Internet Engineering Task Force 103, 8 Nov 2018

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF103-6TISCH-20181108-1610

Description

6TISCH meeting session at IETF103
2018/11/08 1610

https://datatracker.ietf.org/meeting/103/proceedings/

A

Hello, kattiline testing.

B

What am I hearing yeah.

A

I can hear well hey.

B

Super super cool. It's time, let's get started, okay, so.

C

As I was sharing that just before that's why we're just starting right on time? This is the sixth ish working group meeting.

C

Before this, allow me to remind you the well which applies to this meeting as it applies to any IETF meeting. If you know of any IPR, you have to disclose it where it's yours or not. Please read the slide. If you have any questions, ask your favorite, the ATO working chair as a reminder. We are taking minutes, the meeting is being recorded and the president is logged. Taking minutes, dominic has agreed to be a note-taker. Thank you. So much can we have a second note taker in the room. Somebody volunteers.

D

Sorry for the link, it's obviously when those three right yeah.

C

George, will you be a note-taker?

C

Well, you might be in a second ago when you might packing up the unique jabber scribe. It's not was the meet ago. There's really not much to do anybody mache. Could you be in JavaScript? Thank you.

C

So yeah, administrative, oh yeah, I, have it right here. So this is the link to the etherpad. The same link that I said my email saying that you have an e in the the agenda remote. Their participation is possible through over media call for the JavaScript I recommend you just connect over me. Deco mailing list meeting materials will be published there. All the sides are online, I, just updated the security slides that militias are sent. Otherwise all the sides are have been up for a number of hours.

C

Now this is the jannah' that we have prepared. We have a hundred and I think 105 out of the hundred ninety minutes scheduled. So we have a little of buffer time here. Let's try and let's try and finish 15 minutes really, if that's possible, so we're into the first element here in turn and say this blue sheets. Can we get.

B

C

Please put your name note: well describes agenda I'll go to region of the bastion right right away. We'll start by some introduction will then go into the chartered items. Pascal will give an update about the 60s architectural document. The goal of this presentation is to prepare for work who Blasco we put the goal here. So it's very clear, charlie lisanna will then go through the six top protocol.

C

Drafts- although this has been published a couple of days ago as an RFC, will then go through the big section about the MSF, the minimal schedule functions. Simo will introduce the work that we've done on the update, and then we have a two presenters that will talk about its performance, so we have- thankfully yes Luigi, yes Utica, who talked about a simulation campaign that we're conducting on this on this MSF, as well as if they wouldn't talk about the experiment with Henry that we're that is ongoing.

C

The goal here is to gather data from or input from, those different campaigns in order to make the draft better. After that, we'll go on to the minimal security draft. That militia will be presenting I understand you pushed on oh eight just this morning and so you'll be presenting oh seven and and clearly higher than goe. Please don't assume everybody has read it and then time permitting will have a presentation by single. Do you know about a new draft on robust scheduling, a proposal to make it more robust against jammer attacks?

C

Another very quickly about talking about the fragmentation, design team, and then we have a little bit of time for any other business before you get started. Is there anybody raising some issue, something you want to see added removed changed in this agenda this slide at that slide. Please raise those issues now.

C

Okay, here nothing so I propose that we that we get started with with our agenda.

D

So since IGF magnitude, there has been a number of good news for this working group. Obviously the biggest news is that we have RFC 8480, which was published so many thanks to their third chef each him tomorrow.

D

Many thanks- and this is actually a great achievement for his working group- another achievement which is linked to the work in this working group, which we actually kind of push to to 6lu, because that's where the work be known, but it was identified by the 60 architecture. Is there FCAT 505 to be because it's still in the earth 48 queue waiting for some people in this room to do something about it?

D

But but yesterday's this RFC will be published very very soon and hopefully, before next week. I will discuss it at a to to the Chilean for applicability in Wi-Fi.

D

There is also work which I started in an American stylization body, which is called OD di. O TVA is working on industrial automation. They have this command industrial protocol CIP that works of ERISA that an IP and they are considering applying 64 their low-speed automation and so I pick a transforming CIP assistance.

D

Today, which is TCP base, and possibly large payload in the frames trying to reduce the payload a little bit to make it more sixties, friendly and looking at what it means to have UDP go up as opposed to TCP and don't so they are doing this work. I, don't I've heard about the number of experimentation in some a number of companies on 60-ish and so far, they're pretty happy with it and then last but not least, we have interesting news from spread.

D

So spread is a standard which, which builds a stack, which is kind of similar to six dish. They have 6lowpan, they used a 15 for mac and phy underneath, but they are designed for used in in the house. It's home automation thing, and now they are looking at whether they could expand their scope into commercial buildings which, which requires a whole new set of scalability and routing quality, etc, which they have to prove that the current version of thread is capable of any of that.

D

I have in mind that if they they cannot need something which is a bit more scalable and reliable, they could converge a bit more to all six, but we'll see all right. I've started this discussion with the pods right I'm, not intending thread officially so, but but have at the discussion already to start with. So these are basically the news since ITF one or two, which really sounds is compared to the first year or two at six. We are really nearing the end where I actually did we deliver something?

D

And people start using it and including it, which is great news for this working world.

D

There's also work. As you know, six dishes is as much a group that that built an architecture and pushes work elsewhere, as it is a place where we do some actual RFC, so we do actually Isis, but we we push at least as much work outside to complete the architecture, and so a number of our drives are progressing and progressing. Well, so we've got 1/2 to retain and have to bear fragmentation which are now both draft ETS. So the the fragmentation which is part of our architecture is no work.

D

Group document at six load, maybe I'll, discover each it's a bundle of three drafts. One of them, like I, mentioned the first one is going RFC, eighty five or five and the other two are nearing World Cup. Let's go I mean backbone. Router will do the rest call now and a PNG is very close to there. There is this new interesting work as well, which is Chinese in less coal, now deadline.

D

So so all this is really getting your completion. So that's basically the feeling you should get from about everything we are doing. We are nearing completion at roll. We've pushed a number of interesting topics, and one of them is the doubt projection which enables the the route actually push some rods inside the network. We've got the unaware leave document which enables a leaf which is not ripple aware, which doesn't know our repair operates. To use six leper neighbor discovery to ask routing services ripple rotting services from from the first Excel.

D

All that means that, in six dish now we can have nodes which don't participate to repo but still get the packets rot. It's a ripple, and last but not least, we've got this rough Richardson a normal priority, which part of the primes we have discovered the building logic. Six networks is when you've got this node, which wants to join a network and our multiple networks around which network should I join.

D

What kind of criteria should I find in the beacon, so I can find not only the right network but possibly also the best future parent in that network. So this is what is this work is going and then we've got minimal security which we'll discuss at length, I, guess Monisha, so I won't say much, but it's nearing completion as well. With this dependency, we have own custom, underscore: okay, that's pretty much. It.

D

Okay, so news from the architecture: it's it's a white whale, it's basically the document which sustained and explains the whole work that we are doing so this document has been kept alive and, following what six dish has been doing across all those years, and now that we are reaching completion of everything we wanted to do in there. The discussion for today is whether we were ready to publish it so that there are two goals in the architectural document and, like I, said with the underlying idea that it should follow the work of the working group.

D

So it's also an introduction for everybody joining the working group. Where are we? Where are we going? What do we want to do? Well, the architecture was supposed to help people figure that out with one stop document, you have to two steps. The first step is the eye level architecture that raft ROMs asked us to build at some point. Well reviewed this document we almost made last call and wanted to publish several rounds of it. Ralph's told us not to do that to wait for the completion of the working group to publish the document.

D

He also told us that you wanted to isolate the eye level architecture from the more refined components and how they work and what they do, which is what we did at the time. So the document is still structure. This way you get the section three, which is the eye level architecture, and then so some people may adjust up there and some other people may want to go deeper into what are the RFC's that we use? How are they do? They? Are they position with regards to one another?

D

How does the whole system work in find out details? So we have this section for that. Does that as well.

D

So so this slides were made for draft 15, which was published last month, but as it goes, Thomas just made me a great review of the document, so I just published 16 before this, this working group, so Tom I, should look at the tips. You will find basically what you found.

D

So we have a 16 I did publish that 16 because since I had in mind to go and and do the one who pressed code on the document, I thought, let's do it with Thomas video included I supposed to find it the second time so so I did 16 as I clean up Russian of 16 or 15, so the because 60s focused and the distribute the operation. Obviously, the architecture has more details on the distributed operation. We still have some words on the centralized operation, which is enabled by the Mac, but we we don't.

