Internet Engineering Task Force 94, 5 Nov 2015

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF94-INTAREA-20151105-1740.webm

Description

INTAREA meeting session at IETF94
2015/11/05 1740

A

Hi everyone so welcome to the interior meeting. We have a very packed agenda that got modified at the last minute, so we're going to be running very tight on time.

B

So this is the no 12v concert for like few seconds and just so that we showed it. So you know what this is so go ahead thanks. So thanks for en ang in may, for taking minutes and Miguel to be the javascript.

B

So for the document status like we have two active working group documents, so one of them was like GRE ipv6. It just got published as RFC 76 76 and we have the tunnels draft, which is like just got updated in the last meeting cycle and it's undergoing discussions on the list and we have like two new working group drafts. That's the hostname practice and the ad hoc wireless forms. So these are adopted like two weeks ago and it's a pretty packed agenda.

B

So we have like some working two drafts and we'd start off, and then we have some communications from the I Triple E, but multicast, and we have a bunch of individual graphs that are requesting adoption or like at least discussion on the on the working group. And finally, we have another presentation from I triply 80 to 15, for the logical link, control, yeah.

B

Okay, so- and this is like kind of like lightning talks, but the chats are going to show the slide and we get people to talk for a minute. So this is like a draft concerning the broadcast stuff and Rolf, and the back mike is going to talk for a minute about it, and this is just elevator pitch right.

C

So you probably seen my name is Rolf. You probably seen the emails going back and forth on the attendees mailing list on the IETF at IDF mailing list about an experiment. We've got currently doing on the wireless network.

C

That's me, and so we did a previous experiment on a large wireless network and we saw that basically devices today broadcast your identity. Your contacts to other people constantly so links between people, hostnames, persistent IDs, hi, message frequency. So we know when you're on the on the network- and we pretty much know very fast when you're off the network. These kind of things are today very easily to deduct from all the broadcast protocols that are out there and it's not just mdns and ll M&R and at least kind of protocols, but its its application protocols.

C

It's Dropbox 45 BitTorrent sync, and it's increasing its more and more it's how to encrypt, because it's broadcast is multicast and it's easy to collect, because you can just passively sit on the network. Everybody on the network is receiving that at these kind of messages. So we written a draft about this and we think maybe this working group is a good home for this and I would like people to read it and comment on it, and maybe, if you have pointers for related work, that would be great as well. Thank.

B

You all so this is the preceding, so you can see the name of the brass and comment on this. Thank you and another pitch is like for the dynamic GRE tunnel mechanism, so this is presented at the last meeting, and so it hasn't got like much comments and the authors would like to try to solicit more comments on it. So please read the draft, so I'll send a pointer to the list.

B

Soliciting reviews for this and for Rawls strap that we just talked about, but this is just a mechanism for auto configuring, GRE tunnels and to dynamically manage them. Yeah.

D

A

I just say one thing: please use this Mike, it's easier for me to describe. People are standing here. So I can read the tags perfect.

B

Thank you, and so for the agenda, like any questions of the agenda.

B

Going once twice, thrice, ok mark.

B

Second, just like attention.

E

Hello, everyone um Joe touch couldn't be here, so you get to listen to me. um This is a document that he and I started Oh about ten years ago. um It went through various iterations, it SAT around sort of dead for a while, and then it was resurrected. I'm gonna go over just in case. You haven't, read it or have it read it recently, I'm gonna go through real, quick, the main idea, um and then maybe, if we have time for questions lots of questions, um we all love to build tunnels.

E

This document is about tunnels. We build them where, when in where we need them, often for very specific purposes. One of the reasons why we have so many of them. They have a wide variety of uses. Here you see some load balancing.

E

We got dr. black there. um What goes in must come out of a tunnel and oftentimes. You don't have any idea what's going on to your packets, while they're in transit inside the tunnel, which is most of both the feature and a bug, depending on what you're trying to do, ah but.

E

Tunnels must be designed in such a way that we can fit what we want to put in put in them very carefully. Sorry, I hate not having a lapel mic ah mark pick it up pick up the mic, cuz I've moved too much. I can hold it now, I feel like I'm doing karaoke Oh.

E

All right tunnels must be designed to fit what we want to put in them, and that means when they're, when carrying IP, we actually look like a link layer attached to a router. It's a very important point in the sort of fundamental tunnel model described in the beginning of this document when we're using IP. So if we're encapsulated inside IP or somehow put what we want to tunnel inside an IP packet, we're looking like a network layer and we're also looking like a host because we're the source and the sink for those IP packets.

E

So we're sort of this half half thing and by the way this is recursive and if you want to be able to make it recursive forever, you have to be very pure in your emulation of the link layer above or the network, acting as a network layer between your endpoints. So basically, it kind of looks like this um there's three links here. One of them is well three named links here. One is your traditional link and again it looks like a router here.

E

This is your tunnel link and again we have a link here, and this is all the stuff that you're hiding um it might be just routers here or it might be. Firewalls and God knows what you can tunnel through DNS if you'd like. So that's the basic concept which is the beginning of this document.

