have a couple more presentations today. So maybe you can go over the remaining slides, a bit quicker, also yeah.

K

C

K

C

K

Okay, so in the next page we basically fix a issue which is in the latest sec trapped. It uses a different mechanism, as we we mentioned in the past back to extension. So, basically, we fix the issue with a new concept called power resource ID. So this power source on is also used in 0-7 version, but not defined individually. So, basically, we need some clarification.

K

So in the next slide we we we describe the two usages of the past resource I power resource ID, so it is used in the pass filter with response to indicate the resource redundancy and in the sec update stream to the multiplex updates for the two resources we turn by the pass factor.

K

So in the next two slides, we basically give example of the power resource ID. So maybe we can just skip the examples and.

K

So that in the next time we basically talk about the compatibility issue is cost can endure in the there are several version. The integration with cost Canada slept as discussion. Basically, the conclusion is in the you left as a future requirement.

K

So in the latest revision we looked back to a question and and I think, even with the current design, the cost Canada extension can be used directing with the pass back to extinction when it's one requirement that the same, if the same extra natural element in different time intervals, ways different properties must be treated as different areas and if, if also reinforces this requirement.

C

Okay looks like we lost you again, because we can't hear you.

C

Richard, if you are aware of this, can you just.

A

A

Okay, we went to the conclusion slide. Can you can you try one more time.

A

Well, it doesn't seem to work, but I mean we can see the conclusions here and maybe that's a good time to open the discussion or maybe ask the other co-authors who are here so I see what I read from this is this is you did a lot of work and you move this forward and you don't have any outstanding issues anymore, correct Richard. There.

I

Are no any yeah really and so do only one thing is falling and I think it will. Only change which is not make is for the final version of SSC, which should we just upload a few days ago, I'm willing to the name from clan ID because stop stream ID. So it's an.

G

Editorial thing: that's yes, there is about.

A

Nine content.

I

Writing! That's only change. Okay,.

A

But actually I would say for that we don't even need a new version. We do that after the working class caught right, if there's just a name change in SSC, just as long as you have it on your you don't forget about it.

A

So the other question is we agreed that we would move these two together path, vector and unified properties and but Richard you're more, the expert you shall we go for a working class, call on this one and say it's done and then wait for unified properties or shall we wait? I mean.

I

Yeah I think we promise you to wait until we see the unified property and then we can make a decision to see. We was homeless.

A

I

Mean during today's session we cannot.

A

Go into the okay: let's postpone it for a couple of minutes: okay, right.

I

So the only other comment which I want to make is right now, I think I and I. We had a quite a long discussion, I think like a two days ago and because Ryan out of Cass factor, there's one use case about I. Think I also talked to something as well and because there's a one major yields kick.

I

Would you right already instead of getting fully deployed by the DoD by like for bear, lodge I will see our elk, you see and there's some like three years cases and time blog and several ones and which potentially I think he talked about integration with a cost calendar. That's probably related, so I don't recommend at all. We make a change. This document really integrated, but I do recommend that the ulcers and including hikes I mean out there and also start being cost, come to take a look see if we would.

I

Potentially, we might even like a low-hanging fruit. Just email address very quickly, I believe they can be addressing the same thing. I agree.

A

With all that, but maybe they can also do that doing a working blast call yes, I believe so I'm just trying to speed up the process, not for every small issue yeah. So maybe when we start the working group last call officially on the list. Once we do that, you can remind the other cost calendar, co-author authors and stuff to check for these kinds of things. So.

C

Just to make sure because cost calendar has been through the wgl see so there's no impact on.

I

Cost calendar itself corrected no supposedly not, and basically cost cannot be stabilized at Hayes, and the only use case is I was a special tortured by a couple essentially CERN and BOE Department energy people see that you check it see if it can be really use or not. Okay sounds good, so.

C

Then we come back to W DLC after.

A

Impact properties, correct yeah: we do that discussion afterwards, but it generally to summarize also for the notes. Our understanding now is that all design decisions and technical issues have been resolved. Lots of work only super minor editorial issues and some clarifications with other draft or authors that they can make them work in class cause it's done from our perspective. But let's wait because it depends on unified properties how we move forward. Okay, cool.

C

Then Richard Europe next.

L

M

I

I used I hope that it's a real work so therefore, I can really speak and without being corrupted. Ok, it's performance, magic, finest light, I think upload. A couple I upload couple versions.

A

No way, Richard your uploaded versions on short notice before the meeting and I think it happened. I.

I

A

I

You can see that is like a final version from.

F

I

L

I

Down yeah final.

L

I

Swipes yeah I give a title change so.

C

If you keep the title, the same it'll just keep on upload, you know makes.

I

C

So that's why I suggested putting the version number in your sights yeah.

I

Ok, so let me get it started, I'll talk about auto performance and metrics, and it's version 7 and here is least of our callers, and we indeed actually have quite some extensive, extensive discussions in the last couple of days and I. Think, even today, we're gonna talk to some experts from IPPF I. Believe, let's see, let me talk about ok, and here is outline of changes we made in this file in his most recent abortion.

I

Oh ok, so, and for today, and mostly I'll talk about out, is we have made from zero six two zero, seven and I do want to talk about one remaining issue: I cannot really remain. Issues means mainly a single remaining issue which we want to engage the homework you group we're four can discuss. Hopefully we can make a final decision. I. Never talk about plan for next at depends on a discussion on the working group discussion item. Okay, should I come to.

A

I

Okay, you can try really good.

H

I can and I like this.

I

Yeah, that's sort like I cannot I, don't believe so. Okay, your bus, so and here is the really overview over the updates we have made from watching six who version seven. Oh, it works now. Oh.

I

So, basically- and we have made two types of changes and type number one- is we indeed make quite a lot of structural changes, not really content, because with your dad, this is better at the clarified idea and in particularly for people who can really read. I know it's like be small and a media strategy. It's actually true from there and left hand. Side is portion six right hand, side a section. It's about seven, let you explosion. So what we did was essentially the operations, the challenges and so on.

I

We put our very beginning as up front and beginning to talk about all the change and so on, and they could rely less, probably too much diversion and follow reader. So therefore, fourth annual worship, we move it into the new version in your section called section four and we rename the called Operation Torch operational considerations. Surfer. That's one major change from here and to here and the second immediate structural change.

I

Is we indeed make a lot of structure changes on section two, which is essentially bulk apart main technical part of these drafts and we're mostly trying to define the performance metrics? And you know you know in a different structure so which actually the main focus of today's discussion? Of course, we also made a lot of small tax changes as well and which I will not want to go to the details. Ok, hangar. The next slide. Please. So here is.

I

Let me talk about why we make it a structure changes for from watching from version six to version 7 and left hand. Side is the basic structure for every performance metric. We define for each auto performance metric assume. You can see that with if I metric, name, description and method of measurement or calculation and units and marianum points with potential measurement domains and timing and use and applications.

I

So basically, this I be mostly it comes from template of the I ppm and but on the other hand, and if you look at essentially based protocol of auto, which is RFC seventy to eighty five- and here is the cost.

I

Metrics registry part a section part of that based on protocol, and it says that if you define and this Auto cosmetic registry- and you must really provide the following information, number one was identifier and number two and wasn't innocent man takes, and you can semantics really talk about, basically without being utilized and document, and lower value is better and what's the measurement unit and so on, and then to talk about security considerations.

I

Let's be structure of alcohol, essential structure and here is a structure of the I ppm and therefore actual ulcers have quite a lot of discussions. Even we decide that's essential day. They are really out who performs matrix. We promptly even totally restructured, really follow the also seventy to eighty five structure, so next slides peace. So this is essentially the new structure of definition for you to perform symmetric with you find it for auto and from version six to version seven.