D

We didn't work in this working group at all on the centralized operation and on we mentioned PCs, etc, but we never say how the traffic engineering flows work. So it's obviously much higher level. It's just position, but not that this was what we wanted to do. This was the Charter we had so because we are nearing completion and all the components are now in place.

D

My suggestion is to go for walk of Lascaux now, so we we are done and we we don't over overlay with the work group last call of minimum security or hi myself right. So so, basically, we have those three main elements: I myself in more security and architecture and and we don't want to have a same window of time for the for the various last call. We want to make them in order so suggestion, if just dependent discussion and my staff, that is to do the rest call for architectural now.

D

So we do a myself and then security and that's pretty much. It.

C

So I had a question. Sorry I was answering an email, you might have a cell, so you expect there to be little changes to MSF and minimal security that we can publish this now or what do you want to do where you want to wait for before? Basically,.

D

C

D

A different windows of time false way: okay, not just a matter of the group, to decide an order. This one is ready. It might not be the one that wanted to publish first I'll go press call first, but the other ones are not fully ready, went. Why don't we do that? Do the work go for that one? Well, the authors of the other two are completing and polishing their work. The point is for the level of details that the architectural needs.

D

We know enough about MSF and the minimal Kyoichi to describe it at that level in this pack, so I think the spike is okay. Obviously, the authors of the to draft can check the text and give us feedback as part of the last call, but I don't expect that either minimo or MSF will change to match that whatever we have here after the Resco is challenged right, so we can do it in any order. That's why we see yet following.

E

Up question from George from him to Atlanta from Boston from Thomas. What, if in new drafts or new ideas, will come to this to this sixty tomorrow, so you're going? Are we considered to integrate in this architecture, will be like separate I mean.

D

E

The detector should have like all the components of the 16 groups.

D

Well, yeah, we had a story. We wanted the full story, which was a mini-mall story if you like, and we think that our story complete complete. Okay, now we can write a second book. You know a fairly new story, but but we had in mind a certain story for which we did the number of charter item we and we have the story completed. So we we don't want to reopen that story, maybe to another story, but that we opened that one.

D

Okay, somebody wants to work on sixty sugar food. Maybe it's called six dish, maybe something else, but that's a different story. That's interesting! We want that to happen. It's not the 60 Chef Kittichai anymore, so.

C

What I would recommend this is Thomas? What I would recommend is that the authors of a particular MSF and minimal security read the draft and provide some feedback, some feedback, making sure that the text is you know, describes their resolution well and maybe also the you know, RFC 8180, just just to make sure that that it represents what what their drafts are. Sure.

D

Mean everybody during the worker Brasco is supposed to yeah yeah, but but we.

C

Can we can we can formally ask those people to do this and then we'll have three reviews coming in and- and you know it will lower the risk of this being out of you know out of sync, which I don't think it is. But okay.

D

C

Exactly what what do we want to do with terminology? Do we do we push this at the same time as the architects draft? Okay, make sense.

D

Any command subjection any comments yeah. So, let's see at the end of the meeting in which other we do things right, but but basically this one and that terminology are ready for oppress call.

C

Before you move.

F

Yeah I'm serious Krishnan, so one of the things is like I was thinking, is based on the IEA, see statement on support document stuff. Like do you think you can consolidate some of these documents like terminology? Can we put it along with something else? For example, the architecture I don't know, but just think about it if it makes sense or not. If.

D

That helps you. Why not there's no traction at all right, so we do it just because of the informational reference. So if the terminology is part of the estava tract because architecture Oh important my god, we did just that a little bit, but we real. So this architecture is very like the definite architecture and they were surprised that you called for a standard track document and he explained it's not because you get most in there. It's because it has this exposure or to otherwise deals and I said. Well, that's exactly the same thing.

D

We'd get the same exposure with I Triple E. So if he justifies and is correct that he should be still track, then this document should also be started track. So I changed it to start a track. But what did your confirmation? That's fine? So um so will.

F

You put this Thomas or I. Think.

D

F

Just listen spend some time put it in the Shepherd right up boy, I think it needs to be signed, let's track, and that's enough like so it's pretty much the cherished view of the work. No just tell the working group to as well yeah and see if there's anybody has issues with it but I'm. Finally, that person you okay, so let's write down it is for me to.

C

Shepherd document and can.

D

Now you have a neat here.

C

One second and your recommendations to is to roll in terminology within this draft, and so we get one like.

F

You know would somebody need the terminology without the architecture, that's how I think about it. Right, yes, separate, if not put them together, yeah.

C

I think they're both like a wrappers around the work and they're both contexts. They have this own function. You.

D

C

Both at the same time, right thanks, listen.

D

Actually, you will have internal pointers and supposed to extol. The neat I wanted to point out. Is that actually you have my name on it, but there are like 10 authors, and so we decided to throw the rules, like you know, put the contributors in the end and editor on top I would be very happy to put everybody at the front page. But you have to tell me that no- and the other thing is Thomas is a co-author I'm.

F

Not gonna tell you that you.

D

Are texts long ago so you're still on the author list and that's why long ago, when we did the first attempt to work with us co-listing, we asked Shwetha Bhandari to be our Shepherd. So we don't have this this duplication of hat. So you have to sort that out. That's.

F

Fine, if you want me to get another shepherd, that's fine too, like but I can talk to strata officially.

D

She's still, the shipper number.

F

Of one of the documents on.

D

The architecture yeah well, the other one- is just like a collection of words right so.

F

That's fine so like tell me when to go I.

D

Mean you want to shape our data if Suresh doesn't have a problem with you being slightly co-author, then then perfectly field I haven't.

F

Yeah well he's not the editor. I, don't have an issue personally like.

F

Right and and I really want somebody like you know, who's like did Jane all in this, like you know, unless there's like a real conflict of interest, that's really the reason why somebody doesn't want to be Shepherd on the document, the authoring, but he's not. If he's not active in there really shouldn't matter so I don't see a coffin.

C

Thank you. Anybody wants to raise something before we move on about this document.

C

Okay, let's hear nothing: let's move on to.

C

So Charlie can you.

G

H

C

H

C

Us can use.

I

C

Yes, I have not anything works great.

B

C

Okay, your your on Java, we hear and see you, okay,.

I

Great so hi everybody I'm gonna present the latest steps that we had on this sixth of protocol. Can you go to the next slide yeah, so the status is that it was already published as RFC the 8480 on the fifth of November. So that's a very good news for us. We with it in the last three months, some reviews and we follow the indications by the RFC, editor and Ayana editors.

I

I will detail, maybe for me the most important change that we had to do, but I want to thank especially to Suresh for following up with this, to Pascal and Thomas as well to follow up and also the editor Linnaean and the young editor, a man that that did a really good job. On this, the changes there were mainly mineral fixes, typos and some grammatical things, but then there was a discussion that took some time to solve.

I

It was related to section 3 dot 4.7, where there's a sentence that refers to response to return code in a three step transaction and it wasn't clear whether this risk return code was applying to the requester or to the receiver of the of the transaction. So we had to rephrase this a couple of times, so we we end up with the following text. You see here, and maybe here is a little bit out of context. But if you read this section you will find it now very clear, I, guess and I think that's it.

I

So we are now at the end of this draft. I hope it's complete. Now.

C

Yeah Thank You Jerry. He did not order. This would be published in time for, for a man coughs, that's great any and any questions about this before anyone.

B

C

B

Thanks Jerry, so I suppose I was playing with the we're.

C

Going to move on to a large section about about MSF, you know a drive threatened by I think a militia. Sorry C, Mon Diego. It's it's a it's a large! It's a three-step process because I.

C

Don't know, don't you start earrings, you know a introduction by C mode. We can talk about the changes that were made to the set and we'll hear about H I think they would be evaluation of the performance of this of the solution. So Seymour, if you can, if you do hear us yeah I, can hear you. Can you hear me perfect yet perfectly yeah, yet the slides are up they're yours.

G

Okay, great yeah thanks so since the last ATF in Montreal, so the draft was adopted. So we had this MSL version, zero, published the end of August and we just published a new update, MSF version, one just before the cutoff for for this one meeting and what's new in that a new version is basically no new content, but instead what we've done it that we've collected all the open issues that we have from the sixties mailing list and from oh.

