Internet Engineering Task Force 96, 19 Jul 2016

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF96-5GANGIP-20160719-1245

Description

5GANGIP meeting session at IETF96
2016/07/19 1245

A

The presenters please done to the list, but I think the ones and so yeah, so the presentations belong to the presenter, so we we cannot send on their behalf, so we asked them to send. This is my idea: I mean that I didn't somebody can, but I can send it. No problem. Yeah.

B

A

Slides are okay, they're public yeah; no, no, don't copy anything. Just listen! Yeah, okay, I! Guess we gotta start, but we are still missing two presenters.

A

It's little funny.

A

Ok, folks time is up.

A

Now this is a 535 gan and gip mailing list. I. Send you a couple of times what that means? Why I forget myself to IPS basically IP aspects of five team and we have been having these meetings a couple of times in the past, so this is I. Think second or third meeting now.

A

One thing I would like to say is again very quickly: we have we have this invited presentation. Type of you know the meeting, but you would like to shift more more and more towards IETF style meeting that is discussion of the drafts, so we we would like to encourage people to write drafts and have discussion and then this custom, and that is going to help us in the future. In for future steps like going for a ball friend so on to judge if this activity is getting into some maturity.

A

Okay, oh here is done. Ok, we are missing Michael, only I'm, sure, I'm sure, because his last no problem so, okay, scribe Dave Allen, agree to do the video skype thanks Dave now we start, then, with Professor fit texture presentation, so you got to connect through that. Then Oh need to don't.

C

Need to screw some thunder.

A

Okay, listen he's going to present from his laptop because there's a special presentation: I have this two-minute mark here. I'll indicate the kind to the horizontal.

C

A

We're going to be very brutal: okay, just cut good.

C

And so one of everybody- yes and the charity said my name- is Frank Fitts sake, I'm I'm working in the team racing at the deutsche telekom, chairs. You can see for communication networks and together as get advice and leading the fight cheat at germany, and I was asked not to give a little introduction about. 45 g stop what all is and how it links to it already ongoing activities here idea. So um one thing that we always tell people is 5 g is different to other technology, lat 43 good.

C

Now we already know the use cases of my FG and we can already see whether and the money would come from and from the use cases. What you can derive very quickly are these technical requirements right and I think most people understood. Oh, there will be more bandwidth like 10 gigabit per second per user, but what they also. He is this kind of agency, so you have vacancy and maybe many times in context of webpages, latency and Facebook and Google Amazon how they want and reduce the latency.

C

But this latency is even more because we, if Olivo, is one millisecond delay out of my two or three.

C

We at the factory of Germany always referred to one millisecond, because what we want to do is they only want to control, is steer and cyber physical systems, and when you want to do that, besides agency narrate your to need security driver reliable, the bad news is when you think about these, things have to get 10 gigabit per second one milliseconds each parameter C is already difficult to achieve, but you have to get and at the same time, alright- and this is what it's all about, and what we are thinking about.

C

5 g is not only that you have to think about the air at the face right. What it's very important to the network of the cloud, and mainly this is 5g, because this one is just an add-on, but the silicon is also something that really in for providing security negatives and have to have something like fast computing. This is the four tracks that we have identified in the 52 left, Germany and I, wouldn't just motivate a few of them.

C

So when you think about 5 g- and you look at do into the context of 12 25 G, then what you can really say is that everything that was before 5g was something connecting people right. Remember the Finnish company. That knows anything, people that was about 10 billion human beings that want to communicate with each other always sell this YouTube Facebook that right and if you talk about 5g, we talk about five front of billion things that you want to control, and this is already different and controlling. This is a week with this communication.

C

This is a liquid is controlling sphere and you also have to to support the 7 10 billion people over here. That's clear, but there's not the main thing and love to relieve the treaty partner game.

C

So if you do that- and we too often talk about the tech tile internet, so in an internet that gets your feedback in a certain and time, and this is the one really second, if you know think about what is the the pink time nowadays, if you go to 4gb or something like 30 milliseconds, but this is just a pink time on layer 2, so you involve the cloud. You will see higher values there. What we think about is a one millisecond end-to-end delay, and where do you get right?

C

We talked with a lot of stimulation bodies above this IDF is one of them I to prepare etsy and all of them talking about the higher and five teeth, the first network that really supports this kind of tech, cart internet in the end, and and if you talk to sorry about latency, this is the picture I very light. You see some gamers there. They always knew that late. Is it more important than data right and the nice thing? Is they also give you the explanation: how to get no latency you just get it.

C

Opportunity is something that brings you. The latest, unfortunately, I, cannot show you this video here is not starting on the machine, but it's something that we build up some years ago. It's just something like an oculus rift where you have a camera on left and what you do. Is you just see what the camera sees? It's, the normal one, one copy of the real world, but what we can do that is truly to adjust the latency. It's very funny to see people fighting as a mortal 1587 expertly later it's really impossible right.

C

If you want to try it out, we also did it as a catwalk in application. Nowadays you can just download it and just try it out yourself. So if you don't think about what is this mighty kind of use cases, I mean steering something like humanoid robots? Is something interested very much for us? If you think about this industry and 4.0 I psychology and Germany or industrial Internet, if you think about something like robots nowadays, they are standing still.