I

Here's version six just still show here over there and here's a new one for watch and said you can see that there are several changes number one years and follow all the one and we don't have example. Like symmetric identifiers- and here we add a new, a new section- called a metric identify essential to really to be consistent because we really require a identify and then what do we do?

I

Is we talk about in Tennessee men text so therefore, basically, essentially mobile the description and and unit of representation and more including Tennessee semantics, because that turns out to be really intensive antics and then, of course, we move up the use and applications and then what we did it was. We really move a lot essential, other logs and candido's into Central American solutions.

I

We add all those divas over there, so a little bit of restructure repealed and that's essentially, the combination over the past also structure and verses, a senior I ppm, and we were pretty happy about it. But of course we're more than happy. It really take any.

D

Comment or suggestion on this.

I

Part next slide please. So basically here is letting you really go over a little bit about the performance metrics we finalized and we define in this document, and here you can see that essentially can be grouped into a circle. One two, three four and four categories and category on is mostly about the delays and one lady L, a round-trip delay and packet delay variations sometime in the called the cheater. So therefore, if you have feel strongly that we could call the cheater, we recall the GTR is a moment and we'll follow I ppm.

I

We call a buckle deliberation and we have hub count and you have packet loss and you throughput the maximum user abandon where's. The rest of your banner is essentially it can sail. They essentially classified into four types, and here is: let me really clarify. On the month. We really clarified it. We talked about it's kind of essential. It delays, it's a single decent number, conforming to section 605 RFC. Seventy to eighty nine I believe we were using seventy one, fifty nine and I believe now, as 8250 no I has to replace seventy-five 59.

I

So therefore, we change continual version, and we probably we added a must be in non-negative number and unit must be millisecond, for example, and Hopkin is easy in the way we indicate the number must be integer and negative, and packet loss, essentially its percentage and rather represent, and someone if you know the one change is now we you to see the person thing, it must be integer, we remove it, because, essentially we look at some examples. We last no integers will be not good enough. So therefore, basically changing to be must be.

I

Non-Negative is longer integral over there half can. Of course, you must be integer and bandwidth again. Between two 8259 and number must be non-negative and unit is a megabits per second, and I do want to mention a couple issues and which I think if a people want to comment and will be more than welcome to to give a comment. For example, one suggestion these cars, among all authors, is we don't adopt even truly and but suggested is, for example, can we use a number?

I

So basically, for example, if you go to a boy actuallyI, even surely there's a number two: how to cut an eight knot and number, for example, and imagine, for example, be widest useful, for the use case is one way deal a round-trip time and so on often time you can see. If you look disconnected what's the delay, it's a really giant number, it makes them a possible value or whatever to truly whatever 100 power and so on, or maybe even toll. It should really infinity suppose it is, it is, hunt disconnected.

I

So some also feel very strongly that we should really say essentially it's a member to be not a number essentially called positive, infinity or some others. No, no, no, that's too complex and as a lot of passing, complex and so on so called a kernel version, and- and we did not really take this a suggestion, but if you feel strongly about people from I ppm and if you're strongly or whatever and probably won't, we won't hear your opinion and second, the second accurate thing also debated quite a lot or attributes castle. Our cursor is right.

I

Now, essentially, is we pick a fixed unit, for example, here it will pick a millisecond here we pick a megabits per second, they are, they might be slightly not a future-proof, for example. Eventually, for example, the base van der will no longer be a megabits per second and up for example. Eventually, it goes like every seems like it beats per. Second, even till the numbers become always obvious.

I

Very lot, for example, I mean a more like 1000, so the suggestion is, why don't you focus on for making like a megabits per second and kilobits per second K bits per second key VP per second, and so on eventually be computer, for maybe you easy to read, and so on and latency also become millisecond and second and so on, and the eventual decision is no, let's pick a single while, and it's actually always that single number and you need to be defined by the protocol and I will define a pilot by it by pile of little payload itself.

I

So those are two suggestions and we do not I take I think there are minor, but with you or more than happy to take any working group opinions should I wait here for comment yeah. No, we.

A

Can wait, I mean I, see also Ellis in the room. Maybe here's some comments he wants to make or other people I. Just for my perspective is cherries and we wanted to get clarification from the experts in the ITF that the metrics we are proposing here are not controversial right. Yes, and we have maybe just to reiterate why they could be controversial and Alto is not supposed to be used on, and you know on a on your short interval, but only periodically. So all the values you're proposing here would be averages and something alright.

A

So the semantics might not be hundred percent clear and that's why we want a clarification. But if the IDF experts agree that this is something we can do and we are, it makes sense- I mean. Surely these values are useful right. There are many use cases where you could use these where you, where will be very beneficial to applications if they could get these values via Alto I mean from the top of my head. I can imagine 20 use cases. Why would be useful?

A

Yes, so for us, it's more another process, the expertise issue or I ATF general guideline issue: okay, yeah, okay, great, come right as a chair, speaking of the chair.

N

Hi I'm Mel Morton I'm, not a chair for AI ppm, but I'm, trying to catch up with this quickly. I got a junk mail. This morning, yeah.

D

N

Like that didn't manage to show up at the right time great, alright, let me first ask: are these active measurements and that they generate synthetic traffic and.

I

They will talk about that. Basically, just a moment. A semantics would be define essentially be defined conveyed by the because also essentially can be classified customized and essential. I want go to go to that part of details and pork shoulder and the semantics could be essentially our show, the actual, for example, the average number and could come from the arrival results or could come, for example, packet appear and to measure of examples, report and so on, and it could come essentially for exam.

I

Some can marry measurements so may not get me may be active or may be passive. Okay,.

N

So all the all the I ppm metric references are strictly you know, according to RFC 77 99, they're, active methods of measurement, yeah yeah, and because the metric definitions take advantage of that. We also call them active, metrics correct when you talk about doing something else like passive or the hybrid categories that we defined.

N

You know, that's that's going to be it's going to be a little different for the metric definitions, then so, there's you know some forms of mismatch emerging in the you know. In going beyond the active synthetic measurement, traffic can.

I

You hold out yeah I, think for common I'll go to that apart and if they say that you'll see that, for example, if I mean reactive very quickly, it would be, for example, right now. For example, even in one project where were deploying a system is right now, for example, one trend is now you even given it's not activity, I, don't if you called active or passive, for example, right now with a lot of Sdn for examples which is nowadays is you can even do measurements passively by monitor you know traffic?

I

Is you don't really inject because I'm, a probing also or a pure code, not very much to tell even like a right on your capability is therefore I, don't even call active or passive. Basically, we can collect those numbers and try to reserve, for example, I mean reportable, for example, I think that's probably really polar boundary a little bit well,.

N

I think we've got the full text on me described seven.

D

N

Seventy seven ninety-nine so I think that's what I'd suggest that you take a look at that. Sherlyn that'll help the absolute like determining the applicability of the IP.

I

Pm metrics that we've written yes now you yeah, I'm, gonna, sure I ppm in a slice and week, I'm gonna, discuss with you guys and how we want to build on here. But we don't want what you'll say, but this moment and I think yeah. Okay. So you.

N

Were saying the other thing I'm I'm interested in is that I see you've got a registry going here. Yes, we do and we're also about to finish up for registry yeah, yeah yeah, the IP PME, an issue or an issue or educational content, but also the all the registry design, and you know this took way too long, but that means that it had lots and lots of input and it supposedly covers both active and passive measurements derived in different parts of the IETF.

N

People are already using it and I mean as we approach the last call. That was some of the good surprises. You know. People are out there, basically using the registry and had some feedback on that. I'll show you I'm not sure.

K

N

You're imagine there's ways to put these things together, but you know.

D

Obviously we're going to go ahead with.