G

If you go to next slide, where you can go to next fact, yeah thanks so from the sixties, meaningless and and or gates of internal github issues, and we have one of them into a new section in appendix of annex II, which is a temporary appendix with all the pending elements. Next slide Thanks.

G

So we have categorized the pending issues as security performance in editorial, so editorial is of. You are obviously quite minor, so it's just like terminology, etc performance. We have another number of performance, open issues, and many of them actually came from what you're going to hear next from the action and tank phase.

G

So the evaluation in simulation experiments, so we we are going to work more on them, I'm not going to detail the performance issues here then we run unless there is a request, and then we also have security issues in particular, one which we'll be talking a little bit about. Maybe here I think it's a sensitive, actually sensitive issue. So it's at joint time because we have the autonomous service.

G

Now there is a possible attack where a pledge, a malicious pledge, could drain resources of a joint proxy because, with video to normal cells, nodes have to add new cells whenever I have new enables, and so, if your malicious you can have, you could have lots of malicious pledges and you put them out there after to drain my sources of JT's.

G

So that is something that needs to be discussed more Halloween from an attack like one obvious solution could be to have the drain traffic not go in here to normal stands, we're going to be minimal cell instead, but that will come as one with the performance degradation so next slide. Please.

C

So, in summary, you know sorry, so this is almost one clarification, so the issue is that the join proxies are in sending enhance beacons, making the pledges. Where do they exist? The pledges have their MAC address. They calculate the.

C

How do you call them rendezvous cells and to the Germans, the autonomous cells to which the village is listening, and so they can mean a denial of service attack by just sending a data on data frame on every one of these cells are shown up. Is that is that accurate, yeah yeah and you basically starve right memory? You know- and so this is similar to you- have a radio receiver and you just continuously speak to it so that it cannot read receive receive anything else. It's it's a it's essentially jamming right.

C

G

A bit different all right, because here you're taking up resources on the E and the nodes schedule yeah, not on the not.

C

On the air, but if you, if you're doing this, that impact stays I believe at the pledge, because militias going to talk about it, but there is a mechanism by which the pledge does not relay any bogus. Drone requests, that's correct. It stays on you.

C

Sorry I'm asking.

J

Militia so discuss this. This is militia, so I will discuss this mechanism in the presentation. I guess you were thinking of the joint proxy, not the pledge. Yeah. Sorry excuse me, john bursee,.

B

G

And that stated run forever. Yeah.

C

Yeah, do you want to say something.

B

H

Can I hear yes.

C

Perfect, okay,.

H

So yeah sixth Erica from idea the issue here is that the droid proxy needs to install autumn our cell of the pledge. So so, when the joint proxy received a frame, the packet for the mattress pledge, the joint proxy will install autumn cell to the metal versus place that the consuming the memory on the pledge. Another sorry proxy. This a later issue I see.

C

So the the thank you yeah, it's the opposite issue right. The the giant proxy fills up his schedule with cells from the pledges potentially blocking other other cells. Do we have a proposed change to that because I I have the art I have the dress here? There's no proposed change. We have any proposal or just an issue that were fighting. If not.

J

I would have to I have to this is a militia. A no-brainer would be just the don't downwards traffic that downwards traffic goes into the minimis and upwards traffic goes into, the autonomous cell are itself or the drawn proxy. This would be a no-brainer, but we can discuss this further. Okay,.

D

Just two little comments is: usually you limit the number of buffers that that can be used for certain purpose, so you don't starve for the rest of the traffic, but then the attack is if, if the attacker uses all these buffers, then a normal joint cannot happen right. So you starve, you leave the traffic the way it is by limiting number of buffers, but you starve the newcomers.

D

J

Is monster again so I mean this depends on how drawing drawing proxy is implemented, so I mean right now we have two options after working route last call we have a stateful and a stateless option in a state full option. The drawing proxy will typically keep a couple of entries for the pledges say three entries for the pledge to join to it and with the stateless option.

J

There is no such a thing to justify, read all the state that it needs to keep into the onwards, to the DRC and in terms of the neighbor cache that you're, probably referring to in case of stateless. If I.

D

Was talking about the cells that you locate just for accepting gusta skulls so.

J

This is a not excel, so this is at the drone proxy. This is one cell, so it's not that.

D

If a malicious guys.

D

Started guys who want to join because he will turn like a physical champ just to start across sales. It's just genetics on.

G

Eric's our.

C

Manager says yes: I propose that we have you discussion on the mailing list, but but yeah I I can I can actually see in the drafters there's a proposed a little bit lower. But let's go to you discussion on the under a list.

C

To see mod to do we switch to the assimilation results? Yes, you can go ahead and switch Thanks.

C

So you back back to you.

H

Hello, I'm Erica I'm, going to talk about a result. The lessons I learned from the simulation simulator implementation to- oh- maybe that's so now. The next slide. Please.

H

Okay, so to begin with, I'd like to introduce the 60s matter itself, so we release a new version, the yesterday the version 1.1 1.6 and it's implement almost all protocol, which you need to simulate 60 Network behaviors. It equals a ripple, Treecko, Alvin, M, $6 fragment and the fragment warning and also its support.

H

Rfc 8480 is the sixth of protocol, so the simulator supports the 6p pulley, so you can use three step transactions as well as a two-step transaction, and you can see that the sacred joy and simulator try to follow the minimal configuration of the 8180 as maximus as much as possible, of course, is supported. Chch and the simulators MSL is now up to date to the latest version on the draft the test set. Please.

H

So here the typical use case of the simulator. When you come up with the idea, then it will evaluate the idea with the Scimitar, because it is a iteration. The semester is much faster and easier than with a real hardware. So when you get the result, you can share them with others, or if you see the idea is good enough to implement. So you can test it with the real physical hard devices.

H

So this you can run the simulator on the computer classes system so in fact, I'm usually to our institutions the compressor system, to run this simulator. So you can use hundreds of the CPU cores to run the simulator in parallel and also the 60s scimitar implements radio propagation model called a pistol hack, but also its support to use the collective trace which you obtain from the real program or test fit. The collectively trace in this context is that the link PDF and our SSI measurements of all pairs of mode in the network.

H

So six this meter can use that measurement. Instead of using the riddle propagation model, then it can reproduce the the radio environment, the real, real, real environment, of your opponent side in the simulations. This is very powerful and you will how to were realistic. The result next slide, please so that the next slide is kind of the future. Worse. The left-hand chart is from the paper titled assimilating sis networks.

H

This shows CDF of how much time it takes for a boat to get synchronized. So there are two lines. One lines is the result of the simulator. The other is result of the open, defense and implementation. As you can see, the these lines are almost identical, which means the Scimitar assimilate the actual implementation. Well, so we are planning to conduct further performance comparison with the 60 stack implementation.

H

The other thing is the project soda. So now we are building 60s evaluation solution using this test bed and using this simulator, the soda is the project name. So if you interested free, checked the papers shown in this slide, so there are 60 shot. Another action next slide, please so from here I'm sharing the lesson I learned. So these two I think the more important than others in terms of the performance impact.

H

So during the preliminary simulations I saw, only a few were analytic. Cells are allocated even under heavy traffic load. So I look into the log fire and found that the are loaded for the debits cell allocation. Does it work well, the autumn of cells?

H

So, as a result, again, even a heavy traffic load mode can't we have only the autumn of cells. This means without it that they don't have the delicate cell then eventually more to have more incoming traffic than it can handle that it can afford. Then tsq grows, then the it run out of memory for txq. It start dropping the incoming packet. So obviously it is impact on the reliability and I think we need to warp or to harmonize the algorithm with the automat cell thomas comment, yeah.

C

I had a question, so you, okay, you're, saying that with MSF one, which is, if people remember the kind of the mix between the old MSF, zero and ASF. So now we have nose, has autonomous cells and then, after some time they can add that the gated cell, if there's enough traffic and you're, saying that that switch is a little bit too slow that the modes are using autonomous cells. A little bit too long to the point where.

H

You're too long, and sometimes it cannot allocate a typical cell, even if it's e, they did yourself. Why.

C

Would it not be.

H

Okay, so I explained so at first, the the mode held only autumn, our cell, and sometimes they they have collision on the autumn, our cell, once the collision happened, the motor need to start a retransmission algorithm and perform the backup. Wait. The during a pack of fate, the the cell usage ratio Oh the autumn Estelle Goes Down, which prevents the algorithm from triggering the delicate cell operation.

C

Okay, so would you say that this is a misuse of the backup way that we're backing off and carrying himself cutting ourselves from using or not accounting? Well, for that usage is that.