C

They have one program and they're just running that in the future, these robots will interact with each other, and then you need the 182nd. If you talk to cooker with meeting these robots, they will say one incident is unless in there it would like to go lower than that and then, of course, then you have something like this kind of drones flying in the air, not only using this network to control these drones. A lot of people think about amazon just bring in my next new gadget.

C

But here is a lot of these things that are building the network, not only that they're using it. They can also build it right. It's not only the drones could be cars, there's something like smart grids when you want to control different kind of energy grids and wind plate. If you want to bring that together, you know that you have to and look at the faces they have your 50 Hertz and the energy grids. If you want to put them together, one millisecond is already 80 degree of facebook.

C

So if you're 10 milliseconds, this is a shortcut well, we can live with reactive power, but shortcuts it's not what we really want. Also interesting. These were the first people came in coming just saying and when they heard about the one millisecond they said hey, you have a network for one reason: we need this, so the farmers out of the field they want 187 my job, this play or say: if the flowers get it didn't want one millisecond. The rest also would understand. Okay, my father's view and I've just explained is it that.

D

C

With the control of the of the machines to each other all right, this is something the way they have to be very precise and the other thing is they want to read to be precisely on the ground, so the soil is very important. They don't want to go a little bit that a little right out of here thing really work. You have localization and they do, though, the nation but communication plummets, one of the things that I did not mean here.

C

I just make it a little bit more emphasized on this, because we we are in Germany and reveal about with car manufactures is about already control that you have in a car. Apparently, you have controlled between the wheels, and this is already one millisecond or to Billy second, depending on the, where you buy it. If you're a bus system, it will be one millisecond when the wheels communicate to realize something ESC or abs I mean future.

C

What we want is we want to have cartoons right and if you have platoons, that is not only single color, you have to communicate with all the wheels you really want to bring them together to own a clean, very few hours 50 centimeters half are, there has to be really precise communication and the more things you get into your cue, the more complicated because so the one question is: why do we do I didn't really need communication?

C

Why don't you can just you how to know his cars right, also notices for me, cars that are just rely on their own senses to.

A

C

I would be them so what.

A

C

A

C

Robots and these robots are just following the line, and then they have a distance measurement to understand what is in front of them and what you see is then doing a pretty good job right and they do it as long as there is no disruption right. We just have to wait. Even if my treatments are running that now, you see there's a disruption coming in the end and the first one will stop and then you will see.

C

The second will stop and the short one will stop nicely and the force one and after the force one you get a little bit of cakes. This is control theory. If you combine linear system or you go to order number eight, any other problem right then. So, therefore we need some kind of communication, and what we want to have is something like this system right.

C

If you have a network and the cost can communicate directly, but you also would like to have a communication where you go directly to this kind of streets and what you see is you want to have something like this? You have a network that can steer the cars into dating section and this network. What it will do is it will combine all the sensors of all hands together. You make better decisions, remember the Tesla problem. It was looking for a truck didn't see the track.

C

If you just would know that there is a transom so and just to tell you that we have built something like this and you. What if you want the video is there on five th left Germany de web page? We built such a system where you have eight cars, they still have all turnover sensors, but may also on fire. A network share these this high with each other, and you see the intersection commute.

A

Sorry I have to get to have to cut ok.

C

And let me just show you the last: if you don't he laughs and that's it just. Let me give you the last cycle.

A

To last year,.

C

But this is the one where I want to come to where you say you and you have to 5g. You have these requirements and we shall make the other kites. Are the bullets where you would say this is what idea Charlie good right most of them, and this is also what you have to fulfill get 5 cheat right if you think multipass multipass this piece here at half this year, network coding is a group. Machinery is now starting mobile edge cloud mm. Espn is the combination over here network slicing.

C

So all these things are currently they're discussing this already. But if you formal together, this is what we call with five th all right, so I think you will hear about this and I think we are the right place here. It's mai-chi just to understand how we relate to all of the new vehicles. Thank you very much. Crimson.

A

French can go so come I.

A

Think we wanted to have some Q&A, but yes, we live it to the end. Okay,.

A

You don't need.

A

Yeah and then you did.

A

C

Was fine there you get your own one right, no.

A

No there's only one this.

C

One, oh you be the presenter ya.

A

Know the USP now your seats in there was there. I need the ESP cuz, that's the control which PK no, you need. You can do that monday. Okay, okay, okay,.

E

Great um hi, my name is conquerors and today I'll be presenting ila four five Jake. This is some work that Julie's Miller from AT&T myself have been kind of starting, so we present or if you're, that a little bit last time grade the ball n minus their worries. Thank you. Buck and I'll give a little more detail specifically how I la might be applied to five feet. So just a quick recap, my layers and identifiers locator kind of protocol.

E

We split ipv6 address into locator identifier, the upper 64 bits are round of all these are potentially identify each physical host, the lower 64 bits for me identifier, this kind of like the virtual or logical address. um So, given that we have the 500 corgis, parvez, I, Cleon sinking, ipv6 these really good fit for that.

A

How do you need equations for the onion flavor so.

E

A

E

There's obviously, a lot of management role, playing nine, ok, so the component. So when nice thing about Iowa run on imposing devices work from the network, one important thing obviously, is a mapping database. So this is mapping identifiers which again are like the local or logical address to the physical address of devices.