N

The registry that we planned and and and it's it's got a lots of detail in there on some active massive networks. Definitely so.

A

Sorry Mary, just for my understanding you can you can go. My understanding is that our ultra performance, metrics document will refer to metrics. You have been defining so the semantics should be then Chris who cleared this semantic I.

I

Want to go data part.

N

Changes well, I, don't.

I

Want to show that also do so.

N

One one wrinkle on that young is that the IP performance metrics of specifications deliberately left parameters, input factors available and by changing those input factors you can get very different measurements. In the end, the purpose of the registry in our mine, the performance metrics registry, was to narrow those down a bit where we could. There are certain runtime parameters that you shouldn't nail down like the source address and destination address, obviously, but the but others check your path. Example: periodical publish all those.

I

N

Are the kinds of things that you know in order to get comparable results? That's our goal. We nailed a lot of the flexibility of the original IP PM metrics down, so it really sounds as though in some sense, this should be referring to metrics that go into the registry as opposed to the broad IP PM definitions. If you want to get comparable results,.

A

M

A

Long time we were not sure how to move this ahead, so we thank you for.

I

Being here not for me, this was very useful. Yes, absolutely right! I'm gonna go that pop. You won't give a quick comment if I go that part.

F

Don't understand how this like. Finally, it would work because they young you said this is really useful metrics, but those metrics are only useful if you actually have data that is current right like ideally, you would need to know what is my jitter right now right, because these values can change. So what's the what's the intend or the idea here about like when you would actually measure those data or like how static other data? How often is oh.

I

Okay, I think the short answer is appalling. Of course, would you find a product short? Also? Is the you can use kiss its people initially Retriever, for example, from custom, app or any coupon car service and then subscribe, for example, using see sort of what can get pushed to whatever interval? He really is subscribe to it, for example, it could be pushed I know how much we'll get a coach back from Ikea. Basically, the potential it could be pushes by essentially like a SSC, essentially probably can in some kind of real time.

I

Of course, we could really also for exactly the I believe a for example. The I would I show you're like stop all Sprint and AT&T show their performance numbers typically Abbott. Every 15 minutes, for example,.

F

My throughput will very correct.

I

Right for some for examples, report available banner is probably called week on, but a problem area latency promise I'm like a propagation delay problem object unless, for example, there's some kind of, for example, link failure and you do rerouting so what other case you probably pushed.

F

Not like once, we.

I

Are not applying for that because I'm, if you really do that it, we cannot prevent it by the intention is much a longer time scale. So.

F

I think actually, all of these use- all of these metrics are useful in such a timeframe like 50 minutes or more and and for and more important is that you actually make clear what you mean in the draft sure.

I

F

I always automatic, is you mean you do a fresh measurement and that's one of your report, but that's not what you're doing so. Instead of maybe you should want even to rename some of the metrics, because the names are used in this measurement world this way and and what you were add up in some cases you might actually get setec. Well, yes right, you measure your hop count once and then you know the topology and you will never ever we ever like this or somebody configures at once, and it this there right. Yeah.

A

So hop count is not critical.

F

Values, like might change more dynamically. If you ask for these various you just don't know what you have to expect and if you don't know what to expect, if you don't know how people regenerated these values, they are not useful at all, because then you don't know how fresh the information is and what to do with it right and so about some of these metrics. Even so, they would be super useful.

F

If you would have life information, they might be completely not useful if you don't have the life information, so it's actually little bit dangerous to put the minute, but.

A

Have we Oliver Mia? We all agree, we don't want. We don't want Alto for license. No.

D

A

That's out of the question, so the question is then: are there value? Are their use cases where, if it's not life information- and you know how to person sure when this was measured, if it's still useful for.

E

Example, your mom.

F

F

E

F

What what you also could do is maybe I'm reduced to a minimum set right now and then you know and other metrics later on. When you actually see there's a need for right sure.

I

Sure yeah we do have several years cases and well yeah. We can talk about that. I want say which one do you mean? What can we just I.

C

Think this discussion has focused on the metrics. Maybe you go to the metrics like yeah.

I

So we can, you just go I, think I can I can tell where to stop. So basically you can skip this part. Well. Actually, yes, so I. Could. This is a really problem. One important slide so basically is left hand. Side is the central metrics and we have defined and right hand. Side is a back pay pilot.

I

That's we did this caution issue, which we really want discussed is the consistency and reusability and because, basically, what we want is reuse as much as possible and because in as much possible between auto and also a VPN and for people who don't know, for example, the FE p.m. and yet essentially is released of the kind of metrics you find about a ppm, for example, wrongful delay, and you know: I command interminable IP travel UDP for UDP, using periodical traffic and here's a yellow exam pass, save and you learned by PCP and so on.

I

So essentially, all define being. You know, I ppm next sites, please so basically, what we want is this moment is when we need make a decision, and it's really essentially final of your final wording and basically of water. We understand a synergy. Discussion was always the call so I know we want really open discussion with I. Think Y is falling, we've got the IP p.m.

I

metrics, and mostly alpha for infrastructure measurement, man and management and take the underlying network technology be aware, for example, the PCP is UDP and just kind of central technology, and so on an automatic is tended to be more apple application, for example, it should be really technology transparently, you don't here's using IP if you MPLS, and maybe using, for example, and okay. Essentially, what about is user one, somehow slightly technology mutual can alike at high-level abstractions can't go next.

I

So basically, I, for example, some kind of like inspiration we talked about and sony is for example. Here is a data would be posted by sprint and they talk about all kind of Papa metrics. For example, talk about. They are essential at the post of the average latency only on public website. They would talk about for exam, whatever they didn't say.

I

35:56 mean the second units month and basically, if you go from where to hear oftentimes to this very like a core screen and still very useful, and they talk about- was a typical loss rate and so on and next size, please so yes.

F

Because you kind of set this already, you wants protocol independent, but it can depend strongly what the results will be depending on what protocol used to measure languege.

I

Actually, which is a hold on one second, I'll, show you what a slightly also discussion will look like just just like a one minute. Oh okay! Oh please.

N

L Morton their IP performance metrics are aware that there's IP connectivity- yes, that's it! Okay, yes, so I kind of disagree with your slides.

I

And the categorization you're making this sure. Okay, so basically you wanna say it's more generic I peel! That's totally fine! So basically, here, of course that's also spring resolved and you and all the people see what kind of SOA you can really figure out. Somehow, if you want to pick up here, pick a replicate and so on. Somehow you can use, and so on next slide. Please so here is I'm just showing wrapped exam here is potential numbers posted by AT&T I believe, and this is really from AT&T number.

I

You can see, for example, lately and say thirty point.

D

Nine milliseconds.

I

And so on, and here's for every peers and so on essential. There are large school measurements and opposed by Union. So so, basically, what do we one? Is this calico really hello, hello?

I

If you really compare it really, probably into, for example, for example, Europe and versus, for example, North, America and so on? Basically, they can really rather stable and so on next slide, please well. Those are used which we were trying to clarify.

I

So basically, what we one is falling and I'll be very quickly and we can open the discussion. So what do we want many levels? One is we base ultimatrix directory and that's it, which is you the user name use all the, for example, exactly same name is the one simply we serve as some kind like a transport. Would you like? We did for CDI and there's some it's very another one is we don't we don't want to do it, canceled, abstraction and so on and, for example, based on it, for example, Raynald can.

I

I ppm depend on, for example, t UDP, TCP and so on, and if you will forward trend eventually now a meter training along networking to become application where, for example, basically prove or not, you see, for example, a TCP UDP and probably become. For example, a central like an open floor was a match condition and for this match condition and isolation. See you for that map condition. That is the eventually becomes a lot more like generic, like so and so on.

