Internet Engineering Task Force IETF 115, 8 Nov 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF115-LISP-20221108-1630

Description

LISP meeting session at IETF115
2022/11/08 1630

https://datatracker.ietf.org/meeting/115/proceedings/

A

A

A

A

Another person.

A

A

B

C

I thought this was on the other side. Oh.

A

It's going back and forth yeah.

C

Yeah no I was next door, but then I thought this. Was there so I went there.

C

Nothing someone else: okay,.

A

Okay, I think we're gonna start.

A

A

Welcome everyone: you are the at the least working group meeting. This is not a jump in the past is my mistake: I left the itf114 just uploaded a new deck, but apparently the system doesn't update regularly uh anyway, um they're not well. You signed it everything we we speak today. Here is a contribution: okay, um the usual tips, if you didn't do it yet, please log in the mythical in the client tool. You know this is also automatically signing the blue sheets. Okay for people remote, please use the full client, okay kind of contact.

A

I love. To put this slide, just to remember remind to everyone to that. We are here to to speak technically and we have to respect each other each and every time.

A

uh The usual admin slide: we have some news apartment here. Apartment here is the new co-chair. Okay, I welcome her.

A

That Joel was here for the beginning of this session, because I actually have an award for him for a decade. Long working group chair- you cannot read here but as a typical ITF Style I have this for him. If you can read and on the back, can.

D

A

That I will give it to him as as soon as I can he did a wonderful job. uh We reach a good Milestone with.

E

The main specifications.

A

I will do it it's a good idea anyway, on the screen, you have the the usual pointers, the Julie prune, audio stream, mythical, as I said, please log in on the onsite and remote steam. Okay and everything is on the data tracker. Concerning the the material good news, we have eight new documents that have been published since 114., as I said. This is also the Merit of Joel that it brought some some work on that this is an important Milestone, because, finally, we can focus on other work.

A

Okay, we also have the the freedom to recharter and discuss on what we want to focus. Okay, this later on today, uh there are a few documents that are already in the pipeline or have the name encoding, hexagon pops up and young young is close. Actually, we should start to push that one. We will have today an update of a couple of those documents. Okay,.

A

And last slide from our side is the agenda, so we will have a Sharon giving us an update on hexagon. Then we have uh dino uh uh with with the chair, coordinate in the satellite networks. This isn't the right order. You ask it. Okay, we updated the agenda. We switched the the this to presentation, because Dean asked for is better for him. We have uh prakash with the scite external connectivity.

A

There is a last minute request from Alberto to give a short update on pops up. So after the prakash, he will take the mic for a couple of minutes and then all the time left to start discussing on rechartering. Okay.

A

So unless there are through the comments, uh I guess we can switch to Sharon. Okay.

A

Let me Sharon: are you online right, yeah.

F

G

A

G

That I apologize for the travel restriction.

A

G

A

You're loud and clear go ahead.

G

Okay, it's uh more NASDAQ related than uh covet related, but whatever all right so I would like to share with you um a few updates uh on this hexagon since Vienna uh two relates to the ad review and to relate more to Industry developments which are associated with the draft.

G

Next, please, okay! So is everybody's pretty much aware uh by now. The draft combines your special uh grid IDs and these eids for Mobility geolocation networking um just uh starting this effort. There were two choices: potential choices for grids, S2 and H3, and but since S2 is more geometry, mapping, GIS Atlas kind of oriented H3 was better Suited for dynamic real-time conditions.

G

What we attempted to do: real-time, Network, aggregation addressing because of its simple hierarchy and simple neighbor propagation, um as requested by Alvaro, we added an lcaf for cleaning and coding h3ids into eids, uh with the kind help of Dino and Alberto. So you can see, though they are now in the draft.

G

We also updated the references from the Beast to the actual rfcs, congratulations everyone.

G

uh Next, um the network is addressed and the packets are keyed by HQ IDs for values based on bdd taxonomy.

G

So it was requested by Alvaro that these two consortiums review- the draft uh h3o- was kind enough and really excited about the new networking use for their grids, but pointed out one issue which relates to hexagonal hierarchy.

G

um It's basically the fact that um hexagons are great for neighboring and they're like a circle but they're uh like not a circle which is good, um so they tile, but the in hierarchy. It may happen if the Delta between the detection resolution and the Eid resolution is too far.

G

The detection resolution may actually be outside the Eid area, um so uh that was pointed out, and the suggestion by H3 was uh not to go up the tree father father, to get from the detection ID to the EAD, but rather to ask for the center of the detection tile and then ask for the Eid of that Center, so that we get the correct uh Eid resolution, so that was fixed in the in the draft.

G

uh Bdd was asked to confirm that the enumeration of detections out under the draft and assumed uh more or less localization ability around 50 meters that can be Consolidated by eids, is reasonable for removing vehicle with a reasonable on-board Computing.