E

So in order for net were like this, we have to distribute this mapping table conceptually the site is the full size of the table is up to the number of object which can so we're thinking about ways to scale this into the 500 million or training range, and then there's a control lane behind all of this which distributes the map in- and this is somewhat like- a routing protocol- there's actually some national efforts for doing things like this in BGP, but we tend to believe it's more of a very large-scale distributed database will have some probably some more work to do on that.

E

A

Look again so did this mapping database is Indiana's.

E

Wife DNS, it's mapping one 64 bit number to another. Essentially, so it's a database lookup one of the things we're looking at and I believe we're going to start. It's a mailing list on this is something called hyperscale address management. It's basically tackled this problem. We think we believe it's a generic problem. We've already seen this in the context of NGOs 3, where we're trying to map virtual addresses to physical addresses, but in a larger context for object, mapping, shedded STM's scale is off a lot advocacy pipe enjoy number.

E

I should mention that in facebook we are intending to deploy my OA, and this is kind of the the overview of it. Fundamentally, it's simple: it's fairly simple, a source in the packet. Does a dns lookup, usually yes, manager s and doesn't go this address other than the normal ipv6 address sends it on to the network and somewhere the network. We can actually translate this basic virtual address into a physical address by overriding the first verse 64 bits with a look actually I locator.

E

So in this way we sending packets untranslated the eventually hit over Colin I Oleg router. That does a translation and that's and on to the actual physical destination. Where the note with that identifier, please and one of the optimizations we can make. Actually as if the post device supports I la, we can have some sort of redirect like mechanism resolution protocol, send it back to the host and try to eliminate the triangular around so there's some optimization can go n to it.

E

Looking at kind of some of the other solutions or concepts specific to 5g, I would say: there's a few advantages by awake. So one is, there is no consolation involved. If you send 64 by an IP packet, ipv6 packet in I la is translated as a 64-byte ipv6 packet. The network will see this if it's a TCP packet at see the tcp / IP packet, which means it can do all of the normal network, optimizations Hardware off loads of things like that.

E

Also, since ur throne capsulation, we eliminate a lot of the headaches of encapsulation and trying to figure out MTU issues. Dealing with you to protect some other telling another advantage is are no changes to the transport layer. This is really a difference with IL NP, which actually kind of basis violate I like being require some changes all the way up into the transport layer, which really makes it hard to party to plumb the post and then on a router or I'm, not sure what they do.

E

I guess they have some sort of proxy um and then the other thing is.

E

Yeah I look at this is also transparent in the application, which is nice and good. So one of the nice things we want and the mobile cases make the application s tram as transparent as possible to the networking by search. We also have an open source implementation.

E

So this basics of the I la hoz have been up to greenville endings for a while, and we also have a hackathon less on good Saturday Sunday and the cisco guys brought up in il a router / b DP, which is really cold and actually have an intern right now, working in an alley outer and SVP, which is another high performance data pack. So the upshot of that is, we believe we can implement and I only router in commodity, Monday, servers, commodity hardware, basically pretty official. So.

D

We think we can get 700 million packets.

E

Per second gone through a normal server go annihilate. So it's not to say that, were you someone martin report, but we can deploy without hardware support in the initial ground.

E

So, looking at the requirement specific to mobility 5g, we want to consider homogeneous or heterogeneous Gandalf when you crossed networks, IP addresses changing things like that: a single user device via mobile versus a whole network so that cruise ship analogy when the curtain goes whole networks change location in the network. Where would I only run so we actually have several options here.

E

First option is basically run it between the border router to the Internet and gateway somewhere in the middle of a network, and then Iowa would occur between the border, routers and blue or between different gateways that those would be doing they exist. We routing and then beyond that the base station from it and then user devices would see normal packets in the spirit of efficiency latency.

E

At some point, we want to push that closer to the user closer to the edge, so, for example, in this case to user devices could talk just through the base station to each other. The gradation does the ILA translation handle handles that or you can talk between base stations and, of course, we always want to be tough, be able to talk from the user device all the way that internet or external networks, in which case the ILA really occurs at the end. So the middle.

D

Of the network, you.

E

Would see Iowa after five dresses at the edges, you would see the untranslated long. It.

A

Took two minute and.

E

As I mentioned, we have, we do have the ability to push us into the host. My concern with mobile, obviously is: this does require modifications to like the OS Android. What have you it's very possible and it becomes a nice optimization, especially in a multi-homed coast, which obviously most mobile devices are going to be anyway.

E

So in terms of control, plane, pended answer your question: identifiers are allocated globally within Facebook. We have a mechanism to basically allocate unique identifier than the time stamp. So what kind of depend on the use case? We do believe there will be some higher allocations, so, for instance, if true providers we're participating in this, each of them could have one component of the 64-bit namespace. Then they can allocate identifiers within that space as they wish. Now these identifiers are compatible with two different fighters, so they could participate in ila and with the next.

E

That's where we didn't get an interview. Scandal there's some more detail that mostly Julius did is more specific to some of the 5g architectures and.

E

Pertaining to the at the model of 5g, we also want to consider when we do move the applications into the network. I think this is the only ultimate goal will make the service model closer to the actual devices to telling that one millisecond latency. In this case, we can have the application servers and participate directly and.

A

A

Yeah: okay, yes yeah thanks Ella tom, so we go to the next. The next talk is the bitter.

A

Dispute- okay, just clean this, and yours is better playing this fun yeah. Okay, all.