I

So for that says, it's more applicable where for that case, because it's an insane idea as much as possible, I'm symmetric unit, for example, you've already here was different. For example, ok, I, remember it listens it using like a second, our community, second, and so on or pramana we're happy to discuss about to be considering as much as possible. Next I think I'm almost done with this part, so basic clay worth a next place. So what do we want?

I

Essentially moving forward is the following and one possibility if people really feel less, ok and, of course, I think it would be. We talked about it. Yes, the discussion with the world of people as well like a foreign and so on is one way is we make? Applications depends on and basically for exam. Often time is for network. You probably don't really care about TCP you.

I

You can often time you care about what kind of cream and you really get, for example, maybe really like the networking really doesn't get significant TCP UDP perform at me, really gives you only on. For example, this is my gold customer, this, my super customer and so on so often time you were captive in Athens, is different, jitter and so on. So what we probably little magic and your head of hospice.

I

Oh your application one, and what kind of category really have belong and cut you are and then essentially the high level performs, whoever can be the pack and then piece ignition, reflecting like the categories label, the categories and depending on applicants, propagation DNA's, for example. It doesn't depend on any applicator here, oh, oh, maybe, for example, I think, that's a one potential we move forward. Can they go to last leg? I! Think we're right.

I

So basically I'm we're on discussed ways all these I ppm and make a final decision, and essentially it would be what can a matter that we really export I think a mere suggestion. We probably should pick a minimal set I know we all really need to finalize all the timing essentially finalized, essentially the naming ideas of Auto matrix, and then we want to really finish this thing.

A

Okay, so yeah more comments, peace on the mic, but as a chair I see some skepticism in the room and one suggestion that Millia made was focused on a minimal set of metrics to move this ahead and since we are lagging behind with our milestones, that might be an option because the other option I see is having lengthy discussions that go on and on and on right.

I

Right: well, we basically a one with Lenny, just a final come with whatever that minimum subset is I'm, not sure, but right, let me animus upset. One thing is very simple: is we use a very connect definition? Basically, I think would be probably some kind of agreement between the auto server and other client, so we use where connect latency and all stuff. Basically, circular users will be essentially a very likely be authorized already. For example, syndicate sort of for information might be that's a really simple way. We define connector one.

I

We don't really go down the rabbit hole of defining. What's the right green for me to talk about matrix, P eventual drain, Alameda trend, asymmetric GUI become essentially very flexible, based on to match action for from open flow to p4 becomes so fine green and eventually, probably become less, probably very difficult classical down. We just define some very generic classes. Okay, here's current best effort- and here are the propagation latency- and here essentially those kind of very simple things. Therefore, we don't really go down.

I

It's kind of like a fine green application distinction, I think that's probably easily out and Sue probably will be very, very useful in terms of high-level thing and all that other thing is, we probably can try to discuss a little bit which matrix we probably don't, who is too dynamic to hard handle. We could probably can just trim down little bit. I think those that those are the two things I I can try to offer.

F

Yeah I've been trying to think a little bit more detail. How that would actually work right, like you said, 50 minutes, whatever that's the Rena value, but, like you know, would it actually mean that I kind of sent measurement rubric every 15 minutes to kind of? Basically, every input like you know, if requested I could send this to every endpoint on the internet right.

A

No, you don't do if you don't.

F

Do it right right, say I.

A

Have to maybe one reminder also for L important to know. The goal of the alto protocol is to provide better than random pure.

D

A

That was, or that was stress intentionally because we don't want to give precise information just better than random. So.

H

Just to keep in mind that this is what we are doing here,.

A

So and we are not giving precise information, that was never the call also of the original protocol, it's better than random. So.

F

A

Nouri comrade is it still random? What you're saying.

F

You have a source where you can access some data and you don't have to do active measurements. That's kind of useful but given like, as I said, given that the the one who consumes the data then never really knows how this data was generated. It's less useful because sure.

A

F

Really know what the data means it doesn't help you. It.

A

Could still be better than random yeah.

F

But then, okay, let me propose something: I think it would probably more sense. For example, to just say this is probably a high or a low loss link right on this level, rather than trying to give any kind of number sure.

I

For example, right now, the one matter, what you don't like: a cost like a rocking cost, so paste you're, saying somehow, because we do actually introduce one thing called. Essentially it's numerical versus all the notes so you're, seeing maybe wow. It really is, for example, you're just a random rank, basically cosmetic is essentially latency, but somehow mostly focus on ordinal part, which is like this is a higher latency, lower, latency and so on, as discrete vice a little bit often time. One way you want really reduce of damping.

I

You want to template that in process that you'd essentially do some kind of categorization essential to a different like a rounding levels. That probably can also help.

N

L Morton, here's a suggestion. Delay is more than two times the you know: the airline City distance. You know when you measure it and that's you know, that's something that that users might want to know 2 times 3 times 5 times. Something sure. That's sure. That's that's something that you know can be briefly assessed on the network and it's mostly about routing it's not about property.

A

And by the way, is in the spirit of the original, although idea and not give precise information, but something that is useful for the application, but still also doesn't disclose too much internals of the ISP right just the hour. Ok,.

N

That's good, that's good! Keep.

A

Saying that you know.

N

And- and the other thing is, is to you: don't need to go to the application layer to say things like this. You know, I saw you moving up the stack there for a moment. Don't bother just you know, make some measurements and and do the best you can with it in broad categories. I think that's the way to go.

A

Yeah, ok, let's really here, thank you, I'm, sorry I think we would like to wrap it up or would you want to have a comment or you yeah go ahead.

O

One of the very goals of Sabine here, one of the very goals of outside decision-making, which end point to select, and so one option that was discussed is also instead of displaying very accurate measurements. If, if this is too complicated or not available, we can also advertise a score like between 1 and 20. If you want to compare.

I

B

That was I, don't.

I

Know why I essentially give something a score you to normalize in some way, and anyway,.

A

I think we sum this up from a chair perspective. This was a very good discussion. I think it brought us forward, but the process is pretty clear. This one needs another round. Yes,.

D

A

This across, let's that's, we stopped at that. Richard I, think you you, you and the co-authors need to yeah. Yes,.

D

Our mailing, this.

A

Discussion keep Mira and Ellen the loop that we can come to an agreement, and then you need to try. Yeah reiterate as well sounds great right.

C

Yeah yeah I think there is some discussion on the list leader for this Sabine I think you're up next, but your time has been massively eaten into so my suggestion would be to don't worry about the changes between the drafts. Just focus on you know, if you think it's it's ready to send out the working group or not.

O

Yes, definitely I won't spend so much time about the tips they are pretty minor, so well, I've focused actually on the first slide, which is the status on the draft where we stand so since last year, so that was the first is G review which to discuss is the one was immediately solved.

O

It was about the IPR, the other one, who has some discuss about the design that was solved for the previous IETF, so we can go to next slide, so the current status is active, so we discussed a couple of things in Prague, and so this is the timeline, so that was the last working group last call that ended this last Sunday. So next slide.

O

So since Prague there has been an update that addresses Miria onions recommendation about clarifying the the the the issue about JSON encoding, and that was also a review sent by VII. So next slide, please. So regarding the encoding format, the issue was that the base protocol refers to JSON an RFC 8259 that obsoletes sorry. The alto calendar refers to an RFC about JSON format that obsoletes the one that was used by the auto based protocol.

O

So after discussion we did find that there is no extension of Alto that really is tied to using utf-16 r32 for encoding. So basically, so this point is addressed in the new version in section 7 as next slide.

O

So this in this section, we explained that we now use the new JSON format encoding utf-8, because this allows into a better interoperation and so I modified the text there and I send email to Adam Roche to have his feedback so about the formulation. So next slide please so then we had a review sent by Vijay. So there was a thorough edit session that was necessary, especially to clarify the text and short and long sentence and clarify some things and also a question about the content length field in HTTP examples.