E

This document went through a major rewrite since the last time it since, since ITF 91, we actually published this in in July, and so basically it describes this model at the top as sort of here's where we start- and we start with it, looking like a link so that we can fit it in anywhere else and it can recurse forever.

E

And then we understand that, based upon the operational environment based upon hardware characteristics based upon the service of the tunnel itself, delay the traffic that's going to be running over that tunnel, it might break down from that basic fundamental concept and we try to go through and document how. The that's happened with a variety of different types of tunnels, so that this can kind of be an instructive aid for protocol designers operators etc. um So we had the complete revision.

E

Lately, there's been a lot of discussion, particularly about him to you, and Fred will be discussing I suppose we'll bring that up when he talks about him to you later and in the meeting, um and this is our to-do list, so we've got a variety of things that have come up on list. Thank you for the discussion. Please, please bring more in particular on the list, so that Joe can hear it he's been doing most of the heavy lifting on this document.

E

uh I've been now doing pretty pictures and discussing it, and maybe even on the next before the next IDT. If I'll do some real live edits on it, but anything on list is helpful because joe gets to see it. We have some questions about how this relates to existing rfcs. This is an informational document from picasa vola that we sort of expound upon do we need to update 2473 here, etc and there's various clarifications that are in the queue.

E

This is jose note that revs are off cycle actually to encourage on list discussion, because again he typically can't make it to an ietf meeting, because he doesn't have funding for that. It's just not in his sort of day-to-day things that he does. If there's time for questions, but.

B

F

David black is transferred area working group chaired TSU WG, so I've had discussion offline with Joe and want to pass along the conclusion. There are various kinds of tunnels. This draft is about IP in IP tunnels, which in particular is why you heard mark talk about it. Looking like a router, there is another class of tunnels, there's characterized by there's UDP in the shim. These have caused me more adventure like carrot than I care to admit over the near past. In T 2g. We have a draft on UDP guidelines. It's 5405 bits.

F

We are putting material on use of UDP encapsulation for arbitrary purposes. Whatever you want to care, if you want to carry UDP or IP lessons learned and what you should not do will be in this draft. It is a companion to the tunnels, draft you're, carrying IP and IP. This is draft is important if you're using UDP 5405. This is important. I would encourage people to please review both them. We think that before, or five is as close to done, let.

E

Me ask you a question: um actually the document uh tries to capture IP foo IP, ok, for example, in lttp tunnel or a GRE tunnel, or pick your favorite yeah, listen tunnel pick your favorite. Whatever that's got between IP has something between the IP and the IP. Well, and it now doesn't mean it tries to also include all the UDP and considerations, but I think that there's a little bit more of a marriage between these two documents that maybe you're thinking I believe.

F

There's a marriage between ins from the reason I'm standing at the mic and in here and I blew the place where I think we're drawing line is as follows: IP info and IP is the folks this draft, when foo is UDP. There are certain things about UDP that you must understand. For example, you get to go battle. The checksum in ipv6 networks, that's going to be over in the tsv WG draft. In addition, everybody and their brother seems to encapsulate you name it in UDP over IP, the joke.

F

The plenary about the internet running over UDP was not as funny as it should. If a.

E

Great, we need.

B

To make a reference on another.

E

B

Yet Lucy thanks.

G

Lee Howard um so I'm not sure that, on the list of things to do is actually finish. Writing it because they're an awful lot of placeholders still to be written. Yes, do you like to help I'm still I'm thinking about it, but my concern is I'm still unclear on exactly who the audience for this document is I. Think you said.

E

It very quickly, but it says it in their protocol designers, network operators. There.

G

May be something.

E

G

What kind of protocol designer a.

E

Tunnel protocol designer.

G

Tunnel product so so consider for hey. You want to tell something over tamil yeah, but something about something else. Somebody's.

E

With all the new tunneling protocol, because they like to do it.

G

I've been done before, and it's.

E

G

Part of my problem is it's a fairly long document already, without all the words being in it and I feel like they're, probably it's it's generally 325 thick considerations repeated over and over and over, and there might be a more concise way to write. Okay, excellent thanks.

E

That's good feedback. We.

B

Lying cristiana.

E

B

I cut the line we.

E

Reorganized it once maybe, we need to now adjust a little bit more, a.

H

Fred template from boy I just wanted to say that the UDP is one example. What could be included in fool yep, so the document that David's mentioning if they try to say the same things I think they keep the Virgin dangerous ways. So I just want to be sure that you DP and food is one of the examples of food that could be a desolation. That's.

E

A good point: I think that the tunnel generic aspects, hopefully would be captured here and the UDP would be more specific to UDP, like oh there's, a check something as David mentioned. David.

F

Brock, what more, what Marx said if you're running IP in in in for over IP, then this draft is what is is was required, including telling you how the inner I enter IP behaves considerations specifically UDP and considerations that would apply when the inner payload is not IP go in the twg draft and that's the operating uh that's right. Now the operating a a demarcation uh that uh that Joanne I worked at work out offline. That's that's! What we're going to try to carry forward.

F

There is a goal of avoiding duplication, but there's a pile of stuff about UDP. That's got to be said. These people are running into it repeatedly thanks.