F

Right thanks, amor Mele this. This kind of our subject takes about an hour and a half of questions. I've tried to distill it down to 10, minutes and soggy kind of interesting. Then we'll start with just the whys and what's of slicing, the first presenter talked quite well about all the different permutations and combinations of the area faces and whether that requires different aerator faces to support them all.

F

But essentially you can think of it as different qos, different, purely requirements for many different types of new agency, guy or T devices which require low bandwidth and almost a connectionless type interface, and then smartphones, of course, with very high bandwidth. The delay or blind reliability requirements which you saw in some of those first videos, and you can imagine extending that out into a real automotive environment. It's scary, but then you've also got issues like very fast mobility.

F

Mobility I mean dv/dt, right, I, don't mean in terms all our immobility, I mean in terms of the speed at which something is moving, and the current OFDM technologies which are used with 4G. Today. Don't work up high mobility speeds, but if you want something more fun ability speed, that's not necessarily efficient for things which are really slow, so you end up removing multiple different for FGM technologies.

F

So you end up with requiring different OFDM technologies for different spectral bands that are used simultaneously, and this leads you to the concept of slicing off the air interface right from the get-go. So this licensing requirement comes from the problems and challenges of chopping up that area to face and using it for multiple different uses. Multiple different connector lessons, but lots of other interesting issues around being able to create networks quickly and migration and so forth.

F

Another way to think about what slicing is you can look at it from the ue right end user device perspective. So if you group, you ease by space capacity coverage, qos delay, administrative straight to your main, those form a set set can be thought of as a slice lose you department, slice now in terms of the network. What does it mean from the network perspective I, thought long and hard about different ways of describing this. The easiest thing I can come up with for folks like yourself, it's think of it as a set of subsets.

F

All the resources I'll explain that in a minute, but that set of subsets forms a new set which essentially becomes an end-to-end network. So from the yuli's perspective there there are three or four different ways. You can slice eue right Ichi we can be in a different in its own independent slice. That's pretty obvious! Pretty easy! I can come vertically through the you eat and have different, essentially different virtual machines inside the ue. So the radio access technology presents itself to an individual. Sios sees one interface.

D

F

Apps see 10 s, it's like having logical movies, with a non-physical you'll eat more interesting, I think chocolate horizontally where these different radio access technologies that are supported by the you at the same time present themselves as mobile interfaces to a single or less the applications that have access to different q. Haces within the single us see what you'll eat and, of course, you can get into complicated combinations and permutations of foods too. So they're they're slicing from a unique perspective.

F

Now from a networks perspective, it gets a lot more sophisticated, a lot more complicated.

F

The way I'm describing this is the slice is a subset set of subsets, so the universe of all the resources that I have available to chop up to slice who starts with the antennas the analog front hall next to those antennas, and then these three sets in the middle here represent the digital signal processing, which is essentially driving those antennas.

F

So this is the hardware, the CPU and hardware, resources that drive the antennas fabric, which interconnects not that hardware and compute resources, and then here we refer to the numerology, the mathematics or digital signal processing, which is actually implementing OFDM or other technologies at the antennas. So from here, how is sort of an analog interface from here back? We have backhaul pocket interfaces, so this could be a data center fabric. It could be a core.

D

F

Transport network etc, and then we have the actual core network of the packet processing portion of the of the 5g network, where in forging network which is essentially in Italy, the implementation walls, the servers all of the resources and so forth, and then a slice can be thought of as taking a set of these subsets.

F

So some of the antennas, some of the front wall, some of the digital signal, processing, fabric resources and numerology, some of the back all the data center resources for the record- and you do not have to have a proper subset right. This can be you can you can use the complete set?

F

If you want you for a slice which leads you to an interesting diagram which shows you that I've got four different: slices red, blue, green and yellow slice that they're sharing resources up to the point where there's a fork in the resources here for different radio access technologies? Perhaps the red slices using a different type of OFDM different different spacing between sub carriers, which are better for high-speed ability and maybe the others require more normal OFDM and then we're splitting here at the EPC, so I'm using different core processing.

F

Maybe one of these is a traditional IP core. Maybe one of the others. Is this based warm on one of the non-traditional? Ip core? So you have the possibility of running and splitting at different points and I can split into two different further slices after I finish the report pro system because I might have different content, but I wish to provide or different downstream processing. So the slicing can be thought of not only in terms of the physical resources.

D

F

Logical partitioning that happens straightens your almost haunting these treaties, it's nice like treaties, once you have this ability to to to create slices, then the obvious question is that I probably want to be able to move resources from slices lights and again this starts with the driving requirement at the input, which is that I want to move spectrum from one slice to an art all right. So take your point in time. One of my slices is going to use some amount of capacity and they are on the day.

F

I wish the switching the resources to a different slice. I need to be able to move and resign that spectrum on the fly. If I reassigned a spectrum, then things happen downstream of the more capacity downstream to do the digital signal processing and we walk across it down the street to do the packet processing and that may need more capacity downstream for the interconnect. This all has to be make before break right. We know how to do make before break here in a vspace virtual machine move on and so forth.

F

We know how to do think before break transport. Backspace we've been doing that, but what happens all the way back to the front of the system?

F

Alright, so just to to quickly go through some of the areas that are potentially a packet or impact 5, 5 g vice versa. Obviously the address in growth visit is an interesting concern.