O

So because in the draft there are elucidation such as like.

O

Yes, such as this one, okay- and it was not clear in the text whether the content length field was accounting for such elucidation. There a next slide please. So the answer is yes. Actually this file does account for the like real number of corresponding to these elucidation. So it explains that it explains this.

O

There was also a rephrasing where to clarify some text in a rule and which, where I added a must- and so next slide, please so there is a second working group last call as I say that just and eat. So with the chairs comment, saying that been indicated, a couple of edge cases, he put a discuss item that was solved in the version 10 and so, and there was also that IPR issue that was solved so I also received some working group feedback on suggesting on spelling consistency, so I tried to change.

O

I will try to chase them, but well I. Think well. My point of view. The the the other draft is is ready, so.

C

So from the working group point of view, there was a couple of folks that reviewed at Cayenne, Jensen I believe I'll go through one more review and then we push it out of the working group towards iesg, and that should happen. Probably in the couple weeks or maybe the week following the end of the ITF.

A

C

Done yeah, so if anybody has anything post now or we will push it out for sure in the next two weeks.

M

All right, thank you.

P

Okay, Jennsen unify properties unify property. Yes,.

P

Look at better, okay, thank you, syncope, so antonsen I talk about the updater, but you probably it's a major, auto extensions, and because because the current leader, a path vector and I've seen some other document, they become working group document at depend down. Indeed, and so that's that, please think, and in the last few miles, actually we our cars and some other working members have a lot of discussion about how much better clarifies a in fact property tax.

P

So actually, in this revision, weight makes a major a piece of the talking additions and better clarify the requirements and the concepts and the design scope. But actually the specification is not too much modify. We just make some slight modification.

P

Yes, so in this revision we clarifies a unified property extension actually for the client side. The client has three major requirements. The first white client may want to obtain the properties of the generic entity, which is a missing part in the current auto framework. For example, there are some other generic entities like the not suggests an endpoint like the PID is a AIS and concrete, and in the past factor we introduced a track.

P

Metamer elements so object may have a properties, a clan want, a curry and certain crime rates that encourage these properties of the entities. We don't want to introduce any message, redundancy and serve I'd, say for some entity properties.

P

The server me to provide a full might and the clan can just get format. So it's very straightforward solution. In a very early remission of this document we already introduced. It's just introduced a cat can more probably map service to solve this. So you know last rise. We will I will clarify so for the first requirement.

P

What we need to do for the tea time so actually for the first requirement. The aloe plant should help to answer to the following questions. First line is what entity is kind of auto plant Curie's and sacrifice for each entity? What probably is called? Who can cure it? So that's the two major common, the questions for the clamps, so next rage yeah.

P

So, first, for the first time, questions our basic idea is because, in our previous document revision our designs, it's really also server just announced a set of tabs and it's tab referred to the entities of the steam semantics and the.

P

Specified ID format, for example, the ipv4 types will means the entities, it's an ipv4 address of Andy point and they have the identifier with the ipv4 address format and we consider an entity's value. If it's identifier is valid so say, each type will define a complete and did it side and which consider all the valid, and it is for so by just a pile on Stern type. The auto client can know so it can Korean.

P

It is seen in the other entities in this complete on either side, which we announced so but also next back page yeah, but there's a issue that, because not all the valley- and it is a meaningful. So what's the meaningful became an example, so the PID defined by the auto femur, a PID- is only meaningful in your natural mind. So that's a lot of widely the PID names, but the other server only to find a properties for a subset of the this complete hiding inside so which is a associated to some metal knives.

P

So for a given when we observed that for a given entity type and how to information resource to only contains a subset of the complete entity side so which we refer, this subset as a resource page, we can hit domain so that our solution change a to. We allow the server and also said how resource pacific any domain most by speak for the client. Okay, really- and it is so nice page.

P

So, however, the using the just using the entity ID we defined previously to identify these entities may be I'm wicked, so the because the clans, our server, need to distinguish which and digital main entity belongs to so, for example, we have to PIDs ones in the neural map and natural map to define two different apparatus.

P

It means they may have the same properties, yes and the ass number, but in the different natural map. Actually they has a different as number of errors, so to teach distinguish them. We use the undetermined names column, the domain specific entity ID to identify each entities.

P

So here, for example, the the ended domain names can be the narrow map, one dot PID. They are there're Mutulu, PID and the PID whines, just a domain-specific and edid, and we use the nano map one dot, V addict on PID 1, but in Phi H, the PID in the neural map, one so nest page.

P

So we revise the same question this way. We don't need to answer each entities. What properties can carry the other Clank and Kure, so we just ask for each resource, pacific and it domain so which properties the auto climbs and curry.

P

So we have the basic idea: is the other services announced a set of resource pacific and determine to appropriate type makng?

P

So next page, but, however, for the entity domain in the resource, passivity.

P

For the movie called the I've di see the mapping it's a protium, probably type appeal may be ambiguities, so we give an example. The narrow man thought I'd refer to the PID properties, the PID. The value of this PID properties may be defined by either the narrow my map 1 ow it just like this, the current property map itself and we need to distinguish them so the u-p design formulated my thing as the I thought, the I to the out of POC.

P

Here the arrow is elderly, the results are I all just these, and so the server can just announced a two different mappings to distinguish who provides the proper team riders, but in the FC 70 to 85 the other framer, the other protocol. Today's it's not a reversed word for the resurrect, so we just use a delta p o2 to written this top view.

P

Some somewhere can remember suppose. Maybe we can just use a few. Is that the that characters? This just a minor improvement? We can. We can the example body. If you we don't have a younger idea about it. We can I, don't know we want to excite about later body. Oh, we can't just move for sure.

P

So, generally the we make the design clarification guys each probably matched its provide, a set of the IDI Tuesday out of Pio mappings and this mapping.

P

She should be announced in day, probably marked and so hairless for cases for the mappings defined, but we only considered so far only three mapping mode a meaningful, so we called the export mode, which means, for example, the natural map Bank and web app ipv4 entity to a PID properties and extend board a PID defined by a network map plan and they the present, probably my resource, extended allocation properties and this PID, how the the property map just define new actually for Christ and Japan. You probably like the s number on this IP address'.

P

So that's a scope of our design, so nice page based on this but not arbitrary, are I taught the I to out of here. Mappings are meaningful, so we need to currently they announce the mappings should be meaningful. So what me up the so we give some example. Meaningful means the TI must be a spotted by a resource RI. So here's the path example so, which means they never. My window is not meaningful because they, the clad to know, what's and defined by the narrower map and in the export working mode they di to peel.

P

My team must be supported by a resource out, so we have another bad example. It's a the network map went out ipv4 to the narrow method. Yes, and it's not meaningful because they try to know what's a s and defined by a narrower, my plan for and appropriate for the ipv4 so to if the meaningful announcement mappings the way, which is introduced some additional ad and a registries in this new document revisions. So for each auto information resource type, the air which is adding indicated by the imager type.

P

With the fan resource page, we can determine as poverty tree and the entity property mapping as for registry, so that's page will join example. We use a narrower map and in example, so for a network map resource, so I can spell the ipv4 ipv6 our PID hit domain types and it covers the IP refer to a PID properties, all the ipv6 to a PID properties. So that's the initial registry and maybe so other in the future. We have them extension, we kind industry to twist so my speed, so our per code.

P

So if you update the IRD from the previous dynasty, a set of a cross product of the set up and hit domains and the set up, probably who sat on mapping passing a smoke mincing and so next pitch suppose the risk request, we pass a schema, we don't have anything, but we clarify the so. The client only just a still request: the cross-product of the entities, a set of entity and the properties, but the server where we know the mapping, which is not enough in the air D.