B

Dave excellent: let's see Megan quickie yea.

A

Lucy, album Huawei I think this will document the make a very good the backward compatibility for this architecture perspective, seeing ya the tunnel form alinka for this overlay Network. However, moving forward, there's kind of an application doing this kind of metadata, all those things you know make the link to how ammo you know the property there I.

A

No, how you know if we are insist, there's a tunnel just acting exactly as traditional link, women missing something we may need to think about. You know this linger maybe have a more properly. You know for like a linker location or those things they expressed in the metadata for form.

E

Okay, um well, links have been around for a long time and they look. They have some common properties to them, especially when you're talking about how they express themselves in the adjacent layer of IP. um If we want to expand the notion of what a link is then I suppose this thing becomes obsolete, but I don't think that's absolutely necessary. Hopefully we can find ways to incorporate all this innovation that still fits into the basic models and I think that that's possible. Sometimes it's just a matter of semantics to get that right.

B

Thank you very much mark okay. So people please review this document and send reviews on the list like Joe's, not at the meeting, so it'll be really good. If everything happens on this and the authors like are planning to do, updates not on the idea of meeting cycles, but between cycles to increase that. So thank you very much mark and thanks Joe.

I

Thank you are good afternoon. My name is Dan Harkins I'm, not Dorothy Stanley. She couldn't be here so I'm presenting this in her stead of sorry for the weird format of this. This is a I Triple E document, and this is the format that everybody has to use. What we don't get to do. Clever pictures like mark this is uh that's all right.

I

So the reason I'm doing this talk is because those a meeting between some I Triple, E and I t effers and the ITF errs we're asking why multicast is so bad on wireless and wanted know a little bit more about how how multi cast works. So this is the agenda I'll just plow through here in less than 10 minutes, so just an overview of dot11, so it operates.

I

An unlicensed frequencies doesn't have a clue, civ claim on the on the medium anybody can use it and consequently, baby monitors and microwave ovens and things like that will create noise in the medium and cause collisions cause packet losses and pack up, drops I'm sure Evans experienced this you're walking around your phone or something you walk behind this steel-reinforced pillar and hear your Wi-Fi signal drop. So there's a lot of a lot of ways that you can get interference and pack a loss.

I

Nothing to note in that 11 unicast traffic is acknowledged, but multicast traffic is not so the way multicast works in a. Why in Wireless is that the AP? Basically, it's tens, the wired segment, so any broadcast or multicast traffic on the wired network gets a sent out over the wireless network and it's transmitted at what's called the Rose most robust MCS. The MCS is the modulation and coding scheme, and so there's a lot of stuff in the modulation coding scheme.

I

In addition to the modulation coding, there's this probably spatial streams you can have and the channel width and all this other nonsense. But for the point of this talk, just think of the MCS is the rate that the packet is sent out at as I mentioned, the multicast traffic is not acknowledged, it's not retransmitted and this MCS, the MCS rate it sent since wireless data at the lowest possible rate, because it doesn't there may be other.

I

There may be many clients attached to the access point and they have to send it out at the rate that every single client has an opportunity to receive. So the problem is, you could be having some stations they're transmitting at you know, 600 megabits and along comes some multicast traffic and they got to go and send it down it.

I

You know six megabits and what happens is the lower the rate, the larger the frame, so it occupies more of the air as it's being sent out, and that means that there's less air for all the other frames. This is it's a contention based medium, so we you get to get the air. You say: I need it for this long and if you're sending frames that are very low right, you need it for even longer. So it basically just pushes out all the other traffic there are has been.

I

Some work to excuse me improve this I'll. Get to that in those two things in a second one. More thing to note is power. Save stations to prolong our battery life can like do little. Catnaps they'll turn off their their Wi-Fi radio go to sleep, and then they turn the Wi-Fi.

I

Radio back on to say is three anything for me and they do this on a certain period, and what happens is because that happens if, if at least once one client on a connected to an access point is doing power, save that means that all multicast traffic has to be buffered, so that that client will have an opportunity to get the multicast traffic so, instead of just sending it out when it gets it, the AP is gonna have to buffer some of this stuff up and wait for a certain period at which this one single station wakes up to to get his traffic, so basically a multicast on wireless kind of sucks, so things that we can do to address that are, for instance, proxy arp.

I

This is just a nice way to keep a lot of broadcast multicast frames or at least broadcast frames off of the off of the air, since all the stations are connected to the access point, the access point knows exactly who's there with their mac addresses and they can stay, put their IP addresses and do proxy arp and thereby reduce the number of times. You've got to send these large frames at a low rate over the air, and this is the part of the standard that specifies proxy arp and it does exactly what you imagine.

I

The AP is just acting on behalf of the the station so for ipv6 it does the same thing with neighbor discovery. This is the text again he's based in again it's exactly what you'd imagine. The AP is just acting on behalf of the station and are responding to these neighbor neighbor advertisement, message of neighbor solicitation messages and basically a setting these things out on behalf of the client. So these are two of the things that have been developed at eleven to improve the multicast experience.