G

um They confirmed that uh offline. They said they really recommended that when we enumerate speed limit signs, uh we actually read them. What is the speed limit, because these signs keep changing a lot, so we did the ad that uh to the enumeration uh next so so far. This is for the draft related to the review and I I would like just to say, I apologize for the number of the revisions, uh as I discussed with Joel I have some challenges with English written English sentence structure.

G

So sometimes I write something it looks perfect and then it's just unintelligible so apologize for that um next as far as industry. um Oh, no! No up! Please go up. Yes! Thank you. So um the ACC has made public an interoperable multi-party test uh that detecting the free parking spaces in Tokyo, based on hexagon kddi, provided the access Oracle provided the edge compute locations.

G

The network locations between access and Edge were tied up by hosted rtrs next I provided the the AI car cam stack and the AI consolidation stack, but really this is unlikely that one window will provide both because there are many equipment, gear, vendors and multiple Mobility service providers. But that was good enough for the test. uh The RTR pickups uh were analyzed by kddis and were used to prove that uh the steering worked, meaning uh based on your location.

G

uh The uploads were steered to the location of the nexagon, where the hexagon is right now and it could be in either sites uh that subscription to nexagons uh for areas of interest uh uh worked and no matter where the hexagon was the signal free um multicast found the vehicle uh that was, there was subscription, subscription continuity and uh and Geo privacy, meaning no tracking for the based on the vehicle eids.

G

uh The pscs now will be extended to more detections, more vehicles and more special areas. Now in the US next, uh an additional POC that will be made public. uh It will dramatically increase the the Frog fragmentation aspect of the setup.

G

uh Cisco Wi-Fi private 5G breakouts will be used to upload And subscribe to a much higher distribution than Edge data, centers uh more like uh uh servers in Toyota gas stations or dealerships, and that would be a lot more fragmented, and this will push the lisp aspects much stronger than the initial PLC did, um but these reflect uh industry trends for green low cost and high capacity connected Mobility. It's a strong pressure, okay, next, okay.

G

So this brings us to additional industry related uh Trend, which we hope to prove through nexagons and can be generalized later of automatic digital twins. I mean hexagons is the twin of the street, but there is twins of the and and vehicles and the drivers and so on uh in in principles. Digital twins uh are are based on connected sensors um uh for compute objects, which reflect some kind of asset.

G

um So this has a little do with the lisp, as you can see in the drawing in the middle. It's just connected sensor to a digital object, uh but if we examine a a few differences between digital twins in a like a closed domain like a factory versus um an open field like a city, we see clearly why at least and can, if you have been to the buff yesterday, are clearly clearly valid.

G

um For example, uh in a factory image, I may have a twin of um a a an engine or a thousand engines, but in the city I may have uh in the day millions of engines and at night just a thousand engines. So that's the difference between office servers and hundreds of servers uh same for the street, uh a lot more busy a lot more detections during the day. So we need that kind of elastic steering.

G

um Also in a factory there will be a stable access uh in the city, there'll be multiple providers and breakouts. So that's very fluid and requires list of abilities um and as far as sensors like in a factory or a testing track, I will have fixed sensors for every segment of the track or a wind tunnel. In the city, the vehicles are the sensors. So one minute the vehicle is sensored for one segment.

G

The next minute is a sensor for another segment and that kind of Dynamics also lends itself to lisp, also geoprivacy and preventing Tracking not relevant in a factory very well in the field. To that we add the fragmentation we talked about and that fragmentation was brought up yesterday in the exactly uh yeah. It was brought up yesterday in the buff and it's very real uh and it has very strong, economical push.

G

uh So the fragmentation, which is supported by lisp uh nexagon, um allows uh geospatial twins to expand and occupy more Edge locations, uh so educations are considerably cheaper as far as gpus than Cloud.

G

um The reason is that they are priced at 35 cents per kilowatt per hour uh versus the cloud which is a lot more and, um and this has to do with there's no spine in the edge. There is no physical ability to do a spine, um there's, also uh no Monopoly, and um and in addition, uh that that's why the pricing is uh is very different.

G

um Also, as far as access the pricing of uh breakouts, uh whenever they're available in the middle of the city or are are priced per UE and not per gigabyte like the carrier So within the city and busy areas and busy uploads, we would use them a lot, and when we go on the interstate, we will use the uh the carrier. uh One more thing as far as green is uh Edge locations are starting to get powered and cooled by hydrogen and that's a lot cheaper than uh connecting to the grid. Okay.

G

Next, so if we look at the generalization of this approach, if we tie the fog fragmentation with lisp, we can we have significant savings uh compared to uh uh one Spectrum from carrier. uh One consistent Spectrum, which is very expensive um uh cbrs, is free uh and uh compared to compute on the cloud. The difference is: uh is the difference between economically viable and not economically viable for all kinds of reality, plus applications like an hexagram next.