H

I think the accounting of the usage need to be improved.

C

Okay, thank you just for a clarification. What is your traffic pattern going on? Is this bursty traffic discontinuous traffic data.

H

C

Are the modes producing data periodically or is it imperative.

H

Yeah, thank you.

H

And also I found a friend penny big feature. This is introduced to handle okay before.

D

You moved here, you have written them, sell, hardly Richie's limb them sell. What do you mean by hardly the term hold.

H

Each the best chance to trigger their guess our allocation, so.

D

How do you mean XIII? Does it right? Because when you say hardly sometime I understand people understand never, but it's always right.

H

Sorry and the heavy traffic is.

H

So actually, we need more simulations, but in a heavy traffic load under the heavy traffic, the NAM cell use doesn't reach this threshold.

H

Okay, so let's discuss the further on the Middle East and the the other thing is okay, the minor thing, but the the frame pandb feature is introduced to handle the transient the first bursty traffic, but actually this feature could cause the bursty traffic and make the decision worse. So that means that the collision and the chance of collision so yeah again I think the scaling algorithm itself.

H

We need to improvement.

H

So next slide, please.

H

So I already sent other many comments, including the the what Simone pointed out. So you can see the other comments pear found on the the working of midgets archives there, the link to the delivered thread and now we're evaluating the beta semester. Implementation with the the new release and was the result, is ready. We will share it with the working group, so next slide. Please so I think the conclusion of the presentation so again to me introducing the autumn of cell to the original MSF may not be easy, as we expected.

H

So is more the work and also the pending big. This is very interesting, but it's under specified and I'm, not sure our implementation is correct or not so Tara told me that there is the effort to rewrite the text of the 802 50.4 documents about Prendick big feature, so I'm very looking forward to new text. So this path is now the trucks as a CID 93 in the group you.

K

Have a question for no actually I was going to respond to that says there are keeping it. We are going to be discussing that next week. We actually I think they have already done some of the work, but they haven't finished all the th-th-there. So there were so many you know either some of the TRC 8, so there have been going through them and next week we are going to be processing them.

K

I too believes meaning here next week and we are I think we have a six or two hour sessions to work on those items. So we hopefully by the end of the next week, we will have some solution, at least for the TRC, a start. Taro.

C

Can can I can ask you to share those what is shareable with the group? Yes,.

K

Actually, all of the documents are in the mentors. As you see this, you.

G

Know Lincoln, that is.

K

There yeah actually- and there is there's this there's- this is a common database which is big excel sheet, and it has you know, all the comments and all the resolutions to those comments. So it's it's it's it's. You can always go there and check it out.

K

Ok, so so so you're you're asking you have to go there by the end of next week by the end of next probably going to be in about three or four, because you said we usually do and then dead of a day that you know upload the latest version so that the next day people has you know all the you know latest these kind of producers already there. So thank you. Please. Thank.

H

You Tara I will I wrote AK.

L

How many nodes were in the network, which are simulated, I mean using the MSL.

H

We have 30 modes in a random data project.

D

L

Lijo thomas Lido yeah, so how many in.

J

L

Okay, so the do you feel any difference when you run the same thing on the open mode or the hardware, not.

H

Yet but the way we show some lessons from the real opened up: chanel implementation, the next presentation yep one.

L

More clarification, how many hopes you have I mean apply bulbs or, more than so.

H

That's a so now, sir, we are conducting the the actual, the a variation right now so weights ready. So we can share this detail with you.

A

Maybe I can show something my experiment without I think.

A

C

M

Yes, I have a question regarding with a simulator, so we're using the 60 simulator. Can we like generating our own topology.

B

M

H

M

You mentioned this connectivity trace, based on the real deployment test better. You have I think this is a very useful feature but say if you generate your own topology with a node with you know, the transmitters and receivers may have different distances in between compared with your your data gets from testbed. Well, that's the connectivity trace still work.

M

G

M

This connectivity trace model, which I think you built based on the real test pair, that you have right, yeah.

H

M

That testbed could have a like a fixed apology. You know with a certain distance between the devices, but if we generate our own random topology with different distances with that connectivity trace model still work well,.

H

So connectivity so that it's not a model! You used that the measurement value instead of the calculated they say that our SSI or PDF, valued by the the radial perforation moto, so.

C

Maybe again, I can try I see yet you're traveling to struggling to understood here so I know the simulator. So it's either a model where you position virtually different modes and it will tell you the the quality of the links. That's one option. The second option is you scrap this completely and you replace it by replaying connectivity, traces that have been reported before on deployments and test beds.

M

C

It's a fixed, topology and so I I, don't know how many they are, but there's a database of traces coming from the old IOT lab deployments, some of the deployments in in different countries, and so it's kind of a database of scenarios you can play with. But it replaces entirely the models.

H

Thank you, so the last one it's a just for your information, I've just updated the flow chart that I showed in the last meeting, so just for a fixture. Thank you. That's all my presentation. Thank.

C

You so much yet any further question.

C

If not, then we, let's move on to thank phase presentation about experimental evaluation. Yeah, oh yeah, sorry I forgot yet to ask the important question so yeah I know if you're still online, but would you say it in Appendix II of the MSF draft. There's a list of lessons learned basically from your exercise. What I think other people have done is everything you you note it in there, meaning if we resolve that the Appendix C, will you be happy or do you have more things you want to? You know suggests changed in the draft.

H

I think I will be happy.

H

Find something in the new variation but I think I think I already gave all I have right now: okay,.

B

C

Thankfully, the the four is yours, if you can hear us.

A

B

You hear me well yep, we hear you perfectly yeah.

A

Yeah I will present, on the first sight how I run this is a tempeh and from Paris I'm, going to present to the experimental results so sings. For the first time we have the MSF. We did this implementation in a test pan in the IOT site, but since the test pad is deployed in the underground parking place, that that is not very elastic cases when we do relative pointment so to tester the MSF performance in the real scenario. So we build our new test. Bad co, open this bad, that's showing in the picture so open.

A

This valve is testbed, contains 20 Oh tea box. The audio quality is showing on the first picture on the left and we have 20 of those 2 audio out. They deployed in the in Rio Paris office building. So for each OT box it contains one Raspberry Pi and with 4 open mode, be connected through the USB port and so for.

A

The Raspberry Pi is running a person script and with this script they allow the users can reprogram stones or assess moves and also enter a interact with the support from the most like getting the debug informations.

A

And the scripts is open-source and it's showing in the link showing on the slides, so everyone can use it test ban and deploy their owns or using the test bed that we provided so forward. The architecture to tell I think it's bad go to the next size.

A

So for the architecture of the test that it's based on the mqtt.

A

The communication so for each OT box it will running this isin scripted to connect to the MQTT broker and on one side we have the notre addict and so on city. On the IBM cloud, you provide a GUI interface, so user can trying to do the reprogram modes or is that those actions so through the interface and in other hand, the people the user, can use at the API provided by the open test path to int interact with the open test by test bat doing the same thing and for the architecture?

A

The whole test path is running on a separate network from in real sight, so everyone can connect to the internet, and also this internet is is running on 5 gigahertz. So this will be awarded the interference from the test pad itself. Network.

A

Next next slide, please so for the the picture showing the slice is like is what we, where we deployed our OT box. So each point in the figures is ot box and the point with the label with the green items is where the office of INRIA it's a room for the office for office and the blue one is the forward meeting and the red one is for public area, so the workers deployed in the office building, but for me, I was sitting in a 102 where a transcended- it's also there.

A

So this test belt is really deployed in the real scenario for for a office building scenario. So with those tests better, we are going to have the result for real real scenario. Crew go to the next slide and.

A

We implement the sake Taoiseach through the open, wson product, open diversity. The open source project is implements that right now, the sick, tatius, the whole stack and retinal images are released, the 1.20 2.0 version and within version we have the we open support for, don't forget, hurts and because open mode has to redo. So we have a post port right now.

A

For this result, I'm going show is will be only using 2.4 gigahertz and also this release that contains the tones of the feature which enabled notes to have a 100% serial communication with the most and out as well. It implements the MSF dropped and also in this release, we have the fix the box and which is pointed out by Christine agree, and with this we have some work under our imitations and right now to the Box, let's fix and as well those new release with the new release.

A

The user can upload the former they just compiled to upload to the test that, on the right side of the slice, is showing a screenshot. After all, the nose, our flesh with the open division, sixty imitation, we have totally 76 most deployed on beside two buildings, we're good the next slice.