I

One of these things is directed multicast, and what this is the the client can ask the AP I want you to send all multicast traffic to me as a unicast, and this works nicely because again, unicast is acknowledged and it's not going to be buffered up for one station goes to sleep. The problem, though, is this doesn't scale all that? Well, you can imagine if you've got you know a couple hundred stations associated to an access point and they all want to say send me multicast traffic is unicast.

I

That's going to require a whole lot of route transmissions for a single packet, which is not very good, but for you know, for a home network, or something like that that this would be a completely reasonable way of addressing some of the problems with multi cast another question.

J

I

Another enhancement is a group cast with retries. What this does is it allows for acknowledgement of multicast traffic and it actually allows for block acknowledgments, so a bunch of multicast traffic and be sent, and you can get a block act back saying you know. I've we've received this. There are some other additions to what's called GCR. One of them is the unsolicited retransmission. So basically, what happens is the AP if it wants to, you can send multiple copies of the same frame to ensure that it's received by everybody.

I

Another thing is: there's a GCR with a periodic sending of multicast. So what happens is if there's power saved clients? Gcr, oh with the it's SP. What's SP series something anyway. ah If there, if a client goes into power-save, basically, the client is dictating when traffic will be sent out on the medium, but with GC RSP. What happens is the AP says I, don't care when you're going to sleep I'm sitting out multicast traffic in in this period?

I

So if you want to get multicast traffic, you got to wake up and you got to get it.

I

So there is some problems with cheese block acts, because, basically, when you send the access point sends out a whole bunch of traffic.

I

He then has to send out a block act request and wait to get a block act response before he decides to send any more traffic, so that entails some latency, because traffic is going to get buffered up while he's waiting for his black acknowledgement. There's some work being done right now in another task group in about 11 to enhance this using multi-user MIMO. That's basically, I have multiple radios in or in the in the clients, and so they can basically send the request in the downlink and get the the Act back in the uplink.

I

So it's things can happen without what is much latency I guess, one of the things that the IETF is were asking. The I, Triple E was a what's been implemented and, sadly, there's nothing. We can do to make people implement features in the standard any more than you can make people implement in an RFC proxy arp and proxy neighbor discovery have been implemented widely for enterprise-grade systems. But you know, if you go by an access point, that your local frys or you know, whatever your local country's equivalent of a electronic store.

I

Is you buy something like AP? It's probably not going to have proxy arp in it. I'm not aware of any GCR or dms implementations currently being sold. I have seen a GC our implementation, but that was a university implementation. I, don't believe it's a for sale product. So thank.

B

You Dan so edit first and then followed by Lorenzo lines, cut.

K

Eric, not mine, so I didn't know. You had brought proxy arp and Andy in the spec, so proxy Andy. Does it worry about things like secure neighbor discovery? No! Ok.

K

That sounds like a bug, but I mean it's not like secure neighbor discovery is widely implemented deployed, but but the fact that we want to be able to add more options to neighbor discovery and sundays is one set of those and if you sort of blindly proxy respond, you're going to lose whatever functionality we might add in the future and that that makes it harder to be of all these protocols.

K

Just something to be aware of I, don't know how to feed that into the process, but I.

I

Will make sure that message gets received at the burka, I Triple E they're reading next week, so.

K

The other one is since of these optimizations that people have looked at to what extent are they looking at sort of? I have a video stream not being multicast, which is different than the sort of discovery traffic that we see the neighbor discovery on mdns and whatever so.

I

That's what this uh this GCR thing was. It was the way Todd 11 works as they were. They form a group too, and they make an amendment to amend the standard and the group that amended the poochies yarin was dedicated towards video streaming, so the the GCR implementation I saw they were doing a high def video and it it looked great, so they have been doing that. But again, unfortunately, I don't know anybody that actually sells product. That does it does this, but.

K

But even if they did, could you use it for like multicast DNS or something where it's not as stream on things, it's just that you want to somehow make the delivery of those packets multicast DNS is another another.

I

Problem and you know, networks meltdown when people are doing you know, especially like universities and stuff, like that, every every fall we get calls from university saying our network melted down. It's like what he came with five new iPads. You know thanks friend,.

J

Zone um to the perspective of somebody trying to make trade-offs between unicast and multicast um a couple of questions. You said that you know unicast might be 100 times slower. Sorry, balti class might be a hundred times slower than unicast. Perhaps I don't.

I

Remember saying 106.

J

26 verses, 690.

I

Mystics, oh, oh, that was just an example of if you've got a station, that's a when they connect their they're at different MCS rates, and you want to optimize for speed, and so, if you're, if you can get a higher MCS right, if you could do 600 megabits you're going to try and get 600 megawatts. But the problem is when a multicast frame that arrives well, it doesn't necessarily need to be 600, Ackley 100 or it could be 54 mega bits. The issue is that multicast c-can we sent out at the lowest possible MCS, so.

J

To let let me know it's.

I

Not nice, in that it's not really a hundred to one. It's basically just know the lowest common denominator, which sucks soo.