G

Oh, this is a a quick example if I compare the upload of four dollars per gigabyte using uh continuous spectrums um for 1 000 frames per second, the frame is about 100 kilobytes, a detection frame- that's 260 terabytes per month. uh If I do it using one Spectrum, it will cost me a million dollars a month.

G

If I do it using um uh fragmented uh hotspots, it will cost me 72 000 a month, that's a big difference and as far as compute, the the most optimal numbers when all the Ducks are in a row, will cost you a hundred dollars for a million frames in the cloud that's 260 000 per month, but at 35 cents per kilowatt hour plus depreciation. It will only cost me ten thousand dollars per month in um The Edge. The only downside I have to manage my own elasticity.

G

My own uh roaming and I do that with lisp and I can get to these numbers. That's it.

A

A

Seems not so thank you Sharon. We we go to dinner next.

A

The jail coordinates yeah.

E

C

E

You, okay, I'm gonna, present uh the list. Geo prefixes next slide. So um this presentation this is an exact repeat of the presentation given November 2016.

E

um and it basically says there were no changes to the documents from zero two to zero fourteen. So all those doc all those revisions were just references and resetting the timer, so nothing has really changed in the document. Go ahead. Next.

E

um Okay, so the changes from the initial document to zero two is basically, um uh let's see uh changes to reflect the new encoding that was going to match Isis, ospf and bgp encodings of geocordinates.

E

Okay, basically, what's on the left is what the zero zero draft produced um and then we moved it to what's on the right and that's exactly what the other all four protocols do.

E

Now um we made altitude larger and we are allowing for um centimeter granularity and that's why the latitude is being measured in milliseconds latitude and longitude are being measured in um in milliseconds, okay and then there was a lot of commentary about geocordinates and privacy, so we put in a big section uh there and you could read that in comment, if you'd like, basically referring to BCP 160., okay next, so mylispers.net implementation implements it, and this is a use case that I was trying to to use.

E

And what's this showing is, these are entries mapping entries in the mapping system. This is we're looking at a um we're looking at a map server right here that is being configured in VPN, or instance, ID, 1000 and what's being accepted. There are distinguished names as the Eid, so San Jose, New, York, Paris, London, Tokyo, SJC and CDG, so they're cities and airports that are there. Okay, those things are being registered. So if you go down and look at the the um the bottom left hand side, this is an example of Paris.

E

This is the actual mapping. Now we call this mapping a Geo prefix, because it's actually defining a point in the center of a sphere. So building a geofence three-dimensional. So it's just not a circle on a 2d space, but three-dimensional I. Guess you can't go below earth, but you can certainly go up in in the sky, so you can decide so a Geo prefix is what I treat as just like a regular IP prefix. It has a point as you look down at the bottom there and a slash 100.

E

The one 100 defines the radius of the sphere in kilometers, okay and then, on the right hand, side you can actually register a geode point where CDG is the a Paris airport and its GPS coordinates. Are there uh um on the bottom in the.

C

E

Next, so what you might want to do- and my my implementation has um a Geo test sort of thing and you can ask the question: like is the point the geopoint CDG in the Geo prefix Paris right? So when that lookup is done, what what happens is the lookup is done on both of those eids? Those are distinguished named eids and if you look at the bottom, that's the result. The result says the Geo Point 49 blah blah blah 34 is CDG, and these are these.

E

Coordinates and then, if you look up the Eid prefix Paris, you get the prefix and then you can just do a GPS measurement and note is that 49 34 whatever within that slash 100 and the answer is, it is inside of it and it's actually 23 kilometers away from the center of the sphere. So that's a use case.

E

So you know what what's a practical use case for this is uh well. Maybe you have a geofence around San Francisco and you know, there's an epicenter for an earthquake. You may want to send packets to all the eids that are within 10 kilometers of that epicenter, and so that's something you can do. You could retrieve all the IP address Eid, so you can send, pin messages, ping messages or or TCP connections or whatever.

E

Next that's it, and we made a request for working group about a week ago, I guess or last.

A

Week, yes, this is running a call for adoption. So if you support this document, please go to the mailing list and say so other than that. If there are any questions.

C

um Just a couple comments: um it's good that you um meet the formats the same as ospfs and bgp. If you notice those drafts didn't go anywhere in those working groups, mostly because, if I remember, people were concerned about uh publishing location right of what a publishing location, oh yeah, right and and good.

B

C

People know say in the case of bgp that my competitor knows where my router is right right, and you know that, may uh you know create many many things. um So having said that, if the working group wants to work on this, uh you know I can't stop you. Of course. uh Just you know be conscious of that, that you know there's going to be a lot of questions around privacy, um which is my second point. You said uh BCP 160, there's another RC.

C

You want to look at I, think it's 69 73, if I'm looking at mine from here, okay but I'll- put it on the on the chat um that is called privacy considerations for internet protocols and what we probably want to do is besides or maybe instead I, don't know of what the security section says actually have a privacy considerations, so that when people look at this and see Geo whatever they're going to immediately jump there right and then everything is going to be explained there and how it's not going.