F

Information-Centric networking- and this are very interesting topics, because they potentially require a change in the ue and that's always their impediment. For some of these new technologies we're slicing, there will be the possibility of changes in the ue and there will be the possibility to run brand new technologies, experimentally or even to migrate them from experimental into something.

D

F

Worldwide, through through slicing technology, so slicing technology is, is an interesting way to adopt these new and interesting evolution. So I think.

A

You're going on time, minimun yeah, we showed two minute.

F

Now you just distracted I'm.

A

F

So there's also there's also the possibility I could said using these technologies internally to the next generation course as I produce Kimberly. So the sort of two interesting use cases for ICN, listed cetera in this innocence and I apologize if I'm taking a whole lot of people's work and loving into the GTP of course, could potentially be extended.

F

The function chains because the the next generation cocky quorum, in fact in X pocket full of 5 GTA is implemented, will be implemented virtualized within within a cloud interconnect downstream. All those Corban Collins to load, balancers and firewalls and so forth, is a perfect application for sfc I to RS forces and other nasty and related technologies are highly relevant.

D

F

We some of the packet core functionality is essentially virtual. Router and the virtual routers are should be split into a virtual set of virtual control planes and a set of virtual four o'clock, and you need to communicate, of course, between the virtual control planet. We were supporting plane and the idea we do. That is one of these technologies, but.

A

Yes times up times,.

F

Up anyway, multiple other technologies, probably another one of great interest- is multipath TCP because, as I said in my first or second slide will be showed, the ue with multiple slices, multipath TCP, is an ideal way to use those two interfaces and it's entirely. Luckily, the 5g radio access technology is going to come over concavity for prior matter. So and, of course, yang everybody, yeah.

A

Okay thanks man- next is, you know he knows took, is yeah, oh he's going to show it oh and don't know if I have enough USB.

A

My speech space I need. Is this because.

A

Yep you only enjoy all I need. Is this want you to connect to the iphone? Oh, my okay, everybody please! If we don't appreciate via white sheet, if you didn't read your name, please, my team.

A

The Mediacom.

A

Okay, oh yes, sir okay,.

G

So I'm going to present this PID mobility and it turns out that most of this mobility features are across the reader that giraffes and the authors of all three internet-drafts are listed right there, so I'm going to have a longer version. This is going to also be presented a list working group thursday afternoon. So if you want to get a longer version, he's come to that and also I'm doing a Pecha Kucha presentation on thursday night at 1030pm, sorry so late.

G

But it's going to describe the last ten years of list it's kind of this mr. nikken in the fall. So it's going to describe the entire chronology of loose. So I would explain in this presentation why this solves a mobility problem and I'm going to explain that there's actually two forms of separation and use located on the separation and that we also have an optimization column where we can do handoffs at near zero packer class. So I can continue with that or just do you does the audience want to see one minute.

G

One slide presentation of list, yes or not? Yes, look because I think there's some oops that should have candy.

A

Is remote and read Cosmo I.

G

Can the distance this is supposed to be a hyperlink? Ok, ok, ok,.

A

G

The bottom part is list, bob, has a data point and basically host systems that are assigned eid switch, are just regular, ipv4, be six addresses, there's no special syntax. They send packets, postdata host, be sense to each other, and then there's these lists xtr that it kept sleek packets and send them over the Internet in to their our lives in the air, looks addresses that a rabbit in the core routing system to be able to get the mapping from host be goes through up our love.

G

Why you go to a mapping database system, which we have many various proposals on how to do it? The API to the mapping system says that the ATR encapsulator since a nap requests the map resolver it finds where the register d ideas through the mapping system, the captain chris can be forwarded to the ETR to be responded back or he could proxy reply allows the ITR to cache the mapping, and then you cap sleeves all packets.

G

After that, that's it in a luxury off I can not ee trading I can go my cell Hannah, okay, there's really only three ways to solve the mobility problem and if anybody tells you there's a fourth way, it probably fits in one of these three categories. If something's going to be moving around- and you don't want to use any locator I do separation mechanisms, you either have to use mobile IP or you have to inject the movie ends it to you for the rotten system, I'm going to explain each three of those mechanisms.

G

Okay, host Rob pretty much it's been experimented over the last 20 years. It's pretty much a non-starter if you, if I, have an IP address and I want to move around rejected and routing system that single host route for this single entity. That's moving has to be injected everywhere in the routing system. That's going to cause handoff problems too much state and there's also policy restrictions where the horse crops can't get everywhere.

G

You want to get where you want them to get to and if you're multihomed now I have multiple dresses, so it adds to the state problem in the latency. So it's pretty much a non-starter.

G

So we have the notion of home. If I have an IP address, it is attached to a subnet. That means it's part of the topology somewhere in the internet, that why is that useful? If I'm moving around I really don't have a home patio magic right? So conventional wisdom says that if I have a home address that my pecs are obviously they have to go home and that home agent s defy the floor. Agent ascent packet, see conventional wisdom, says doing triangle while robbing as latency why you can start there.

G

We want the shortest path between any two points for communication, regardless of their moving. Okay, now as multihoming work with multiple homes, what, if I'm going to what? If I have two IP addresses and once at my physical residential home and ones that Mary work and I want to use both of them? Do I now have to route packets to both homes, one or the other? It's confusing. We want to be multi home.