J

Know so suppose, suppose, for example, that I had a multicast packet, it was slow and I wanted to send it to a lot of people. How many people does it need to go to before it's worth sending it out multicast as opposed to unicast, because if it's, for example, if it's simply a hundred times slower, then it have more than a hundred devices on the link. Multicast is better than unicast, very, very simple.

J

What sort of you rule of thumb could I use as a protocol designer to like figure out what it might be 50 x, 100, x, 20 x? I.

I

Have no idea I mean that's. That would require some putzing around.

J

I

J

A good answer, because I don't want to chew up the unit cost. You know if I've got 20,000 hosts on my on my link and I want to send an RA to all of them. Unicast, that's gonna, chew up the channel for a very long time, because they're all acknowledged and if I send out on a multicast are a you know, might as well 10 10 or 10 of them in the same bandwidth, and it's actually saving on channel time.

J

Even if they're not acknowledged the probability of one of them getting through is very, very high. So, okay, so okay, we don't know so. Another question is like um what is the efficiency rate for a lot of this stuff is for big packets. What is the like? Assuming look like a lot of the control protocol, stuff that in ipv6 uses that uses multicast of small packets? How is this sort of?

J

How is this math different between small packets and big packets like, for example, if there's a big channel set up time and a small packet, then maybe multicast is a little bit more efficient than you might think. So.

I

Gcr can- and it put it, can aggregate a lot of multicast packets into one aggregate frame and send that in the future. When.

J

It's implemented yes, but in today's networks.

I

Today's networks, they're just gonna, be sent at the lowest MCS and all right, but.

J

I

J

I

J

But like if the preamble is big compared to the packet size right that that's what I'm trying to get at, like, for example, of 1,500 by packet versus 30 by packet, how's, the math different.

B

Thanks Alonzo Christian makeup on quickly we're out of time, Thanks.

D

Yeah I mean basically what we're hearing there is that there's Christian with Emma what we are you guys that there is multicast and there are laws of physics and the laws of physics are telling us that vertical sucks. Well, it's not.

L

I

D

Again, if I avi it might be, but well we have usage patterns that are that data usage pattern. There is a usage pattern. You wish you send multicast, because the data is going to be consumed by everybody we are going to is going to receive it.

D

That's relation commit patterns in that case, while we can be better what you said about that, what's the size of the green really makes sense at all, and we have the usage pattern is which we send a question to everybody, because we want to reach somebody and that's a pattern that actually sucks and, and that pattern is where is something we should take on in the interior, because I mean we could achieve the same result which I don't know DHT or something like that and which will be using unicast? Okay.

D

So that's something we might want to consider for stuff like nds of like a dns, and should we shift it to something like a a unique aspect on, but.

B

I'm sorry the line is cut pologize, but one.

I

Thing like safe: if you take anything from this presentation, it's not a hundred x, you could have people speaking at a gigabit and you send out a frame at six megabits or you could have everybody speaking at six megabits and you send a frame out at six megabits right. So the the point is that multi gasket sent out at the lowest possible rate and if you got a whole bunch of people speaking at high rates, then that's going to be proportionally. Suckier, okay,.

A

Michael Lorenzen here I went to the I Triple E about this. There is a non working group mailing list you being formed to handle this. Please join this list. Let's get it worked out, what di trick they should do and what they attempt to do. There is a lot of things to do here and to consider we're not going to solve here in the deadly. Please will be created and will be announced in the interior and all the other relevant groups.

A

I know there's work in the 600 also related that solving some of these issues, and so so we really want to take a look from both. I triple e, an ietf side, their solutions, there's what has been specified, what has been implemented, what has been deployed- and it seems like there's a lot of education to be done. So please do join that list if you're interested in this topic so.

B

We'll send this out on a day or two thanks.

H

Okay, so this is fred temple from boeing. I'm going to go fast through this they're going to be cases when tunnels have to do some fragmentation, maybe just a little bit, maybe some more extensive fragmentation and there are three different ways of doing fragmentation.

H

We all know about fragmenting the tunneled packet, which is alternative one, in which case you have oh shoot. I.

H

Guess I don't have a laser pointer, in which case the encapsulating headers appear on each packet. The inner IP header appears only in the first packet and any shim headers in between the inner and the outer packet appear only in the first packet.

H

But here we have some alternative one issues, things like the ipid being too short, and things like that that make it owners in some such situations you cannot have alternative to, in which case you tunnel, you fragment the inner packet. When that happens, you have the encapsulation header appearing in each fragment, and the only problem with this is that it's only allowed for ipv4 alternative 3 is what's called tunnel fragmentation with the tunnel fragmentation.

H

You have a fragment header, not inside the outer IP header, not inside the inner IP header, but inside of the tunnel shim the encapsulation xem, so the shim had or will have identification and more fragment offset values instead of using the IP header at the inner or the outer packet, and so this avoids all the issues identified for alternatives one and to allow specifications of a minimum reassembly buffer large enough to resemble a 1500 byte packet, which is a kind of a magic number in the internet.