E

To escape and how so distinguish it from the security consideration section.

C

Yes, exactly I've seen some people make it a subsection or a completely different section or whatever just to to track the attention of people that want to look at this. So.

E

Can I comment on your first point? Yes, um uh I assume, are you done? You have yeah.

C

E

um What you register the mapping system about location can be static. Things like the example I used right, everybody knows where Charles de Gaulle Airport is and where Paris is, um should you assign geocoordinates to routers and and hosts and things that's a completely different matter? um If somebody decides to do that? Well, it's just like. Should you sell some a gun to somebody and that's.

C

E

Who pulls a trigger, but if the gun wasn't there? That's your point right, yeah.

C

Right exactly that, that's the point that that, um yes, there are many use cases. There are probably fine. um There are probably use cases that are not yeah.

E

C

Or that it depends on who used the uses them and who sees the location and what.

E

They could do with it.

C

Right so um just make.

E

Sure that now the mapping system can restrict map requests, so you could be so you hit using the ecdca draft, you can sign map, requests and, and then you can, you know exactly who is requesting it and they can be authorized.

C

Right, so all those things you want to put in the consideration so that uh that way you you limit, who can see this yeah and.

B

C

Know Etc just you know, General comment right thanks.

B

Hey um do you know one thing I wanted to say: is that, specifically, if we're talking about different kind of families of of what can be like, for example, Charles girl, there's no privacy on it, it's a building that could be a hospital. It can be, you know very specific um public places, so uh we were having a discussion before about possibly we could actually group those in very specific instances or very specific ways of identifying them and I. Think then you could actually have your drill privacy kind of Soul.

B

By saying these are the class of objects that are already known so there's no privacy, yeah.

E

And yeah we can use, um maybe instance IDs and say the all, the public ones. If you want to publish yourself in to the mapping system- and you want to be public and you don't care where your location is, then you register it. To instance ID 1000, where, if you want privacy, you do it to instance, ID 1001 and then that's severely restricted on who connects 1001.

B

Tell me about you know special emergency response. Those can be really good use cases where you actually want to see them. You know you want to know, and it might be that only certain people can actually look them up and have some restriction there. Yeah.

E

And if you want, you can obfuscate it or um uh aggregate it where you don't have to give the Geo point of where you are. You can give a Geo fence, that's 20, kilometers and you're somewhere in there, but they don't know exactly. Of course a bomb could take out the 20 kilometers, but knowing exactly where somebody is which room on which street that could be obfuscated a little bit yeah.

A

There is a question from Sharon.

G

uh Yeah I'm wondering do you know if we can combine uh Goa IDs with ephemeral and uh sort of handle the Privacy meaning keep changing. They can coordinate with the person just an address, which is right for a while.

E

You catch all the words it broke.

A

Up a little bit, can you repeat, can you repeat the shot on please.

G

Yes, I was wondering if it's possible, to combine geoeids with ephemeral eids so that they are good for a while.

E

Oh, you mean they're only they're ephemeral in terms of how long they exist in a mapping system.

G

Yeah and uh if you want to keep uh being known to be in a location, you have to keep uh generating new ones.

E

Absolutely you can you could send a map register once with a short timeout and when it times out the map server will just remove them so that that that will solve. We can solve that pretty easily with existing machinery.

E

A

E

And if you want, if you want short timeouts, but you and you want them to time out quickly, then you just send refreshes just like any other map register so or you can send deregisters as well from the the point who registers so yeah, we could we. We could certainly support that.

A

Okay, satellite.

E

Next, okay, so I'm just gonna um go over some high level goals, there's a little bit repeat from the last ietf, but um a list of overlay is something that can run over an IP delivery Network, which is we call the underlay. And today, when you run this overlay over the capital, I internet, we just use the capital I internet. However, it looks with independent of its interfaces and types of routers and everything. That's underneath. We just use it and we know it'll deliver packets to our logs okay.

E

um If a satellite Network can deliver IP packets, it's just another example of an underlay, and so a list overlay can run over that. If you wanted to exclusively run it over a constellation, it will do just fine. If and also if the satellite network is part of the capital I internet, then it should work. Just fine as well um point being here. Is that there's really nothing that has to change in the satellite Network to make lisp run over it?

E

So very much like lisp runs over the 3gpp network um by running it at the very edges or actually in the notes in the 3gpp network or the nodes that are in the uh in the icon, Network the airline Network.

E

um It could just it just works, because it's just different types of components that deliver IP packets. Next, please, okay! So how does it work? It's pretty straightforward! There's no Eid state in the satellite underlay right it only routes to Arlo, so encapsulated packets by lisp get delivered from our look to our look as long as the satellite Network can deliver IP packets.