A

So for evaluation, the secretary mutation, we agree running with the open utilizer so with open which wiser it can connect to old mode through the test pad and getting the information from the mode who is going to stand at the bug. Information through the C report and as a well as I can stand command to the mode, for example like it's that one mode as a dagger root note and also it codes to the open visor is trying to receiving debug information from all the 76 possess.

A

At the same time, so we optimized open riser, trying to get all the information updates it in in time. So now it's for the 76 most sending one packet, every 100 milliseconds.

A

So with those features we can start to evaluate, what's the performance, citation and also as well those the invitation, citation and open utilizer posts are open source. You can get all the code from the link, as shown on the slides next slide, please, okay! So now I got the experience results as showing in this slice. As you can see, this is the routing topology.

A

After running the six, a stack is totally 38 knows why, sir date, because it deploys the smalls in two buildings, almost half of them are out of the range of the one building.

A

So it is a basically form a network is with a one at all the most in one building, and for this topology we can see the maximum hop is four, and most of them are staying first to a sex in the hops and now and in the next slides I'm, going to show the end-to-end reliability of the each nose you can yeah thanks and for for the result.

A

First, we configured the invitation to using slavery twice so, which means, if I transmit three times, and they still not able to transfer me out, then I will drop the package and the results showing on the right side. So you can see almost all the odd nose has a 100% end-to-end reliability except the mode B 618, so I after I checking the source state data from the old packet of received I see, there's one Church in needle it's a mr. several package.

A

I believe this is because during the traders are forming its route for me, is reading, it's choosing a bad liver with a lower rank and it try several times and it failed, and then it will change back to a new parents and, and then the those packets in between were lost and makes this and non 100% reliability. I think this is imitation issue.

A

We can trying to mint, maintain still keep those package in queue and once they have changed the parents, it can change the destination and trying to rest and to the new parents to save those case. So I think this is a reason why one note is no 100%.

A

Okay, your go to the next slice yeah, so throw the implementations of the MSF.

A

Here's the lessons, I learned from formate first I, think I gather the similar result with a judge for the apparent LMS at zero one and here's a some Commons I have some of them may be the same with watch is a present ate it before first days we have the MSF zero one. We have the probability broadcasting on G, B and D iOS, and and and this makes it a synchronized station of the most take longer depending on topology I. Think this arrays more.

A

C

Yeah, you have a you, have a.

I

C

A

Please Chris George quiz.

E

Mmm-Hmm the first one is.

E

Devices or even the relay devices are act as a source like all of all devices, all 37 out of 38 they transmit, or only the leaf devices. That's my first question and the second is: what is the traffic? Is a CBR I mean bursty, and it is it's one packet every how many seconds or minutes for thanks.

A

For a question: well, the first one I think here you are asking like why the number is 70-37. You know.

C

Whether do they all transmit to the old generate data or are only the leaf nodes generating data yeah.

A

All the node except the de Groote, generate the data okay, and at what speed and the speed is, you know the world generates a one packet every one minutes. This is on the application layer. Meanwhile, each node will also have the keep alive, a packet which will transmit every twenty seconds. If there's a no traffic traffic in 20 seconds, he will transmit a cable ID, which is local yeah. One.

E

More minor question: what was your slot frame size, 101? Okay, thank.

A

A

Resume on the last slice for the second point, in my invitation, one node can hear kind of more than ten of the nose, and in that case, if I reserve, but marseilles TX tells to each neighbors, it will cause my ass gather overflow problems, though I can increase the power of the scatter button. I think that's not very cool soldier and as well the TX or autumn RC. Is this. The SMA is in CSM in fashion, so you know which means that the app the application package also could be transmitted on these two cells.

A

But since this there's a lot of other also traffic on transmitted on the siblings and this will coast, sometimes the ipv6 packet dropping and for and the last one is a similar- is the same. What a yacht has present that previously and according to those comments, so I have some proposals for the next version with MSF the first days.

A

I think it knows only can only reserved only reserved automatic cells to there parents only one and so basically for each node it will has to Ottoman cells and the one is reserved as a tx/rx shared Ottoman cells, but but only to the parents is a unique a cell. Another one is to reserve tx/rx, not share Ottomans itself, but it is any custom one. So in this case, all the node will still standing the six top request on the Unicode master unit yourself and for the parents.

A

When trying to standard sick tops response, it will transmit on the bottom are cell, which is Anna cast you. You notice that any caste is no shared, which means if the parents six response field, because collision, it will transmit immediately next cell, it won't back off, but for the children who apply that into a backup. I think this is a reasonable.

A

This will savage the six top transaction to success because in the prairies, if experiments, I was a look at this issues. It takes ages for the children get to the sixth out response and, in other words, I want to bring back, though the matter man, one managed txl before we merge the automatic cell at enormous the forum, work, ASF and MSF together, so which means right now we have the autonomy announced, tells and then after it is installed.

A

The first thing is using this autumn, normal cell, to install another txl for for later the application to transmit. So in this way we can separate the application packet from the autumn ourselves to increase the reliability, because it is not in contention fashion and with those settings and in currents the MSF zero-one version. We are saying sending the EBT house once it has the cell in in I. Think.

A

We send evn the house after we have the one amenity Excel, but I think we can still standing the EPS even before we have a manatee XL. So for that case the EP will help all the notes to be synchronized as well. It won't influence the note like a building its routing topology, but before is stabilized yeah. So this is a some my proposed for the next version and that's it any questions.

C

So some remarks for me so same questions for ya. Does the appendix e contain all the recommendations that you're making or are there some that you're they have discovered since or that you want to change.

A

You see complete some of them. Yes, some of us know so I think a neck version will March posted a recommendation from the current and the one in the opinions, a section in the current CMS. Okay, it's.

C

Really important that you, you raise those those issues that are not Appendix II on the main list, so that they can eventually make it into into MSF. Now Simone you had a question yeah.

G

So I'm just trying to understand the like the problems you were having with the autonomous cells and the proposals you have so I feel like the main issue. You are having it that you end up with a lot of T X cells in your schedule, right.

A

G

So maybe what we could do we think about the conditions when we actually installing the TX cells. So maybe we don't. We don't install txl for each name on the table.

G

Instead, we could lessen so TX our on demand whenever the packet two cents to a given neighbor, so because so the the idea in in the autonomous cells really is that you just have one cell to to listen, and everybody knows what that cell is because it's it's the hash of your Mac and then, whenever you want to talk to someone you just hash their Mac. So I agree that in our current draft we may take too much resources with having one txl for which neighbor.

A

G

Let's just have it on demand: whenever you have a packet in queue for neighbor B, then you either, except whenever B and you send it over there and then you can even have it timeout for that Sam yeah.

A

I agree: it could be several proposals and I think that's one of them.

C

Yeah, there's something I don't understand if you proactively install all these th cells you're, not using them anyways until you have something to say. So what is the difference between C Mon? What you proposed, and the current state is that that we prevent installing additional th cells when one is already in place at that time. Slot is that the anything yeah.

G

It's the only thing like just save some memory on the schedule side; okay, because that seemed to be the issue rights. Thankfully,.

A

In that fashion is like eat, my schedule have the TX down like one time. You do up here here and later up here in another place, because.

G

I mean that's a stoma, pointer I think there is no difference in terms of what's happening on the air. It's really just about how you manage or your schedule internally. It might even be up to implementation. How you do it I.

A

Think saying, what do you propose am I, so you are saying both the sending in store the sales on demand and actually, in the most case, that the children will actually only talk it to their parents. So I think.

D

Guys, please, let's take it to the meaningless, because we are running out of time and safe from any shadow. Let's question really so.

C

One more remark: have you tried consolidating your findings with and let's see, moulded a bunch of work on the similar things? I know that I mean judge did. Did you try to consolidate those things yet? Are you finding the same like issues that yet, for example, had we haven't tried yeah I.

A

Just more like a talk to offer tough Allah privately and listen, what did they learn and then I at these experiments? I can have similar results. It is yeah.

C

We can Pascal is proposing to do an interim dedicate for this to have some high bandwidth to the bodies money. She had a last question so.

J

Christian Hoffman from Jabbar. The question is: what about installing T excels only to ripple neighbors.

A

I'm, sorry, who is all the Excel to ripple neighbors only triple neighbors ripple.

J

Parents as well.