H

Today, and there are two drafts- the first proposes adding a fragment header to the GRE header. The second proposes adding a fragment header to the gue header. Both of these are examples of tunnel fragmentation. The original concept was actually discussed in RFC 2764 and, as Mark Townsley mentioned to me, it's also done in pwf tunneling, that's head.

K

So I would go through this quickly, but the whole purpose is standing up here is to get you guys to set mail on the mailing list. So since most of you are reading mail anyhow, you might as well go do that now AB so because this is out for the question is whether we should adopt this as a work of your document and I'm going to try to give you some motivation if I can figure out the direction of these arrows. So so so. Why should we care about this this issue?

K

Well, so there are these technologies that are being used, which I will explain briefly, that sort of partitions the layer 2 domain in interesting ways from the perspective of IT and and this is being used in some deployments when you have multiple routers, so that could be with vertebrates, RP or just sort of two routers that are on the link and and I think that, so there are some hard hard problems in this space and even though these technologies, some of them, are not that widely deployed are they their deployment might be limited.

K

It might be that we will run into more issues here, because people will want more redundancy out of the network. There is something that applies both in the ipv4 and ipv6 space. So let me illustrate what one of the key issues is in this this space and if you didn't catch it on a title slide, it shows up to some extent in dsl and in cable they tend to have a simpler topology. The most complex one is what private viewings.

K

So you have some form of topology hear that, from the perspective of IP, looks like a single sub. Not so IP thinks that this is one subnet. It could be before and or v6 right, but then it's been partitioned and in this case there's two routers that are connected to this. So you have some setup. You know some sort of hosts that can talk within their communities at layer 2.

K

Those guys can forward packets at em, but- and you have some other ones that are you know single host it's attached to those, but all these hosts actually see that they're connected to that subnet.

K

So when people want to deploy this and actually want to have some control of our host parking to host end up, deploying it with some prox Europe set up in neurology and when there's a single router, this is not very complicated, but in this example, there is like a dual router. So you know be in this case. Host 3 sends out an arc across four hosts for, and it thinks it's because it's things that directly connected well, it doesn't go directly at layer 2, it gets dropped on the floor by the VLAN filtering.

K

It gets delivered to both router one and router to they've been set up to proxy arp. This thing and the the host might get to our perp lies back right, but then it can actually send packets in the pockets, make it to the router the roddick in and have policies to decide whether it's going to let this packet through or not, or you know, filter it out based on something but the.

K

But but you end up with these risky behaviors right and, and so the packet is coming from the outside of the network and it's being forwarded on to the link and that router doesn't have an entry in its artcaffe or neighbor discovery cash. It ends up sending in ARP requests. Well that is sent down.

K

They exactly sent down to host to host one on host to as well, but it basically gets sent up flooded. Everybody sees it, but rather to seize it as well, and rather too has been configured to proxy arp for that address right. So now you could actually have that body one and rather two ends up sending that packet back and forth to each other, and there is some description I found in some obscure email from a while back where people said.

K

Well, if you do this stuff, it seems like people who are making I don't know if that was in the context of hsrp setting it up, so that if the packet ARP request came from rado, one rather two should not procure up respond, but this is not written down anywhere where people can actually find it. So so that's sort of a key thing that can actually break in these networks and that's the overall motivation for you to go, read this document and say whether they should be working. Your document or not thanks.

B

Eric, so there is an eruption column going on the list like so it ended like just before the idea, so I'll probably extend the call so but feel free to respond on the list. If you think this should become of a group item thanks, Walker.

M

Hello, everyone I'm workhard I'm, going Jim from Holly and because also we were to not to come here so I have her to choose a plantation. Oh, this plantation is about a young model, oh gee, I dunno. So the motivation, definitions, first date, definitions, the data model for gie completion and amanda manager and young presents a very popular choice for completion and the moment asking what's the address in revealing 0. Firstly, we define the young deal model based on oxy 278 forth and defines the young data model based on oxy to 80 90.

M

It's mainly for the key and sequence new number of this of the genus kelitha finder and the divine young data model based on ave c 2-6, 7-6 6, which for ipv6 and the ab c 2 futaba each what's not addressed in the revision 0. The frosting is the winter Priory primary features and not in this group and in future sinker is that the winter king could be supported at we could be supported the window. Priority features by augment. The second is the data information which will be added in the next revision revision when the Yamato.

M

What a what a the content of the Yamato pasta is: a supported delivery protocol configuration a second, there is a path MTU, an evil entity, support and further been teaching anybody see Bo and that checksum enable/disable, another key enable/disable anarchy, video completion and a support of the sequence. Number enable/disable here is the hierarchy of the Yamato theta.

M

What about next steps? I think you're faster? It gets the comments and the seeker contributions and collaborations and intro groups the adoption. Now we are open collaboration and welcome all comments. Thank.

B

You very much again so like this work is here because there's not a home for this, like the GRE extension stuff, so I think we will have to discuss the eighties like on how to proceed with this there's like 10 on off or two off its ok, but there's like lot of documents coming through, so we probably had to discuss the deities like how to go forward to this. Thank you. Thanks.