G

Most devices have multiple radios, so we're going to need to support that and when you run, um you may roam too high bandwidth area, but your home location has first family, so you're going to be effective. Why can't you take advantage of the higher bandwidth place that you roam too? So the notion of home and assigning an address?

G

That's something associated with home is probably not a good way to solve the mobility common locator ID separation really is the only solution because, what's moving is something that's fixed that never changes which is called enlist, input, ID and point I. He is always fixed to that famous movie, and what you do is you reattach to the number of getting a new binding to routing locator? Okay, the moving entity has a single address that makes things very easy for debugging.

G

Even though you're attached to multiple radios, you can have multiple our looks in an hour load of stuff. That's a single tid maps too. That's the way to solve problem, and you always get the shortest paths between our ones low latency.

G

So we have two forms of separation in unless the idea now looks that can be co-located in the same moving entity. That's documented in a list: mobile, no giraffe and the ideas of moving at s a moving entity and the our log is a stationary. Why? Which means the eid moves? Does not support less and the are low gap discovers the IE does support less and that's an eid, mobility Europe. So what an example of that is.

G

Maybe a virtual machine in a data center has an eid and it moves for one day, etcetera to the other. The outlook that changes from is one top Iraq switch to another top of and search.

G

So this is an example where this is running on the ue it's assigned in eng, maybe to its loop back address and it retrieves these outlooks 2ghz piece from these different providers. This is what's documented with mobile node. This is not necessarily what we're recommending the list working group is recommending for the thigh g solution. You want fast deployment, even though we could Peter sent me to support and using slicing that's something we maybe could experiment them, but we probably want to use this solution.

G

This is what the IP mobility draft says. The eid is the ue, but the Arlo's are actually in the network. So maybe an access point for Wi-Fi, maybe a you know: Bjorn s gateway in the 5g world.

A

G

It do it ok, so if you want to know the details, the dress I'll go into detail at the list. Working group, so I'm just going to fast forward to the most important part.

G

So we have, this is the third draft that does make before break this is going to cause near zero packet loss for handoffs. We call it a predictive our lives. If you know the direction you're going in, you have the ITRs encapsulate pack and send it to all of them, and rather than finding the moving entity, the packets are going to be wherever the movie yet to be makeup. So it's going to go to to be dropped everywhere, except the place.

G

It's currently f, if you want more information on predictive analytics, we're giving a presentation on that in the Miss perky group.

A

You know go backward okay,.

G

Okay, so basically, how is this fit on the current five-year proposed 3gpp architecture? Now, if you look at this slide, there's a lot here. This is existing Nautilus sites, branch offices, data centers, where you see these double color things these, where it lists can run so those are list sites over there. A branch would list a data center with lists, and now we can look at all the possible combinations of where you could put less okay.

G

So, first and foremost, today, list was running in these environments, where r SI PE routers, encapsulating to a top iraq switch, and so this internet does not have to carry the IP addresses of each of the course okay. So the first thing we could do is we could put this peer into the encapsulation between the s gateway in the peak a way that would be. That would be good and then what you get is whenever a uni moves, ue, Simon, Eid, it's an IP address, can be v4 or v6.

G

One question that has the answer would find cheap is how you explicitly not support, acted before very important question anyways. If you eat moose from this, this SMS gateway to that s gateway, then the new binding is done so the PK menos and caps like there versus there okay. So that would be a good solution.

G

Another solution is clicked closer to the movie andy. So then you get much more granular mobility, domains or Vig moves from one note to another. You could have it done there in most cases where the decap happens at the peak gateway, it could rien kept to maybe a branch rather than its talk to. Maybe this phone is talking to something at a branch, this guy in caps, lates d caps lakes, the packet from here re encapsulate here this would be called an RTI function. We encapsulate internal rocker that could be done.

G

Probably the best solution is to keep it close to the moving entities, so you have good granularity and that he had caps lakes all the way to where he has to go. This is supporting an overlay where the functions are in the network. As close as you can to the edges of a network, but not in the edges, yeah there's no reason why that top Iraq switch that's doing the D capsulation can't be in a vm on a server somewhere. Today's but I think this is a network function.

G

Let's say the CPU resources in the you ease and then the servers for doing applications and let's put risk in these various things. So I want you to. You know when we have the time I want to entertain thoughts and where people think the best place to put.

D

In you can see.

G

That there's a lot of flexibility there. We really want list to be able to go from this point. To that point so could support all the various flows is.

A

Michael here oh.

G

A

Okay, so we're gonna have to call them yeah. There's another released talk. So just hang around and Michael Michael meant ok, Michael.

A

And we need the USB yeah it's in here, Oh No. Where is the aim USP that comes into somebody took? It is not in session. Maybe where is it?

A

Then I'm will have to go your speed that comes with this. Anybody has AI.

F

Using this one yeah.

A

I know it was here something anyway go ahead. Michael I'll change from here. We.

H

Agreed on my name is clear.its. Microphone is like a fun yeah.

H

This is slightly different, but it's work him. This is also located identifier squid, but it's a little bit of different flavor than a lisp and I LA, and the application of 5g context is probably pretty much the same. That's why I will concentrate on the differences.

H

This is a work that was discussed about, say, eight met eight years ago in the research routing group, and it was published some years ago in the future internet journal. So next slide, please here's the all of you. We have an ipv6 internet backbone. We have some local domains and the idea of GOI split is that we have a routing locator that is specific to a local domain and which, which is the local locator and routing locator.