E

um Then there should be no problem with this encapsulation note that the satellite network is unaware of all the ID State as well as any overlay, that's running over it. So it knows no difference of a TCP connection is going over it or lisp encapsulated packets and, of course, the underlay Network. Can transport ipv4 or IPv6 packets and can be dual stacked just like the um the capital I internet is too so we can use you can use either address type to make it work and a combination of both as well.

E

When path optimization in the underlay is available, the our local record could help the satellite Network route packets through it. So if you're, using a satellite network with lots of isl's inter-satellite links- and you want to select the path, if you seem to have more knowledge than the cell network, you can make suggestions, we've also talked about. Maybe satellite networks will be using GPS coordinates to find the down link, which satellite is the downlink to where you have to go.

E

So maybe if those GPS coordinates are in the mapping system, that's associated with the rlook of an XTR, then those that pair can be used together. That's something we could. We could add as a a use case uh in a paragraph in the satellite spec next.

E

So this is what it would look like. The top square is basically space and you see the satellite networks that are up there and there's lasers that connect them together in a vacuum, inner, the isls or the inner satellite links, and then we have up and down links that are usually run with RF type technology and all we're doing is we're defining um a new type of XTR called a GS, a ground station XTR, it's just a regular box that sits down there.

E

That's a lisp XTR and it just has an interface up to the satellite and, of course what you do is eids are attached to the GS XTR they come and go, and those are the ones that get registered with the Arlo. That's associated right here, so lisp runs in those boxes and the mapping system is on the ground next.

E

So what? What are the sort of features you get well? Edies can roam from gsxtrs to another and keep their connections up. So, just like an Eid can roam from uh once, like a VM can roam from one server to another. The eids are assigned to the VMS, it's just like a mobile device or anything else, and that's um uses all the Machinery. That's in the Eid Mobility draft right and if you want the zero packet loss things you can kind of assume where the Eid is going to move.

E

If it's moving, um you know across GS xtrs, and you know the direction you can use. Predictive, Arlo's and packets could be going to the places you're about to arrive at, and so you can get zero packet loss. That way, just like we did for the um other use cases.

E

um Gsx TRS can load split traffic across uh different R logs. They could be our looks of the satellite link or other links that the um GS XTR has. So you can actually load split between space and ground if you need to or use it as backup um and all packets that are encapsulated can use list crypto and therefore, if you want everything in space to be encrypted from XTR to XTR, that certainly is available.

E

Okay, now GS xtrs can use an alternate path and use most of the traffic engineering stuff that we have in the traffic engineering. Spec so, for example, that first bullet there that says an XTR may send a packet up to the satellite, that's in its aperture range and that satellite then typically will send it over a laser to another satellite. Maybe you want to reach something: half the circumference of the earth, and so we would go up over and then down.

E

That would be the the typical case when it's not doing the even simpler case where it just goes to one sat and then comes down. Okay, but for policy reasons you might want to have the satellite go up and come down to an RTR and I'll show you examples of the testing I've been doing. Why that's necessary, so you may go up to a satellite back down to the ground back up to satellite and then back down.

E

So that would be something that would be needed and and as we realize, that's something that might be needed, because maybe the connectivity is maybe there's congestion in the in the in the particular orbit that this satellite is in and it doesn't want to switch Orbits for whatever reason or there's congestion or there's lack of connectivity, because what we have now is we have um a bunch of systems on the ground that are either static or moving around using a network. That's totally mobile, all the time right.

E

The satellites are moving 2.7 um meters per second or like 17 000 miles an hour, so these things are not standing still and they're moving very fast. So if you try to do anything clever, like synchronization by the time you synchronize things when you're ready to use it they're going to already be out of date.

E

Okay, so what we can do is we can use the list to underlay Telemetry by using our load probing to actually measure the performance between xtrs on the ground, and we could actually find out how much variability we have in the satellite Network as well.

E

Okay and then the second to last, but GS xtrs can offer Eid multicast Services by doing head-end replication again, head and replication is there to offer a multicast service model to eids, even though that the underlay doesn't support native multicast, so it just does head and replication to unicast arlooks, where there's no reason why it can't do this over the satellite network.

E

If that's its only path, and also note that eids can talk to non-eids and the path may be either the encapsulated path, maybe over the satellite Network and then it comes down and then it gets decapsulated and sent to a non-eid um or it can go even farther. So all the inner working stuff just works as well, because there's no there's no special properties that make things difficult for an overlay to run on other than the satellite Network just has to provide IP address routing next.

E

Okay. So the changes from we wrote the zero Zoo draft early this year and presented it um in um in Philadelphia in June or to July, and um there were some comments about. um Should the mapping system be put in space? um Will it be? Will it be needed there because of better latency if, for some reason, list pass to be put in the constellation?

E

Well, at this moment in time, we see no reason that lisp has to be put in the constellation, but there was another common saying is what if a GS XTR only has a satellite link and has no terrestrial links?