B

N

Okay, so this is another yang model and in the routing area working group we were told to bring it here, both the GRE model, and this one is more specific about IP over ipv4 tunnel. But you know we're told to bring both here, there's a actually a separate GRE tunnel that my co-workers have written.

N

So the point of this presentation is actually to solicit feedback about how you want to view tunnels, both IP tunnels and GRE tunnels, the IP tunnels I'm talking about our the ipv4 over ipv4 ipv6, over ipv4, auto ipv6 over ipv4 manual.

N

Ok, so the IP tunnel actually came from merged rats and we had sweet we're having a discussion about how again how you want to look at tunnels right now. There are two ways of looking at tunnels: one is sort of tight and the other is not so the the one on the right is or left this it's separate list.

N

Each list has a only consist of a single type of tunnels and the other one is obviously a single tunnel that, with all kinds, a single list with all kinds of tunnels, so their consequences of doing things like doing things different differently. I'll show you that, and there are also in implementations of doing things differently.

N

The first difference is the naming of the tunnels, because the separate list is tight, you can you get. You can have a you know the first type of tunneling I guess it's hard to see, but this is ipv4 in ipv4 it that is named tunnel one, and here you also have ipv6 ipv4, that's made tunnel one, whereas you can't do this in this other organization.

N

The the other consequence and other consequence is that when you want to change, there's actually a line right there, but it's when you want to change, for example, an ipv4 and ipv6 in ipv6 tunnel. What you have to do is explicitly remove this tunnel and then create that the the second tunnel second type type of tunnel, um whereas again with the other organization, you kind of the back and implicitly does that way.

N

So I like I mentioned earlier. This is very simple. When you look at the drafts, you'll see that they're very similar, the the IP tunnels and the GRE tunnels. The yang models are very similar and so.

N

The question to you is: how do you want to look at tunnels? Do you want to consolidate everything into one, and or do you want too tight or whatever, and that's it um thanks.

B

A

B

O

Per question yeah, you may want to look at how the IP tunnel lib works, where the management area already tackled make sure actually in turn it off, but they already tackled this problem there, where you have one generic model and then extensions that are on a tunnel protocol by tunnel protocol specific type, and so you put all the generic stuff in one place and you extend another ones, and so my recommendation would be try to follow that as close as possible. Given that that's already a standard, okay.

N

That we, that was discussed among the off all the authors involved and so sure.

B

Thank you Thank singler, so like sent a mail to the list like get the discussion on there, so we don't know if we can take this work, but we can be the discussion forum for this I. Think it's like what doing it again. Thanks.

B

P

Hello, I'm Charlie perkinson, I'm here to present some information about a new effort and I Triple E 80 2015.

P

Basically, the idea is that we wanted to provide a logical link, control interface, definition for attitude up 15 dot 4, and this is relevant to several efforts in the ITF and subbed and the earlier this week, um I Tripoli IETF coordination committee. That was suggested that she should be presented in this group int area to head to a wider audience, instead of just, for instance, sixth ish.

P

So, basically, what has happened is that this year earlier in the spring, there was an interest group forum for the purpose of seeing weather related to that 15 would like to provide. They make an effort to make this LLC in this interest. Group in July indeed stated they voted to form a study group, and the purpose of the study group is to create a project definition in justification.

P

So it can be. The work can actually be chartered. So that's that's in process. The study group is meeting having a second meeting next next week. Our first meeting was in September at the interim eating and next week is a trip later to wireless plenary meeting.

P

So since we have the group- and since it's of interest in IETF Here I am um we want to discuss the goals and try to find out what kind of feedback we can bring back for discussion next week, I presented this information just last time in sick station, so it has relevant comments from them as well. So what's the goal of the study group?

P

Well, the goal is to try to create the the project and the project idea is to take the current to specifications for attitude up 15 24 and provide this link control mechanism, because 15 double is a very minimal.

P

It has there's a lot of functions left out that are expected to be carried out by higher level protocols.

P

So, as a result, since those functions are not specified, some people have found it to be more difficult to use and then it also seems to encourage proprietary solutions, but in particular it's also the case that the group's an ietf need to make use of some functions like what would be in the LLC. And so that's that's why we want to make sure that I TF is able to benefit from this effort. So, in addition to the other functions are within 80 2015, there's some layer to routing and KMP for security key management.

P

So we want to head this LLC but make available to higher level protocols all these sorts of functions. Yes, I need to have my little handy, dandy, switcheroo here. So this is a picture given to me by Pascal, Pascal, d'hubert and sort of shows the situation we have down here. Oops, not that one.

P

We have here several kinds of mac protocols and numerous five protocols, and this LLC should sit on top of that and make all of the functions that are necessary, including, for instance, adaptation for ipv4 and b6 available as such as standardized LLC. We have six tish here and there's some sort of best effort protocols and deterministic. Sixth ish is an example of deterministic Minerva, low energy solutions and so on, and also non IP. So from the standpoint of 802 dot, 15, there's a number of potential customers and nevertheless, ITF is a very important potential customer.

P

So this is the this is sort of the the map and here's some, the initial list that we developed a few weeks ago in our first meeting about the kind of functions that need to be done right away.