P

So next page well so for the second requirement is more likely improvement for the current design, because the current exam may introduce some redundancy. We identify two kissably redundancies, wise they request and it is, may be redundant. I'm the return entities may be redundant and the next page will shew example. So for the first case, because the appart property map may announce a natural, my IP refer to the hpid and may also announce another natural map to deliver d.

P

But if a client able to curate the IP address for both days, PID properties defined by the two different network map in the current exam, it had two Curie to innovative entity identifiers. So it's redundant, so we introduced a next page. We introduced the improvement solution. It's a. We allow the problem map just use a the nth item domain type map to a set of the properties which present the equivalent entity problem mappings to Bob Patrick land can just curious single entity. Identifier took at post, two different properties depend by the different mappings yeah.

P

Next page. We see Harry's example. If we, if the we use such a design, we can revise the arity to mal compact representation and nest page to pull the second issue actually very early. We we can already identify its for some type of entities.

P

For example, the ipv4 entities- actually maybe I, owe their IP v4 addresses in same say, IDR may have the same property values such as a as members, but to obtain the s number for all of this happy for entities the current land, the the client have to enumerate all the individual IP addresses in this CIDR. So next page, the unified properties introduced a hierarchy and inheritance approach, which means which means they.

P

We, we can't define an entity block representation for some types of entities, for some of the CIDR can be a until about her potential for the ipv4 addresses and the yoga can define a property inheritance rules on Z and hit block fermentation.

P

Okay, yes, I. Think that's a last standpoint.

D

P

And the next page yeah, but in this revision, with make some clarification, that the property inheritance must not be angry. So.

P

Mixer make some important like the forgiving entity types. If it's sensitive, loving definition, it Japan you're very much be hierarchical and in other words it cannot be partial intersected. So we came up an example like so if we define the.

P

Ee1 didn't block even I'd, say: 192, dot, start out start and one another entity block I say when any two thousand years of Stardust are so they have the different Essen properties.

P

So the class cannot know what I asked and published for the IP address in there intercepting set so the little introduce some application and the property curtains rule. You should avoid this ambiguity, so so next page we make some concluding fold. Our current document revision and.

P

The hope is a clear to better clarify our current GOP design and if it can nitrate, if I can move this document to they work on black car after they missing so.

C

How many people have read version eight of the document.

D

C

C

Given that the authors feel that this work is ready for working group, last call I think just to be conservative it'll be nice. If at least a couple of people can first read version eight post their comments and then we would like to go for a working group last call now it could be that if the comments are deep enough- and you know reasonable enough- they could serve as a working group.

C

Last call, but at least the chairs field right now to try to take a home for working group class called seems a little premature, since nobody has read it so can. Can somebody volunteer to read it.

C

We can start picking victims, awesome, yeah sure. Thank you. We have thank you to volunteers. Thank you. Both.

A

Mike and say yeah yeah.

A

Q

Good trade-off on telefónica. Thank you. Thank.

B

You thank you Barney Bowden, from Sprint, okay,.

C

Thank you, excellent. That would be very, very good, and then we can move ahead depending on the comments yeah, but.

A

Still I think from a chair perspective. This is then ready for being send out. We just want to make sure that you know we get a review done, but you basically resolved all the technical issues correct.

P

A

So, but that brings that back up the question: do we want to do a working class call for path vector because all the dependencies have been resolved or do we want to well? We should then wait on the comments, I guess, because they have a high dependency.

I

Yeah so yeah, so yeah, basically I think the ACA's feedback I should be slowly and it would be right so I think problem right thing to do. Is we don't help you of your comfortable bit about unified properties, because past record does indeed write you more kit. So, therefore, you should just somehow get four dollars. Yes,.

D

I

M

Jiang says that I will read the draft in post comments as well great.

C

Agree good, thank you for all of those that were on here then so.

P

Next one I see sitting I yeah, it's a bit quick and because we've seen is almost ready and just from next page, so in the latest revision for the CDI apps, a document may be. The major change is about the edges and some minor improvements, the change, the naming and the in front of wordings and updates and examples and education to matching the latest GOP document and the SSE document also and make some clarification on the vertical arrow handling and revise a example.

P

So next page is quickly go over the repeated, if you remember they post too many issues for the last in the last night. A meeting at for the absolute document, it's very issue is servicing the city. Ifc is not okay, whether stuff format and the clear alistair format, because in the auto protocols a map service looks like they were from Stouffer might like the network map service and cost map service, but the city I FCI object defined by the FCS on a under its own seats.

P

It's not a cursed or so, but we still keep the message steamer with not change the protocol specification, but we renamed the city FC, I'm a service to the justice city, FC a service in the latest document and for the second issue.

P

It's that we want to allow the curie by the not not not only by the footprint on all the only possibilities it can purify, join the footprint and capillaries, but in the working to have any strong pinning to allow this use case so kindly readjust to now address it and if appeal happens, companies they can propose. Another extension is minister necessary.

P

So in the next page we did such renaming and in the cover they all the the whole document and renamed such my service to the adjuster called a service yeah in the most immediate, haven't a data format. So next page it's also written renamed the new footprint type from the auto natural map to the auto PID, because it's more convincing it means the foot footprint Weiler in such a type.

P

It's a PID defined by another Auto, never map and nest page. It's.

P

Because we, the City FC I, also use it probably mark to better provide the a client can cure curative care police by the footprint. So we just update the example to use the latest to tablet. Map from the I mentioned is from the course product purplish to the Isetta mapping yeah. So next page is the update example a bit in the request and the response. So next page at issues a.

P

Example, the relating example: how to use the increment updates, the assessee document to to ingram abbot the city I've see a resource, but we know notice at the kissing Matt. Merge patch will always includes a whole city, app CI resource. So we made some correction that it's not commonly to use. It is multi-strategy GFCI.

P

So next page it.

P

We also made some clarification on the request process and the error handling part so because, based on the protocol arrows defending in the section fat16 baby, the undefined capita type and weathers is invalid in the erequest and if the plans, since at acrylic server will master returner and invalid felt, whether arrow so for the we mentioned only for the valley today accomplished.

P

If this server does not define any footprint white, the server test, images from the response and nest page its way introduced open discussion that shall we allow the city after a service announced a spotty, see di capelli types and values in its IRT, because in the app see, is on six and I thought eight in the in issue with I own a registry so native NJ, a given example. The only defend the HTTP 1.1 atps 1.1 as the spotted.

P

Providers that may be a new prototype like the ATP to may be really stirred, but the point is old: Auto server may not be advocate to slowly so, which means for this auto server. It will consider this wider, it's invalid, because it's not registered into the energy stream and in its implementation, consider its invalid, but for the new Auto clan, if you consider it's valid, whether it just may be, the auto server doesn't define the footprint for this value, so it was surprised why the server will return the invalid field, wider error to this plan.

P

So if you, if we announce the spotted cavity type and the wailers the we can prevent a grant from sending such wasted requests, so it may be just a increment. It may not be very necessary.

P

The LAN, but we want to introduce a tit of discretion, play and.

P

So basing the class are also think it's this document already so almost down. So now we can go to the working Glasgow after the meeting. Yes,.

A

So and I'm, a co-author of this and I totally agree. So we we, as authors, think this document is very ready and, as chair now changing my head and what we discussed yesterday, also in with the authors, but also previously today's we want to involve author, the CD and I working group to have a final look again theoretical they should. Theoretically, there should be no issue because we're not changing the semantics of the CD and I Jason format, we're just transporting it. You know just to be fair, you have them look.