C

Do you mean how or why he's he's proposing another? Another alternative is to install not only myself only two nodes ripple, he says neighbors. Do you mean parents and children, yeah? Okay? So let's have an interim to discuss on MSF specifically have some high bandwidth, you know a new but but I want. You know, I want to stress how how imported or though is it is because you know we're getting almost live feedback from from our sanitation work. So that's very, very cool. So thank you so much for going through the exercises in this.

C

Thank you with this I'd like to move on with our agenda militia representing the mineral security draft.

J

Yes, so we public mean we had the working group last call.

J

It was before Montreal in July, I suppose, and ever since we published two versions of the draft, basically resolving all of the issues that were raised during the working group last call and during the working group. Last all we had a lot of reviews. We had seven reviews from neuron terror, Cioffi, klaus dreams, I think Faye and William, and then Thomas Thomas did a review yesterday, also which is incorporated into the Oh 8.

J

So my goal of the presentation today is basically to go through the major issues that were raised during the working group last call and to present the contributions to the group, so I said: I will present major major issues that were a result. There is a whole bunch of fish, other issues and editorial changes that were made to the document and please follow this link to see all the issues that were that were addressed in the latest o versions.

J

So the first issue that I'll present is on stateless proxy and we had a lot of discussion. This was with Tarot and also with Klaus heart cave Klaus from core, and also your on cylinder from your on cylinder brought up this issue. So essentially, so.

J

This work has now been outsourced to core core Klaus published a draft which has just been adopted in core, which defines which basically extends the base coop header, with an extended token, in order to support our our use case for having a stateless proxy, and you can see the format of this new coop header on the on the right on the right and basically just some reserved bits from the base.

J

Co-Op header are now used to encode a longer token and what we did in minimal security is that we completely removed the definition of stateless proxy option that intent had the same purpose.

J

Basically, carrying the longer token that would anchor that we carried the state of the proxy, and now we have a section in the configuration part of in the application layer configuration basically recommending they use recommending the the drape the Jaypee to operate in a stateless manner and to operate in a stateless manner with JP needs to do is basically just wrap the state that it needs to keep prayer pledged into a quad token.

J

For this it can use the the coop token is defined in RFC, 72-52 or if this doesn't fit, it can use a call size draft that will go for working group last call quite soon, and basically, with this, we now have only an informative reference to the core to the core draft. Then also after long discussions with Tara, we we, this is now the recommended behavior.

J

We also have a disclaimer there on the implementation stacks that where it may be difficult to implement, the statelessness add to the join proxy and I mean statelessness throughout the whole stack and there we note that in these cases the proxy can operate in a stateful manner, but the disk comes normally at the price of an increase of an additional dose factor.

J

So the second issue is on the use of confirmable qualm message reporter on the quest. This is something that was brought up during a core interim meeting sometime this summer in August when discussing the operation, also in coordination with Klaus. So basically previously, what we had is that the pledge would send a non confirmable request in order for the drawing proxy to forward the request as known confirmable, and to not to keep any state for retransmissions at the proxy. So later, when discussing this, we came to know that, basically we can. There is no.

J

The proxy is not required to forward the message in in its original type. So what we do now is basically, we simplified the document by requiring the pledge to send a confirmable co-op message, but then a proxy forwards it as non confirmable and therefore there's not need to keep state for the retransmission.

J

The retransmission is then completely handled by co-op, and what we need to do on in the minimal security document is only override the values that the cooperate transmission mechanism uses for for our use case, which is join and I, give here the the default values that we recommend on using third third major issue is a bandwidth capped at JP.

J

This is basically as JP a forwards on attend traffic coming from unauthenticated nodes. This traffic consumes the bandwidth upper up in the in the in the mesh, and basically, we discussed also I think with Tara. This was one of the comments on basically limiting how much traffic the JP should forward when it operates in a stateless manner. So, basically, a good good thing with this with this was also discussed during the quarry interim meeting, was that co-op already provides the congestion control mechanism and basically in to two parameters?

J

One is n start, which is useful for the state, the proxy that operates in a stateful manner and basically and starts just specifies the number of requests, number of requests or number of ongoing transactions with a given endpoint. So basically, by setting and start to three after the drawing proxy drawing proxy will never forward more than three requests before either getting a response or timing out by some mechanism.

J

That is not defined in in coop, and the second parameter is probing rate is useful in the case of a stateless proxy that is specified in bytes and basically, and it's essentially a bandwidth cap that the proxy should not should not exceed when forwarding to a given endpoint. So the good thing about this when we use this there is absolutely no for us, no need for any additional lines of code, or all of this is already in 72-52, and we need to define values so for this.

J

I will now present what we have innovate and at the end of presentation. After some discussions that we had during this week and also today, I'll present an additional proposal that we can that we can do so.

J

What we have innovate is basically this table with default call with default values of what parameters for our case for join, where we specify that n start at the pledge to build normally one and but at the joint proxy should be three meaning that the drawing proxy can forward at most three requests without getting the response, but then for for the stateless case, where we need to specify the probing rate.

J

What currently stands you know: eight is basically the equivalent in bytes for these three for the N start of three to calculate that equivalent in bytes I had to make some assumptions on the drawing request, message, size, etc, etc. So that basically results in a probing rate of 12 bytes per second at the join proxy.

J

That should be averaged over a window that is 10 seconds long, ACK timeout, so I don't have time not want to go into the details of this, but please bring the question up if you're interested we can discuss after essentially I mean with this 12 bytes. We consume around 18 percent of the available upstream bandwidth, assuming one cell with preferred parent on the path to the 2.6 l BR, and this can be and can and should be higher, but as we discussed also on the list, this is something that the Jaypee might not know.

J

So you will see now in the proposal we will be base. We propose to make this configurable by the jrc so that the GRC just can send configure the drawing proxy with a given cap, depending on the current status, for example, if network is forming or if there is an attack, or we give basically this control to jrc. So this is the proposal 409.

D

Just to think I would read that what in list? Because we would like people the mailing, is to be able to come out and feedback.

D

The other thing is, you know that we are working on getting from repo the capabilities and the situation at the root going down through the dials. So we don't have that specified yet so you cannot point on it, but you could have some losing world which say: okay, if you have some indication eg from repo about the saturation of the the node of the route, which is basically the bandwidth that is variable the wasteful, because the road is the gating factor, then you can use that to automatically tune. And that's that's a.

G

D

Would be a slave of the route in some fashion right in implementation of that right now is the route would give this information to Jesse you get from the GRC, but you could also directly get it from the route, because we will show that that signal in the iOS. So it's just matter of running that open what the source is. If.

G

You don't specify.

D

If you specify the transport, like you put that the GOC, so here's the field so we'll do it this way. But if you say oh, if it's variable from the routing protocol or something you may use that to turn dynamic. Oh.

J

Okay, I see yeah, but I mean the intent was just basically have. This is a field to configure the drapery value.

D

Which is static, then they expect you, you figure. In your mind, it was a static value that the administrator.

J

D

What I'm saying is more like a dynamic observation by the route, which will part will be part of the DAO, and so especially that would tell you not a static value of 3 4 5 6, but oh, if the route is loaded and lesser than that, then you can have so much. You see some logic. So basically, if I just say hey, you can get base. Don't Monday that it's hard coded by the GRC is not a good idea. Well,.

J

I mean I wasn't going to hard-code the value of course, but I mean just means for Darcy.

D

To communicate, it died, which is good, and then you can say it can also come from ripple.

J

Yeah, so, okay, then we should think about this on the from the on the text. Basically so yeah. This is the intent 409 basic to introduce. Did you know nine? Because right now, it's not you know it. This is something that we discussed this week. Then a fourth afford major issue basically was raised by traveling an error handling section on the quadric protocol.

J

Basically, what happens when an error is encountered at the Sieber level at the payload at the level of the payload of this configuration protocol and quadruped, when processing, sim or objects are processed Bo about this part of coop requests and responses for requests? It's quite easy, I mean we. We just need to transport. We return for 0, 0, bad request response, but for responses, it's more tricky as, for example, you get so the player gets a response, a configuration object and it cannot process it.

J

So this was a co-op response now, so what does it do now? So what we meant a it now is what we recommend basically, is that, in the next request that the pledge will make to join to the same drea are see additional parameter that should be included in this drawing request. Object is basically the error code from the previous from the previous attempt to join and.

G

J

Through this I mean we give some hint to Dre RC on what went wrong with which the pledge may be. Some parameter was unsupported, so maybe we should exclude it or just help the debugging information, and to do this, we define the new support structure, called error, basically carrying the the code, the additional information and human readable text description, and we give like some values for that.