H

That is specific to the global routing domain and within the local routing domain we use addresses that are composed of the local locator and the identifier and within the global routing domain we are using an address that is composed of the identifier and the global and a global locator. That's the main idea and the attachment point between the local routing domain and the globe will go out global routing to main a so-called GLI gateways. This is very similar to the other approaches like an IT, RS or so interest handle router.

H

We have a interworking.

H

Concept with exit with classic ipv6 domain and within a GLI domain, we may use classic ipv6 nodes and upgraded G&I ipv6 nodes next slide. Please actually.

A

Like you can change it, yeah I realize the DSP is already there. Okay and you can use those as a point. Okay,.

H

So some nomenclature, we have the identifier, we have a local locator and the identifier is used to look up the local locator in a local mapping service. The global locator can be looked using. The identifier in a global mapping service.

H

So this is the address structure. We have three different types of addresses. We have identified addresses that contain only the identifiers and they may be used for lookups. We have the local ad. We have local addresses that consist of a local locator and the end identifier that is used for local routing and we have the global address that is composed of global locator and the identifier. So this allows us to have an attachment attachment attachment points within a local routing domain that are independent, independently recorded in the addresses.

H

So we can change the attachment point in a local routing domain without changing the attack without it, without changing the global address of the device.

H

Okay, so how does this work then? Within a sink within a local network, we have a node, 1 and node to a node one wants to communicate with node 2, so it looks up the local locator and pushes this local locator on on to the packet header, and this packet can be used for forwarding in lowering context. This works. Similarly, so node one wants to communicate with note to.

H

It looks up in the local mapping service, look at the local locator for note 2, but it cannot find it so it carries the global mapping service and finds the global locator for this destination. Node. This packet is forwarded through the July gateway. The chill I great way substitutes the local ohaider of the sender by its glow locator. That packet is forwarded to the destination and yeah. That's about it. So we need this July gateway to hide any local structure from the outside world, and this is a address rewriting mechanism which is stateless.

H

So this is the big difference to nap.

H

H

We have preview, have quite a bit of flexibility with this GLI domain approach with this jeedai approach be kept because we can use a different path, and this is all transparent to the end nodes. There are also.

H

Interworking mechanism between classic ipv6 domains and Gia Lai domains- and here is the summary, with GLI split, which is very similar to I LA and to Lisp. It is possible to change providers without internal re numbering.

H

What is the difference or the commonality the similarities to list in Lisp we have, we have used encapsulation and in classic list the identifier and the local ok de, was kind of in the mingled in ila there is the global locator and the local locator combined in the locator part of the address, and here we have two different addresses used for it, the log, the local address and the global address. So it's just a different approach. The benefits are very similar because of the locator identifier split approach. Multihoming is facilitated in for smallest july domains.

H

It can be exploited for multipath forwarding in addition to traffic engineering, improved reliability and mobility. Support july split is incrementally deployable on a per domain basis and also within a single domain. The migration from non upgraded, July notes to upgrade chill I notes can be done gradually, so yeah.

H

We demonstrate the incremental deployability of cheer I split by means of a proof-of-concept implementation in on net plus plus GLI gateways perform simple address, rewriting without the need for session state. This also holds for inter working with classic ipv6 internet.

H

In contrast to many other proposals, chill Isley does not need triangle routing, via exit devices for internet working for, inter working with a classic ipv6 internet. The full set of benefits is available only for communication among chill I nodes with upgraded networks tags, but a chill. I split, provides also advantages for upgrade GLI nodes when communicating with the classic ipv6 internet and even for classic ipv6 nodes in GLI domains. So yeah thanks.

A

You finish doing on time, even earlier yeah, so which leaves us a couple of minutes for discussion. You can go to the mic or just stand up and if you haven't asked questions, Mike or Michael or anybody else should we have a couple of minutes to for discussion. The.

H

Main point I want to make is located. Identifiers split is a very attractive mechanism, but their value. There are various approaches to do it. Yeah I.

F

Just a quick question about the global mapping in in any talk of Lisbon ICN or these other next-generation data planes, it's always about the data plane and there's not a whole lot of talk about how the global mapping works seems to me. That's the Achilles heel. So again, please could you could you talk about the the global mapping or.

H

Altered to do you know, I mean if you understood or not, we do not require a specific mapping system. We could use any yes.

F

That is, you need at least one there.

H

Needs to be at least one. The works of my goals in the this working group, DNS, like systems, are proposed, the what name the DDT, for instance- and we also presented some years ago, firms as a mapping system, so there's in all the new paper about different mapping systems.

H

But this is not very specific to the actually locate identify a splitter program. Now I realize it's sort of orthogonal. It's.

F

Just it seems to me it's the Achilles heel, it's the it's, the part that you know the miracle happens box yeah.

H

It's a mapping system, it must be a very powerful, it must be reliable, it must be secure, yeah yeah, but it's a different problem. Somebody else's.

F

Brought me it's it, is, it really is? The miracle happens probably assuming this, then all this other great stuff happens, but how does that work.

H

Well there, for you know, oh.