A

So what I'm thinking currently right now as I, will send an email to this CD and I work in group and tell them guys. We are shortly issuing a working class call and give them a heads up, so they have more than two weeks right and it might seem a little unfair free issue. Working class call right now and just give them two weeks, so maybe I'll send them an email guys that we are finishing this piece. I have a local working group.

A

Last call is coming soon, will inform you to, and maybe also one of them can be reviewed during the working class call. So my guess is we do issue the working class called in two weeks or something and I tell the city and I give this either networking the heads up, and then we have another two weeks for the working class calls. So they have a month to to look at it, and then we ask one of them to one of their participants review it and one of our group and that's my plan. Yeah.

A

The question is Jensen, but we can take this offline if, if we issue a new version in the meantime, because the the last issue you had there at the IR T changes, yeah yeah, all right, we can we can issue a small update in between, but I don't feel strongly about that.

C

Yeah I think I'll leave it to you guys too version or update or not that change is fairly small. But again, as we look for working group here, working group last call here and coordinate with CDN I, please, you know, need participation from the working group list, members to also review this. Assuming we get one from CDN I. We would just need at least one person from auto working group to take a look at it from W GLC perspective.

C

So for the notes, we're going to give two weeks for the draft to percolate here will immediately inform city and I that we are going to have a working group pass call in two weeks, and then we have two weeks for somebody to review it internally in auto, and then we issue a working group last call all right any feedback on this. In that case Danny you're next, and if you can get us a couple minutes that'll be good.

A

I think we're doing great in time. We still have about 20 minutes, so maybe Danny you have ten minutes and then we have another ten minutes for the general discussion.

G

A

Think you're doing great yeah. You will keep a couple.

G

Of his lights, maybe.

A

By the way for those of you, who've missed it, I forgot to mention this. Actually, we had a paper on the a nrw workshop and many of the co-authors are here, and this is very related to Dennis presentation right now and I think it's a good success, also for our working group that we had a pepper there and you know show that the community our work. So congratulations to the authors which I.

G

A

Said actually, this is doing work with.

G

People at Yale, University spring Erickson telefónica, no cam. Okay, I will skip a couple of.

G

Yes, okay. Next, please keep maps X next.

G

Here here we have a high-level overview about the key alto mechanisms and constructions. In particular, Alto provides different mechanisms such as information resource directory, to provide a list of available information resources and to provide an infant consistency. In order to allowed, we specified the a dependency between differing information resources and information at a model to to controller to a string control push an incremental update of the information resources.

G

Alto also introduces a set of instruction modules, such as the network map or Kozma to to provide a network location grouping occurs between then the patch vector instruction to provide a more detailed routing information and using a step net or elements and the fruit broke. The capability map, such as the unified property maps or the fruit peelings capabilities X, please.

G

Basically, we have the overview about the all alto documents in the nest column, including some potential topics and the discussion to be addressed, such as extension from your detectors, attention for in settings on alto survey. In order to try to to summarize the coding and implementation in real use, case scenarios etc. Next next, so, basically, currently many multi-domain use cases are emerging with the development of new technologies such as as the egg and a free.

G

If IG example of such use cases, including the multi domain collaborative data science, multi domain serie function, chaining, flexible, inter domain routing control. In the past few alternative meeting difference draft were submitting summarizing the the experience of developing multi domain application using alt of Hira. We have an incomplete list of the X please so here. Basically, we have a couple of example of premier scientific experiments, the Hadron Collider and Square Kilometre Array such experiment pushed scientific discovery boundaries and rely on workflows that coordinate geographically distributed resources.

G

Next, please- and here we have the moving of the data from the LHC tier 1 tier 0, to t 102. Basically, this same network like weight, multiplied demands. We get the ordinary resources to run multiple different applications or jobs of the key requirement in this type of distributed with workflow is the ability to straight multiply resources across multiple domains for heterogeneous applications? X please.

G

Here we have the multi dimensional function, chaining, where basically, young and doing network services is often required. The earth dweller physical network for showing on a specific order, a network operator has stepped over in the to the customer. The possibility of configuring network service was a specific requirements, his term of resources or performance. Such resources are expected to be available across a different domain with different technologies or administrations. X please, he connects.

G

He will have another potential potential use case, that to expand, define great net capability of FDM from into the main set up to to multi domain setting to provide flexible inter domain routing a traditional internal matter and the protocols are limited at because they provide single parroting and limiting the decline spot choices. So flexible, multi domain routing allows to user to specify ratting action at the provided level, with more flexible matching conditions and more choices on roads and not only currently use at roads, but also available yet and use it ROS.

G

Next, next, full of key design requirement for alto to support new multi-domain use cases. The mother use case required information on properties and capabilities on different networks, resources, including transport sources, storage, resources, processing resources and such cases name need to network astray, multiplayer resources in multiple networks, so that a unified resource, the presentation is a key requirement. A four for multi Dominator information extraction, an existing representation tend to focus on a single domain, a multi domain just case related information can be retracted from multiple network to compute.

G

Entering information as such that support that super aggregation of multiple network into a single built entities are another key key. This antiquorum next, please next to to solve the the the multi domain or to solve the or we use the multi domain extraction to solve the application layer, optimization across multiple networks. Typically previous case examples can be modeling as an optimization problem. F where were where F is an optimization function with to tie off parameters X representing the network parameters and Y representing the application parameters.

G

The optimization function is also next, please is, is also composed, is subject to to a two type of constraint: the constraint related to the natural resources and construed to the application resources, the lack of full knowledge make or control over the underlying network. It can limit the potential values of of X, for example, a delay, bandwidth packet loss, etc. Next, please using the multi domain. Adore abstraction is a method to provide such cost. In order to solve this, this optimization problem next, please, and what is the divisive formulation? Basically applique into that?

G

We denied we're asking for to carry traffic us for a set of flows sample. He will have to two flows: a F, 1 and F 2 on each flow will travel different networks. Each flow cutters had a set of network properties at delay, available, bandwidth, etc. Hanging a multi domain settings, a another property may involve nether property of multiple component networks. For example. The delay should be the zoom of the delay on each network that the flow will will travel next, please.

G

So the basic idea have to provide the ability to discover aggregated, expose information of multiple networks to provide a single, consistent built on network view and represent this information using a generic compact mathematical programming constraints. Next, please just to give an analyst at the example.

G

Here we have a single neighbor where, where a user want to reserve two circuits from search squad to his nation, one confluence from source to destination, to Bob Sieck, we share common links, your link, 3 and Link 4 a so that is not possible to reserve a hundred gigabit for both circuits next case. So using the mathematical programming constraint week, we can generate linear inequalities to provide a comparator presentation of the available bundle of flows through a network example. A x1 and x2 represent the available bandwidth for for flow 1 flow 2 respectively.

G

Next, please, this restaurant represent the resource availability, radio of the neighbor for providing resource to a set of flow. As such, some a linear, linear inequalities can be redundant, takes place using the redundant inequalities, egg can be removed using a polynomial time, optimal algorithm next, please. So, let's take the previous example from a single network to multiple network. Here we have a user that went to reserve to three circuits from source to destination.

G

One two and three: each network will will using the mathematical Brahman constraint, which each member neighbor will disturb the bandwidth sharing for its circuit into a set of linear inequalities. Next, please, although each domain being already conducted on the redundant systemization, there can be across the my redundancy, for example, the flow to and flow three in a member tangled and but never drink and eliminate the flow through and flow. Three member and I work to next, please using the multi demand, I, don't and Septimus a show. We can remove this.

G

This London and next place just to tourists, reverse view, basically, the key main idea to aggregate the extraction in multiple net or into a unified single single ambient WA representation, and next please just just to quickly summarize a multi-play import. A multinomial ocation can benefit from alto. However, new this attic warming also merge and different graph graph have been proposed it to address some of the design requirements as net test.