E

Okay, will it be able to resolve things using the mapping system, and the answer is yes, because the mapping system is accessed via Arlo's or routable IP addresses and if the, if the satellite Network can reach that part of the network and and the services that are offered with most of the satellite networks today, is that you can reach the entire internet by just sending something up.

E

We said that um that the gsxtr should have no problem, and the mapping system on the ground should be able to serve all these different combinations, okay and then what we also added to the zero one. Spec was a bunch of test scenarios that that I'm about to um present to you now.

E

So I went out and I bought a starlink dish, I mounted at the top of the roof. The roof of my house and I started doing some list tests on it and when you look at this top diagram, there's a there's, satellites, they're, starlink satellites, there's about 3 500 of them now roaming in different orbits um in Leo, and when you buy a starlink kit, you get a dish, that's about the size of a pizza box, that's labeled dish there and you get a Wi-Fi router okay. So what I did is my lap.

E

My personal laptop is basically acting as the GS XTR running lisp. Okay, so I was able to do experiments. I connect my WI-Fi to the starlink, Wi-Fi router and I'm off and running. So these are all the test cases that I have tested. The top diagram is the is the case where it's GST xtrs are directly connected to the satellite network using non-nat. This is really important. Okay, this has not been tested because starlink router is a Nat device. The Wi-Fi router is a Nat device.

E

Okay, so you you it's hard to reach each other without doing the natural versal logic and lisp. Okay. Okay, if you look at the bottom half, this is how it works with Nat. So if you say, if you say my laptop is there on the left hand, side the only way I'm able to reach anybody else, is to encapsulate my packets to the RTR. So actually, when the packets are encapsulated to the RTR, they go to the rtr's Arlo. So the satellite network is delivering it to a Colo down where the list RTR is located.

E

It turns out the list bar TR is running in on Amazon AWS VM, so it actually, wherever the satellite, wherever starlink decides to downlink to whatever Colo I, think it uses a lot of Google infrastructure that the packet then is routed from Google to Amazon, to get to the lisp RTR, and then the list bar TR then can encapsulate to the GS XTR on the other side, which is also doing that traversal so that that has been tested next.

A

Slide, you know we should.

E

Speak up: okay, okay, um so this just shows the case where the laptop on the left hand, side is talking to a list of xcr, that's in a VM just doing encapsulation. So this is the case where the XTR is not directly connected to the satellite. We tested that.

E

um And then the top one was inner working so can can an Eid the laptop on the left hand, side talk to a non-list post and that's usually done through a lisp RTR, where we're using listnet, where the Eid is being translated to an hour load. So packets coming back can follow that path as well and so that's been tested and the bottom has been.

B

E

As well next, this is the case that hasn't been tested. This is where we'll have just the phone.

E

That's roaming, that's not running lists that will roam back and forth from a Wi-Fi or GS XTR, AP and I can do that um on my laptop as well, so we can have Eid one move, or in this case it's showing eid2 is moving from one one: GSX tr2 another one and um all the packets are being delivered by the satellite constellation and the TCP connections are staying up because the Eid address doesn't change, but just the our logs do so.

F

E

More testing of that's going to happen, so this is just a call for if um anybody wants to do, testing I need somebody far away from the bay area in California, so I can do some um testing and when the um satellite offerings do ISL, um it would be nice to do an ISL testing if I can, if I can make that happen or judge that it's happening.

E

So if there's any volunteers that want to join testing, that's fine, I'm, really I'm, even willing to offer to pay for your kids as long as you pay for the monthly costs for the service. So unless you can convince Elon to sell you free stuff, maybe if you're any of you live in Ukraine, maybe we could do some testing huh is right.

E

So here's um and we're at uh oh, so the question here is: um is this interesting work to make it um a working group document.

A

um First of all, there are any questions.

A

uh Lihan is in the queue.

E

A

I, don't know so I have a question in the meantime, maybe uh at the end of it the document in details. But my question is: does this document modify or extend the lisp uh specification in any way because it looked like just uh informational deployment document.

E

It's it's documenting the fact that lists can run over a satellite Network and it requires no changes to the architecture protocols. That's what it's documenting.

A

Okay um has been this discussed extensively on the mailing list, I.

E

Don't think it has.

A

That would be good if we discuss a little bit more about this document before going to the working group.

E

A

E

You want me to initiate a discussion of some.

A

Sort that would be great, for example, I, think, can.

F

A

Me, yes, we hear.

F

You now okay, so uh my question is: uh I saw that XTR and the eib bounding doesn't change right.

E

You want to go back to the um yeah on this on this slide yeah, so.

F

uh When the satellite link uh uh with Dish is changing, do you trigger any updates? Absolutely not. Okay, okay,.

E

Okay, good, basically, what happens in this diagram is eid2 is registered with the Arlo of the GS XTR um on the it's, the IP address of the wireless interface to the Wi-Fi router, when the IDT moves IED two moves. It's now registered with the IP address of the right hand, gsxtr's Wi-Fi interface and.