P

So one is, it would be something like an ether type to allow dispatch to our level protocols, and we want to have this make visible the attitude up. 15, 9 and 10 on layer. 2 routing provide the link control for that and also to allow extensibility in the future so that you can define a function by using the features of this LLC without having to do it separately for every different mac and five protocol regulatory configuration.

P

If you'd like to be able to avoid having to do separate solutions for every mac and five diff, no matter which, in other words, take care of the regulatory considerations within this this layer and then also well. We have obviously going on protocol extensions from other organizations in particular idea.

P

So that's pretty much the quick description of the effort here I put in super fine print that you can go. Look at a longer presentation and you can see all this stuff from the LLC group by going mentor is a document auditory for all the well for a whole lot of groups, but judge me in particular 802 wireless and so our repository. If you go to this, but URL you'll, see that you can pick out the group and if you pick out, LLC you'll see our documents for that next week.

P

We're meeting- and we have already the project requests in the feasibility report so though so we'll be refined and then, if all goes well in January, we'll look at that again and submit that for consideration by the executive committee which meets again in March. So this is really the first time when there could be, you might say, an officially chartered task group to make this LLC, but that doesn't mean you will just sit around waiting until then. I think that the time is now to to do work on this.

P

The subject in particular in 66 top, is already have a lot, a lot of definition of work, and- and so we expect to take some involvement in reviewing that as well. What else oak? So we have a lot of work to do soon and we don't have to wait till 2017, but the way it goes with a 802 wireless is we have many many comments, usually on the the document and the specification, and they will all have to be resolved, and so we'll see what the sort of process that's followed to do that.

P

So, within a few years, we'll have a document specification that can be published. In the meantime, we have a lot of work that we will start to do. Questions.

K

Arecont mic so I didn't see fragmentation on the list so and and I thought 15 dot. 4 has a larger small maximum frame size. So is this not something that's part of the LLC or will.

P

There is already specification for fragmentation with within 80 2015 the so, yes, we might have a LLC entry point for that, but it would just be a matter of directing to that other specification. Okay,.

K

So the other thing that wasn't clear to me was maybe that's the picture, but whether I beat when you do ipv6, whether that would run using 6lowpan or whether you would run ipv6, be able to run ipv6 directly over the LLC, because the picture had Lee for ipv4 and 600 planet. Didn't how ipv6 well.

P

So the answer that question I can I I think it makes obvious sense that we would provide that yes, but it's not on this slide and I, don't like I'm, not speaking for the group, I mean I'm, presenting information about what I've seen so far. So yes, I, believe that should be a priority item.

K

P

Bring that feedback to them. Okay, thanks.

A

Ralph oops.

P

A

B

Off the mic doesn't look, I. Think yeah thanks.

G

Thanks for bringing this to us Charlie to give us an opportunity to anticipate, what's going on? Oh the very first slide, you said something along the lines of oh okay: use of higher level protocols without changes, no I'm. Sorry, it's the bullet up from that make it easy easy to use as 802 11, 8023, Wi-Fi and ethernet. Does that mean the the underlying network characteristics? The the sort of network characteristics provided to IP will look just like Wi-Fi and ethernet no.

P

It doesn't mean it, and actually this is a an editing error by me. The intention. Basically, what I should have wanted to express was that already the what people are able to use ATT, 911 and 802 dots 3, without changing, depending upon the underlying characteristics, for instance, of 80 2011, and we want to have they wrote about 15 dot for share that kind.

O

P

Hiding of the lower level features.

G

Okay, is that explained in more detail in the the presentations because there's a whole? Does it in different ways? You could do that, but so, for example, will this LLC if I have a mesh of 80 to 15 4 nodes with this LLC? Well, that all looked like a single link and I and I can I can essentially consider get from any device on that mesh to any other device in one hop I? Don't it's not forwarding through the inner I p forwarding through the intermediate devices.

P

You're asking about the LLC entry point, for instance, layer to routing yeah.

G

That would look to me like making an 80 254 mesh, look just like a Wi-Fi mesh or a fat yellow, cable in the ceiling.

G

Jenna wonder if that's that's the the idea or or something different and without going into the technical details, your work and I find out about that to give that feedback to the study group next week. So.

B

Charlie, do you understand the question or pardon? Is the question clear to you I.

P

Believe I understand the question you so.

B

Can you so I think like crowds question, it's like it's explained in more detail so like can you look at the slides and send a pointer to ralphs on the list about like okay? If it's answer Sandler like because there's a longer version of the slight said that he pointed to I, think I'll sue question is: is it being explained somewhere, but.

P

That the longer set of slides is not answering your question and I know what's in there and so the exact. um Obviously we don't have the LLC yet because we haven't actually started working on, but we had some places to start from and the specific answer to your question. First, as you might want to say, you want to have a mesh that is associated with the pan ID or something like that, and we don't have that information yet.

P

But that's the sort of thing that we do need to provide whether or not it fits in the initial list of priority. So I can't tell you.

B

Okay, thanks thanks are 12 thanks, charlie, and thank you everybody, especially the presenters who kept to the time. So thanks a lot for letting everybody speak thanks.

L