J

Fifth major issue was on a EAD known 3u that was brought up by Jim shot. So basically, the case here is that jrc, for some reason, completely fails and loses all mutable context parameters, for example the sequence numbers in the replay window. So in those cases this was also discussed with Tara. We mandate now that the jr's that the network, when this, if this happens, that basically all the networks that are managed by the GRC need, must be reinitialized and all the nodes forced to join.

J

But then the problem arises when these nodes rejoin, if the tray RC attempts to update the parameters on any of them, and it lost the previous security context with these nodes and as we don't want to reprovision every node that was ever deployed. The problem is that when this first parameter update message from the GRC to the nodes goes into the mesh, there is an imminent risk of this of non 3 use.

J

Basically- and this was brought up by Jim, shard and dream was also kind to propose a solution which we now included, and this portion is as follows. Basically, the three RC knows that the new jrc that takes up from the old GRC knows: ok, there's been a failure. There is now the risk of distance. The nonce reuse will happen so now in the draft.

J

We mandate that the payload of this request we just padded with byte string that they generated uniformly at random, and we send this request that will cause monster use to the node who will detect the nonce tree, use, as the sequence number will be largely below the one that it expects. So this is all already handled by our score and basically so what we do is that when the node detects this significant mismatch of the sequence number, for example, just reset of the context 1.

J

It respond with this error that we defined in one special error case where part of the additional information of this error is. The next is the next sequence number that the node in the mesh will accept from the jrc and literal with this. Basically, we transport the lost state from the node back to the plug and play, and the jrc receives this error, which is encrypted with with the new nonce. It is not reused, decrypted gets the state and recovers the state, and then it can continue as follows.

J

So this is in the text I mean you should probably read the text for for the warding dream was dream seemed okay with the resolution ever since I did some editorial comments, but the solution should still 6th major issue was on the blacklist parameters. So taro was arguing that in k1 likely case is that there is a Miss configuration of platters that we continuously try to join. So, basically, to support this, we added the blacklist parameter that jrc can include and send to any pledge or the drawing proxy.

J

That is just basically a list of pledge identifiers that it needs to blacklist and how this is implemented, of course, is implementation. Specific I mean many implementations. Do this in hardware, and but essentially with this, we give control to the GRC to blacklist. Can.

D

You and blacklist yes,.

J

So yeah it's a good question, so basically, I'm blacklist would be sending the black the black list. That is empty.

D

So that means that you always send the full blackface. Yes,.

J

Yeah, this is currently the case. That's right so so the intent was for the last blacklist value to override the previous month. Okay, but yeah I would yeah. This was the intent. I would need to track the text if it explicitly denotes this, but this was the attack, basically take whatever you receive and just configure. It is that okay or just.

D

Depends on the number of devices you will have to blacklist, but yeah yeah, maintaining addition and synchronization is so.

D

J

So I'll check I'll take is an action point basically to check the text explicitly state this. This was the attempt currently. So yes, then, the seventh resolution was I think this was Terris review. Yes, after the last IDF, basically, we extended the link layer, a key structure, and now we support all modes of 15-4. We add an additional parameter, which is a key additional info.

J

It is a byte string, it's optional and depending on the combination of these parameters that are included, we support different modes and there is also the mode, the implicit mode where we essentially, where the what the jrc acts as a trusted third party, but we do not detail it. How this is done and.

J

Basically, I mean this key can then be used. What we discussed on the list for additional protocols to bootstrap the keys between or or for future extensions before pair, wise, king, basically so taro. Could you check just this text if it's all consistent what we discussed I think because we had so the you know, but this this is. You know seven. This was done in oh seven, yeah, so yeah yeah and then yeah. So there were a lot of other miss comments. Updates of the security considerations, unique quest requirements were explicitly stated.

J

Privacy considerations were updated. Some overhead optimizations were done. Like short address. This time is now encoded in hours, not in seconds. So basically you can get. You can specify the least time only in multiples of hours upon ID clarifications that it's not table, we mandate. So we reverted.

J

We went back and forth with this, but so finally, we mandated unique a pre-shared keys, but we added a lightweight implementation of the option, appendix the basically details and implementation options where the implementation of HK DF and she is not necessary in favor and also we removed the network.

J

Prefix parameter as discussed at the last ITF, because we decided not to go into the 6 lb arm and configure configuration management and, of course, I we great if people could just see if the Fisher's or just go through the document for consistency enough for consistency, but if everyone is happy with the resolutions, so, regarding the next steps is I, we had a lot of discussions also this week and today the two pending the two pending proposals is basically to add.

J

This bandwidth cap parameter that the jrc can use to configure JP, and basically this would just extend that configuration object. That goes into the drawing response or in the parameter, update message and then the second one was. Basically, there is no. We just noted that there was a remnant on this 6lb, our management.

J

Basically, we still have a network identifier that can be returned as part of the join response, and this only made sense in case of 6lv are joining and to bootstrap. 6Lb are completely I mean it needs much more parameters, so we decided not to go that way, and so the proposal here is to remove the network identifier from the join response, and it requires the 6lb are to be provision with the network identifier before it starts advertising. So these are two I mean short modifications.

J

I mean I can probably do this tomorrow and after that by week we can discuss on the list. So this would be online and I'm just request here: input from the trace on the button. Next steps for the.

D

Well, if you thought just the proposal, like I said, you need to go to meaningless. Just summarize that just like you did, and so people will react, I get it like. The route could also give you hints now for the other points below we have suresh in the room, so I hope Jones to Mike, because basically, once you're done with this you're ready for let's call right but but then you've got this. Do.

J

We need the natural last call or how it goes normally, because we had last call in when was it in July.

C

The history about this is we have it. We had a last call. You had lots of reviews, lots of change that you just in in fixing that you realize there's two little things that you still want to change. Yes, because you're going to change that you have to do the proper way. So that means discussing at the mailing list getting some feed some feedback, some ideas implementing it into the draft, and then we do a short last call at the end.

C

Okay, because we don't, we certainly don't want to ship something we're at the last moment. Something slips in and you know, there's no proper review.

B

C

But but I you know I for having read the draft very recently. These things are tiny little changes. I think we absolutely want to do the last call, but but we will keep it at a short last call. The question then is then this thing is ready. The question, of course, is you know the ops core kind of blocking there. Yes,.

J

Guarding, what's going on this promise, but I mean right now escort has. It still has a discussed to our knowledge and progress.

C

Since Prague, so it sir but L hawk, is is not in this. It's.

J

Not this rounded yeah, it's important for yeah.

F

So a couple of things first of the last call like Thomas, said most of what I wanted to say, but one added something in case: you wonder, shot loss call you can limit it to like the changes that are there, so not opening up the whole document for comments again. If you want okay, if you want to keep things simpler, but that's fine, if you want to do it, Oscar, like I, had a chat with Alexei, so he thinks it's progressing like he. You know he had a chance.

F

He had a chat with about this as well. I was going to talk to Alexei, like you know, but we kind of kept missing each other because he had to present now attack me so I can really get to talk to him, but this is something I want to talk about tomorrow, like we have a IES, you wrap up like and now we'll talk about this tomorrow to him and but don't block the Stockman because of us car. Okay. So that's like my responsibility.

F

I'll take care of it when the things to go so like when the working group is ready to push it push it and if Oscar changes significantly before it gets approved and you're worried about it, we can bring back the document it. What you can do stuff, but I want to just go through the process like when you're ready to go so I don't want to hold the document here. Perfect, okay, perfect! Yes, this is great. Thank you. So much.

J

And then just the last point is basically on the future. Extensions I mean Michael, I. Think he's in nice, easy.

G

They say because we.

J

Had the collision, so basically this is no zero-touch. This is regarding at Hawkes.

C

So so we're going to go to a separate talk to these chairs about it. So no input yet.

B

We'll have to discuss okay, yeah.

J

So, that's all for me great.

C

Thank you so much.

C

Okay, we're we're over our official time, but we builded a 15-minute buffer, which is exactly how much small needs for presenting this dress. So we're now entering the the.

C

Non-Charter work new work, you know if you can hear us floors, you're.

G

Presenting that new draft, which is a proposal, the main author of the draft, is marco de luca. You might you might know him wasn't really in six stage, but he's very involved at ATF and security he's no sitting in ace. That's why he's not presenting, but he's actually in bangkok, so you want to have a chat with him. You should manage to meet him, he's been also involved in court cetera, and so what what that draft is about due to security draft and it's about a selective jamming attack and a mitigation against it.