F

That's got it got it, so actually you uh actually push all work to IP layer to uh uh uh uh to advertise or connection between GS XTR to satellite.

E

uh No, it's in the mapping system, the the Eid to Arlo mappings are in the mapping system and the mapping system doesn't know how um the IP address are logs. It doesn't know if it's attached on an Ethernet, a Wi-Fi interface.

F

Yeah, that's what that's! What I mean means that? Because the satellite network is not aware.

E

That's right, just like just like my eids attached to my Wi-Fi router, that has Comcast service and if I talked to Padma and she has Comcast, you just change the satellite Network there to a terrestrial. You know um cable network and it works the same way because all list cares about our eids and where they're located on the network, not what's underneath the network and how the packets are being delivered.

F

Yeah got it. Thank you.

A

Okay, thank you. Let's move forward, we are running late, so now it's time from prakash I'm going to share this slides.

A

H

Hello, everyone- this is prakash Jain from Cisco Systems I'll, be presenting on the list of site external connectivity draft. It's been there uh for a while in the uh uh published for a while co-authors are Victor Moreno from Google and Sanjay Huda from Cisco. Next slide, please, at high level the draft addresses the Dynamic external connectivity for for a list site, and especially it is for the eids or the end point which are unknown or which are not, which are known but not registered.

H

To the mapping system means they can they can be external to the to the mapping system or external to the site, or they might be, might be in the site, but not registered. Yet uh so draft uh essentially uh suggests the mechanism how to register the proxy ETR for those endpoints and how to how to update those proxy TR dynamically at the ITR or ETR, wherever it is needed, and also how to request the PTR and also how to specify the PTR in the map replace.

H

Now one one thing to note is that, as part of this draft, we are not changing any of the procedures or any of the record formats which list standard specified. We are using them, as it is only the contents. How how the contents should be to address this dynamic, dynamic, external uh uh connectivity or or have the external connectivity dynamically changing now the? What draft suggesting is that Eid prefix is the distinguished name either configured or it can be agreed upon uh distinguished name as per the list name encoding draft.

H

Now the uh the pxtr or PTR should be specified as part of the r log set and those uh each locator in the unlock set can be or may may be encoded. As per the list. Vpn draft to add uh to cater the VPN environments means your uh your pxtr could be in a different world or different VM.

H

Additional information may be encoded as uh as a vendor specific alcaph type regarding the registering ETR or or the PTR itself. Anything if, if needed, that's a specific I mean it's very specific to the implementation. Next.

H

uh Similarly, the notification there's no change, as in the procedures or the record format, the Eid prefix can be distinct, will be a distinguished name and, as per the list name encoding draft uh uh ptrs will be will be included, will be as part of the r log set. Now. The only difference here in in the uh uh in the additional information is that now this information will be what map server or whoever is notifying, wants to communicate to the ITR or ETR.

H

As part of the notification awesome yeah means we are not really changing anything in the message type. So, whether the message been used for uh if the pub sub is used or whether the request replies, we use it's a.

E

Question or should I wait, go ahead. Okay, um so I understand that the message format's the same, but what's the reason you want to send a map notify.

H

Okay, so because it's a dynamic in the what we want to have is a dynamic adjustment of the PTR. If there's something changes in the ATR, it has to be updated at the ITR.

E

Change or the metrics are changing.

H

Anything changes at.

E

H

Given the routes.

E

So that functionality is documented in Pub sub, so it means when you register, when you register the distinguished name, you want to be able to set the end bit. So it's a subscribe request. Okay,.

B

E

H

It's not limited to it, but it can. It can be used. There's no notification can also be used, but the pub sub is also.

E

The answer should be yes,.

E

Because if you don't use it, then you don't know what the mapping would have changed. You gotta wait for timeouts.

A

But maybe let's finish the slides.

H

And we can discuss this also yeah. We can. We can discuss that more. We can talk about and talk.

H

Okay, okay, next one, please the map request, nothing is changing, except that if it's needed the winter specific Healthcare can be used to communicate the information regarding the ITR next one, please the main changes in the resolution in the map reply today when the eids are not known to the mapping system uh or if they are not registered to the mapping system. Negative map replies is sent in the negative map reply. The r log count is zero.

H

The the graph is specifying that, in that case, Arlo count must be non-zero, and it should include the PTR information as a part of the r log set or the locator set, and with the r log encoding can be used either way either way, as per the VPN draft or as a regular R log uh Eid prefix is same as the prefix uh in the negative map. Reply means it is a non-list hole as per the procedures as the standard procedure.

H

The the another thing is the TTL.

H

The draft is suggesting that the shorter TTL should be used because mapping can change, and that's the one of the your question was that how it will be updated because the the TTL we are specifying is this is the smaller one. Now it may not be immediate you're right that pop star will be more immediate, but it can be.

H

uh It can be taken care either by that uh additional information uh regarding the uh regarding the mapping itself can be communicated from the map server or whoever is sending the uh the map reply to the uh to the ITR next.