G

It is to make a systematic investigation of the deployment conserve of alto economic incentive, fifth ability, scalability privacy security consideration and to make a systematic design of the attention to address the reset requirements.

G

That's it as just to mention that we have regular wear quickly, meetings every Wednesday night, 30 us ester time for feel free to join us. That's it! Thank you.

G

C

Know you have a draft associated with this and for a long time, obviously we are not going to adopt any new drafts in the working group, but considering that this is something of interest both to the research community and perhaps the operator community. One thing we might want to think about is moving the draft head as a independent ad sponsored, but we can talk about that later. Clearly, you put some work in the draft, so I think you know we can. We can move it ahead if you guys are interested. Ok, thank you.

C

Alright, right.

A

Yeah, you are Jenna I, didn't discussion on topics beyond the working group Richard, you want to say something or others.

H

C

I

C

You have some time.

H

I

We have five more minutes if you wanna video talk about some ongoing things which were working on and the issues which want to address.

A

B

I

So clearly, this is just mostly trying to really define a little bit the issues or items which we might want work and, of course, there are beyond account working group deliverables. So therefore they are not really in the Charter and so on, and it's a result of a lot of discussions. Yeah a lot of discussion with a lot of people and I think that if you, the wonderful job to show people about it's like a regular meetings, we have every Wednesday.

I

So so other people are more welcome to drawing all these discussions and I think that they're, wonderful, wonderful, these couches. So next is please so, basically at really high level- and this is an beyond or current working group items. I said we're trying to address two issues and one: yes, Network information, abstraction for applications, basically, what kind of content? So, basically we're really talk about abstraction.

I

So therefore, what kind of abstractions and the content we should really come up and the second one, of course is we have all the content and how to really carry those content from the network to the users and that's essentially, information, retrieval or transport and that's essential. The protocol path. So therefore, essentially we're focusing on these two aspects and about work. It will be in the context of application or aware networking next slide.

I

Please and I probably can skip this one I think most people are already familiar with, essentially, let's Auto Protocol and as the transport layer thing and just pull the slides over there for people to take a look. We have all kind of things. I. Think Daniel also had this like a very nice picture, showing a central of the bigger pictures. Next slide. Please, and here is a set of abstractions, which we, the auto working group so far, has defined already.

I

So basically, what kind of abstraction Yury one abstract you don't want to go to like details, for example like a tunnels and all like a very, very fancy like forwarding tables and so on. So here is a list of abstraction basic, you have nodes and you have a pass. You have Co flows and so on. So those are most basic abstractions, which we have introduced in working out.

I

A working group next slide, please so this one probably slightly easier to visualize about working abstractions, which are the auto working group has introduced, and you have boundary of nose and the taste and I think that's the general concept and the and the entities would essentially, you can read a great office entities to define something called networking map and of course, then you have like either, for example, from the entities or from aggregate of entities you can define in component car services, essentially the link property and of course you can also get all the properties on demand.

I

Of course, you can also have Co flow, which means you really worry about set of Transportation's and really go together and that's essentially called pass vector, and you really have abstraction called call flow next slide. Please so I think we can skip this one. Those are all the current working document and so on, and we can skip this one okay, so now, I do want to talk about a few ongoing efforts which potential it can be interesting and follow group.

I

So one of course the item is the alpha deployment, so the basically setting is very simple is focus, for example. This is one example of deployment. I think Daniel already also mentioned is a wonderful is, for example, focus is collaborative at the Shubert hit, execute data sciences and, for example, one driving project is the project called sense, is team phone and in network the sciences at access, skill and basically application already planning to be supported. I think then mission I would see here is essential, black hole, astronomy and they really Clark.

I

You terminal data, doing all these like analytics and one meaning work. I think, for example, using this current group is to solve the issue. Is a car service is reporting for exam time block maximum bandwidth by name is sliding window I'm time banning its production zone basically is, and those are wonderful, wonderful test cases to see if Auto can really support. Also, use cases are not so I think those are really will be should be one more type of which I think working groups are really looking to a lot of features.

I

Can we support these kind of things there already a lot about patient and for this earlier part, but we need you message if we can reduce all these, like a true use cases on that next ice piece, so the second one I think the items and beyond the current deliver both would be the following item essentially focusing on the protocol on the transport layer and those essential on top all the green buttons. Essentially, they define the current transport mechanisms.

I

For example, we have information resource directory where information resources and immediate hives and consistency and have grammar. We have all this stuff and I'm fortunate victory and so on.

I

So if you look at looking back- and you ask yourself, what can the transport entities might be missing, you can identify a little bit, for example, IRD with context information, I think that's one issue which something has a tremendous increase and which I think be very useful and in current, for example, currently we have single media type, essentially with the about one media type for one and more and more for all these consensus, consistency requirements or we premiere war about multi-part information resources. I think that's.

I

Another item would be interesting and right now we're focusing on single or JSON grammar. Maybe we should really look at, for example, like using Llamados as a tuner, for example, grammars, and so how much we should go down with it and iconic transport layers use SSDI, which is useful, which turns a very simple, very easy to use, and so on but Ryan, for example, there be a new, more flexible and more potential more efficient. The transport protocol, HTTP 2 and those external potential can also change our transport layer protocol. They are potential items next slide.

I

Please so here will be talk about essential all the new abstractions, which will be potential, should overcome I, think on life and say we talk about all kind of abstraction we'd already mentioned briefly, but there are a lot of like anything domains and beyond ipv4 ipv6, for example like a cellular and so on, all the domains which we have not defined and we probably define in some way and then we should have probably more generic unified.

I

You know the code flows, multicast multipaths, we're not really dealing with some huge systemic over yeah, so I think those I central abstraction we proudly introduce and we have XD IMAP, we probably for the go to the generic network capabilities or for all the network functions and so on. I think it become very useful, and currently we only have very simple filtering capabilities, and one thing we really have a lot of interest which problem it can be very useful.

I

Like networks, sacral, language I, think that's a really beyond anything which I know and IETF will potential. It can be very useful and, of course, also like Canada concept and so I think no very useful. Please so of course, last one which I want to mention about protector items which people are working on and really what I should mention. Is it's a back in and realize realization for exam, even today, a lot of discussion about what can imagine meant, and so how do we do measurements?

I

For example, there are a lot of cushioning already on, for example, by Lao and also tactful you very credit, triggered measurements and collect all the data start and collect. Madam smartly I have formalized protocols and real, accurate meaning, accuracy and freshness, and so on, I use proxies and so in central. There are a lot of these conditions and really back in his work, which can be very useful for the success of this particle. Okay and I believe that's all I want to say, I think I, probably finished Brent right on time. Yeah.

C

Perfect, ok, I think this is very interesting thoughts on what happens next. All I mean for those of you that have been with us the inception. You know the first idea for Alto was to just get a course network information to applications when we reach our turd. Although we moved from you know, a single computer view to what used to call it Richard, my network view or yeah might have could be of yes right to path, vector and and and other things, so the next iteration of fall tour.

C

Once you finish all of the documents that yeah right now, some of us were thinking, might very well be operations. You know, focus on Alto, but they're. Also ops, well, coops, yeah, okay, so, but that all depends on the energy of the working group and once we are done with our deliverables, that's something that we can talk about at that particular idea.

I

Wonderful right.

A

Because what what you described in the last night is pretty much in line of that with that, like.

D

A

The operation of this right, exactly provisioning interface, was intentionally out of the Charter for now. But you know in an ultra operational setting, that's or Alto offset, certainly something correct.

C

Yes, so if there's no other comments, then thanks for being here and we will figure out what to do in Singapore, so thank you folks and have a nice day. Thank you. Thanks.

H

So yeah, thank you. Everybody.