G

So the motivation is that in th, the cell utilization pattern is predictable, because you have slot frames with the given schedule in them and a repeat, and so that versary can easily derive communication pattern by listening over a long enough period, which actually consists of several odd frames and after that it can extract the full schedule of a given victim, node and then do selective jamming against it and why it would do selective jamming is for tourism. One is to be unnoticed, that's why that for the selective power and also to save power.

G

So if your attacker needs to be on battery, so that is the motivation for that work. So what makes the attack easy? There are a number of properties that you have in a sixty schedule, it's periodic, and here by your equipment period of slaughtering lengths times, channel opening sequence links at that period, cells with peat in the same actual frequency not just channel offset.

G

There is the users property that tells us that, as long as your stream length is co-prime with your channel opening sequence lengths, then all the cells are going to be used during one of the periods above they're going to be used uniformly.

G

Then you have the offset property that state that all of these cells, basically repeat with a given fixed offset about slot frames and the predictability property were you can predict the sequence of channels once you have figured out a timeslot so below our number of equation, that I'm not going to detail now, but what that really tells you that once you've once you've figured out the physical frequency forgiven at a given time and channel offset for a given cell, you can then infer for other cells which frequency they are going to run out to run on and give an ASN I'm going to exemplify that little bit more in the next two slides.

G

So the attack outline is the following. So you start you, you started a given slot frame. It does not need to be straight from zero, but you will you will count it as zero in your internally and then then you pick a frequency, so you, let's see frequency one, and then you start listening and you listen for one period and period here again is start from lengths times channel hoping sequence next and then from that, with the equation of the previous style, you can derive the full schedule.

G

Basically, so you know exactly exactly one and on which frequency the node that you are targeting is going to transmit.

G

So you have that full schedule and you can selectively jam it and if you go next slide, you'll have a visual explanation of that. So thanks. So here is a schedule. You have a stock frame length of three and the channel hoping sequence of four, so we're going to keep listening for 12 three times four, and in that example the node the idea, the attacker picks frequency one and is going to listen on frequency, one for twelve slots and what you see on that schedule. In each cell you see F equal.

G

That means the actual frequency where the cell is actually executed. So if you start listening here, we start listening at a century and we listen for twelve slots and we we notice, if you start from a few cells, that executing on frequency one so and there is one on a yes and for one day s and five and then one day is six and they need to repeat after twelve. So that is for the scanning part.

G

So you do that's listening on frequency, one for twelve slots and then you know the full schedule and you're able to able to selectively jam it from now on. So that is what the attack looks like. So, if you good next slide, I give you the overview of this solution of proposing.

G

So the the idea is to prevent the attack that constructions, so we're going to we're basically going to do shuffling of our own schedules. So we do permutation of our own slot frame in the time and in the channel of domaine both well in the time domain. It is optional and in channel of such domain it's not, and that is done locally. Using a secret, so if you go next time, I can talk about the secrets that we need for that. So we have actually two next type it yeah thanks.

G

We have two permutation keys, one for the time domain. It's KS and one for the channel of set domain. It's Casey and then we also have counters for the time and and and channel domain reduce. The counters is just internal accounting. It's just a variable internally, so it doesn't matter as much, but the permutation keys are secret that we're going to obtain at the joint and the secure zone you included.

G

So if your next slide, it's.

C

G

More detail how this solution works.

C

You have a question.

M

M

So basically, I think it's a very interesting idea to reverse engineer the tsth pattern, but I still cannot really understand the rationale behind to use it in the you know, kind of attacking scenario, because basically, if I'm a smart attacker, the purpose of attacking is focusing on the entire network rather than a particular.

G

M

Yes, yes, so that's usually what happens right? You are targeting for the entire network, so.

G

You could imagine both.

M

Cases right so.

G

I could end, plus you could do that on the entire network. So if you have enough so.

M

Yeah so if I say, if I can know so, for example the accountant, so so, if you do selector jamming, for example, the purpose is why you do select jamming, because you want to deal in. You want to jam some content right from that particular note. But if you already know the content, which means you already hacked into the network, then you know the everything you.

G

Want to basically prevent a node for communicating or together you want to disconnect it so you're dreaming. Every single transmission of that particular node right.

M

And in that sense you use, do you need to know the particular channel and time and node is transmitting rather than you know, just keep on jam? It yeah.

G

So the advantage of doing it selectively is to be unnoticed and power efficient. It's a bit more secondary, but it's to be unnoticed, so nobody will notice there's an attack going on his wrist. Oh that node can't seem to talk anymore. Why is that? But you don't you can't notice? Oh, that seems to be a jammer there, but.

M

Would that the parent node can also detect that if the parent node didn't receive anything from a child node, so.

G

The only thing you can notice is that, oh, that particular node cannot talk anymore, but it's it's not as obvious that it's a jamming attack as it would be if you would be just jamming aggressively yeah.

M

D

Discussion, the mailing list.

D

People want to stay in the room, I think we don't think so.

G

So I conclude here so the so it's not frame you basically need to schedule for the next slot frame. So if you go to yes, okay on that slide, so you you first permuting the time domain and that one is optional and that's optional, because we were aware that some schedules may have end to end delay. They need to guarantee. So if you would shuffle the time domain, you would lose the properties, so we make that part optional, but when it can be enabled it gives you extra latest resilience.

G

You can do that via rack and on step. Two, you permute, the channel of, and these two permutations are done using the secrets that I mentioned in previous slides. So you have that secret obtained at joint time. So next slide, please so for the key provisioning. So how we get the two keys, we get them at joint time, as I mentioned before. The proposal is that it's an additional cozy key set in the drain response and that set has two or one key.

G

So if it's two kids, then it's Miz looking to permute on both the time in frequency domain. If it's one Keith means it's just going to do the channel frequency domain and we still need by the weight who update the draft to the new minimal security, triflin, the end egg and the core and the cozy section next slide, please.

G

So to summarize, we introduced that attack and we want to hear if a working group is interested in having me th and I, guess that I can, if so, so, we also have a proposal for mitigation that has good property, that it preserved the collision, free and consistent consistency from schedule, etc. It has no communication overhead, so it comes quite cheap, so it has all these advantages so which really ones here is to get feedback on.

G

If, if the group is interested in these and yeah the next step as well, we have that update and for the new many more security with the cozy keys.

G

Thank you very much for granting some time on this today. Yeah.

C

So we we still have two minutes: oh I wanna make. So when I read the draft I I thought you know, I'm I'm, still not a hundred percent convinced it is. It is a neat little trick, but how real is that attack? I mean I've, never I mean we can imagine all kinds of things, and so this is a.

C

Is it the solutions looking for problems, what I'm, what I'm asking myself, but, but actually the the previous discussion I, can see a little bit better in the sense that you this attack would allow you to basically have one note disappear from the network and not just have a signal, generator blast, energy and just cut the wire so I, okay, the I'm, a bit more convinced now just wanted to raise. Do something like thanks.

K

To Munich Westin, how do you actually find it correctly? You shuffled a you know the channels based on the keys that joining know doesn't know, for example, or the attacker, which means how do you actually join that kind of network? Because normally you do this that you have. It is one fixed thought that it's our stint a beacon, and that has to be there if that one move somewhere yeah.

G

It's it's a very good phone, so it's something I haven't mentioned today, so we do it.

G

So the idea that.

C

Thomas one I mean the attacker doesn't know the network, it doesn't have the keys right, so you might indicate those numbers in some encrypted portion of some of the messages so that you would learn it during joined. I, agree that with eby's it you know, will be heard, but.

K

If it's, so it's actually not that problem, the problem is, if anybody see the first and Hearst beacon, that's in clear, of course you have to, and it doesn't matter which channel or which you know frame it comes in. You just see it, but then you need to know what is going to be the you know, the initial schedule of various the one joining node and.

G

K

G

Absolutely that is why we do not show for the mini myself. We do not show for the minimum the Stars frame zero, so.

K

You don't stop all all the slots because I mean with that, might not be that you know the first one because there might be multiple slots there. So.

G

Its first loss Primus those first frame 0, we do not show fall because it's the minimal Rose frame right.

C

Thank You Seymour any any more any more questions for Seymour, no any other business you want to discuss before we close meeting where we're at at time.

C

Okay, I mean nothing all right thanks, so much meeting in the meeting is finished. Thank you.

B