H

This is the example use case. It's not part of the draft, but uh it's here to show that this use case is using this, this draft to communicate the default uh PTR information uh to the ITR or within the list site.

H

Now this has been fully implemented and deployed in the field uh here, uh the Eid prefix has been used as a as a default DTR distinguished name, uh saying that hey this is the default path and our our logs there can be multiple PTR in dialogue as part of the locator set and those uh multiple locators can be in the redundancy or it can be in the load, balancing mode based on the priority and weight as in the existing uh uh lisp standard.

H

Now, the map request here is the uh is is so there are two type of map requests here involved. One is that if somebody wants the default ETR information before the packet arrives, then map request can be done using the default PTR distinguished name as an Eid and the uh and uh means with this. If there is an additional information needs to be communicated, can we be communicated using the winter specific type? Now?

H

Another map request is that when really traffic arrives at that time, when the traffic arrives, the map request goes as a regular map request. Now the reply will be a negative Maple in earlier case that replies the regular reply, because this the Eid has a distinguished name and the r lock is the PTR. Now when the actual traffic arise, because the Eid is the actual destination, the mapping system does not have that that uh that destination.

H

So the reply is the negative is kind of negative map reply, but with the modification of our love is non-zero, foreign.

E

It's a contradiction of terms. If you say you receive the negative map reply with a locator count of non-zero, but.

H

I'm saying that it's similar to the Eid, similar to the negative.

E

What's most important is what's the action value in that negative map reply.

H

So the uh that that.

A

H

Can we can call it like a regular map reply with the uh with the Eid prefix, similar to the Eid prefix of the negative reply or either way like, but to the point? This is not the uh this, not communicating any action values. It is communicating the ETR information. Wait.

A

Wait wait, wait, wait. You know we. We should actually wrap up because we are in the last meeting of the meeting. So do you mind going to.

H

Quickly, I think this is the probably the last slide I have and next is just requesting for the working group adoption and any comments and questions or can continue.

A

To add something.

B

Hey um Praveen, I I do have a few comments, so I went through your draft actually and I came up with a case where um I want to know exactly how you're doing this is that if you go in the case where you have a negative map reply right, it's going to give you a range which is larger. If you look at 9301, it's going to give you a range which is larger, for example, if you're asking 4.1.1.1 it's going to mostly respond.

H

To you, that's the point: I was trying to make that Eid prefix, though we should not call it negatively. I agree with. What do you know is saying, but the the reason I'm saying it looks similar to the negative ambulator because of that Eid yeah prefix whole prefix calculation. It will be in in in that range, which is the larger range than the requesting preference.

B

Yeah exactly so exactly this is where I think we need to be extremely careful, because at this point before you may have the negative map reply. But you still don't have anybody attracting the traffic here you're going to have the Petr attract the traffic of everybody else who are not yet up, because anything that's going to come up now are going to be actually attracted by that. Just like you were saying is that you have a non-hole.

B

What if somebody comes up with 4.2.1 Point, there's no way that they're going to actually have a more you know, a shortest prefix is going to match it, but that.

H

Is not going to change any of the thing which we do with the negative epileptic. So, let's say, let's say today is default forward. Right. Look at the action code map server can say so here. The purpose of this is that we wish we fall forward but encapsulated to the board.

H

That is the purpose. So, instead of taking the native routing path, we want to take the overlay Parts towards the Border and that's the that's why it is functionality wise. It's not going to change anything but other. Instead of taking the native path, it is taking the capsule.

E

Capsule, yes, yes,.

A

We might want to continue yeah, okay,.

H

So one of the purpose for for that is that why is it deliberately done? Because we assume that when you are on the Axis or the edge side, your box is smaller, so box may not have all the information, but when you are on the border, you have maybe Pub sub already there, which already know about this.

A

The discussion is interesting, I'm, sorry, but yeah. We are really over time. Thank you. Prakash um continue the discussion on the mailing list and we will reach that point about the working group adoption now it seems there are a few points to be clarified and uh and that we will probably usual. Thank you very much thanks Alberto 30 seconds without sliding. If you want to really to have a quick note on pops out.

A

46., you are asking too much go ahead. Yeah.

D

So, first of all, thank you Alvaro for the review. Sorry, it took us a while to get back to you. We have done that uh and then I send an email to the list about moving, perhaps up to propose standard. So if you have any opinions on that, please go to the list, because this is something that has come over and over and yeah. We want to to see if we are going to move it to our proposition.

D

A

Okay, so please go to the mailing list and explain to your opinion on this point. Thank you very much for participating. Sorry if we are at three minutes over time enjoy the social event later. Okay,.

A

We had, we don't have time about the shelter. We will take it to the mailing list that site to do to make progress uh before 116., but now we are over time. So officially we go to the mailing list.

E

A

We can we can, we can usually notice yeah.

B

Thanks everyone.

A