G

Well, that seems to be the theme of all the overlay stuff. That's been going on the last five years. Everybody wants to argue over the data planes and then they say, control plane has the scale. Sb robust magic goes here and somebody else is going to worry about the magic. So over the last 10 years we've been working on list, we've had five or six different ways of doing the mapping system and we've concluded. We want to use DDT, which is a DNS like approach. It does not use a DNS protocol. All it uses.

G

Is the iterative look up procedure from DNS. We can give a presentation on that at a deep dive on that any time you would like it.

A

Is GDP DDT global, a delegated.

G

Database table Oh.

E

Hi Tom Herbert, so I think my question going to be a little more general. So we had you know great presentations today. I think the the case use case for 5g is pretty impressive. Yeah it's going to be if it materialises definitely a very large scale network. Unlike anything, we've ever seen. Yada yada yada, that's great um simple question: where we: where do we go from here? So will this become we're? Gonna have a buff, oh yeah,.

A

All or okay yeah, that's a good question. Now I. We appreciate that you wrote the draft about your presentation and there was some discussion, but you didn't give you didn't you replies to those questions by the way. Also, please everybody do their homework right. The drafts and submit you can submit anywhere, but I mean anything, but if you submit draft your name and then 5 gm, an IP or something cuz.

E

Even bother submitting graphs to a working group, it doesn't seem like no I mean you can.

A

Write something else, but you can right now: I'll group name mosa, but you you added you an acid refrain, name right and somebody commented or commented. I think, but.

E

You didn't reply yet.

I

A

We need discussions on the mailing list, yeah, so actually the idea all right. Now we may be this year really we're going to continue presentations like this, because we have not yet explored all the five cheese's space, but next year. Hopefully we want to you know, stop it. Maybe I'll MP is one good area to scope. This related issues, rated scope and so on. So.

E

A

Go for above yeah, so.

E

We have a buff next IETF or not.

A

Not nothin next year, not next IHF I, don't think so I mean we would like to have it. If you, you know, if you're ready, if you have drafts well.

E

Yes, so the day I've heard for 5g is 2020 and if you do the math yeah.

A

A lot of races, yeah, we slept obviously.

E

That's that's pretty ambitious.

A

12 years we have not even touched upon is IOT for 5g, the other one is via cooler, 5 scene, I. Think France, professor vtec mentioned a little bit. Those are issues maybe next meeting we can have some presentations and then you continue more of these current things we have and so on. Okay, yeah.

G

Like the restate Tom's question, when is five gang IP going to be a working group, because until it's a working group, other standards bodies won't take this work seriously, so my question is even higher level, maybe than Tom's was when will 3gpp take this work seriously? Yes, that's a good.

A

We have gsma yes ma person here. She can say something.

B

And not the 3gpp spokesperson, hi Natasha, greenie, gsma um I did speak with some people about this work, and they were the message I received was that everything below everything below IP they were dealing with in 3gpp and IP upwards, as they felt like, should be dealt here. I, don't know if that's gonna make people feel warm and fuzzy inside or not, and the act I did want to come to Mike. For another reason, I actually just don't know what this group wants to do.

B

I all the presentations are gray and all very relevant pieces of work. I've actually like to see a draft charter, even if it's a crap version, just a joke charter to know what the group, what group walk off right above yeah well I'm, not even just going for a bob. Let's just put some staff down on a piece page make.

D

Some children and then and.

B

Then we can argue about it on the list and then we can see if it's worth doing with with summer being where it is your prob. You probably are not going to have a botnet off next time, or at least not a working group forming one because I don't think anyone's going to approve it, but you could aim for March and I. Think that's probably a better time frame. So I think you're right about that. But let's start arguing about a charter because there's enough people here so okay.

A

Thanks Natasha: well, you busy will do ok.

I

Don romesco know he might be kicked out of here. Go ahead, all right done before I'm kicking being kicked out just my observation. It's from the presentations and the draft that I have read right now: I'm not sure that I know, if I'm, to write a charter of if I'm to do a ba victory exactly where it would be long. But I have seen like disparities, the pieces of works and maybe long in routing may belong in the internet area.

I

Some of them may be actually incremental work due to existing working group. So there is still much work to do and I believe that Beckett is perfectly right. We need to actually to do show activity on the mailing list right draft and especially comment an answer comment in order to kind of help crystallize and so our layer of the ideas here and understand what what what we, where we can add value. Eg I gave absolutely.

A

Thank you no way. Yeah.

D

You know y el amor here, 22 short comments. It's yeah! If you want to do something for 5 g's, not time force like the slideshow presentation, it's a time for defining what you want to do or what it is needed to do with in IDF and to define the scope of the work, because 3gpp will not wait for two years to have any solution. There's already working on the zia closing the architect network and the even even we have a pretty work on pre 5g work, ray-j.

A

A shake you well yeah. We have.

D

We have, we have also protocol safety.

A

D

A CSP for security, but for the protocol aspects in city level. We have already some work on going on separation between the control plane and say we.

A

Are following closely what's going on a trip, of course, the.

D

Point is that if you are not coming up with a in in a short time with a charter and a and a concrete work plan, I think it all. This discussion will be a useless because 3gpp will not wait for any solution.

A

So so so you're saying you should hurry up yeah absolutely so then, then then, do like a dancer. You.

D

Can sum up it like that? Do.

A

Like Dan said, okay, I guess: that's it folks, thank you to all the presenters and great presentations and then those who didn't write their names. Plz, like your name.

A