Internet Engineering Task Force 97, 16 Nov 2016

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: IETF97-MMUSIC-20161116-1110

Description

MMUSIC meeting session at IETF97
2016/11/16 1110

A

So what I'm going to.

A

Describe to the people I'm going to do that.

B

Very much real.

A

Difference in our.

C

Alright, we're going to go ahead and get started. We have 12 after already, so this is a music IHC, ITF 97, your chairs. We got the yeah note well statement, hopefully everybody's familiar with that one shouldn't be anything new there, administrative stuff as always sign the blue sheets. They should be circulating right now. We have a escribe, that's Jonathan note-taker we have soo has and we need one more note taker. It's pretty easy. It's a short meeting, not a whole lot on their mouth on, look away!

C

No we're not going to proceed before we get a second one.

C

About Brian, would you be willing yeah? Thank you. It's not even safe in the back.

C

When you speak, please make sure to use the microphone and state your name first, and also, if you up here, you know, try to stay within the little box there that works best. With the camera document review request, we are trying to get a bunch of documents. Wrapped up. You know some documents. We have good reviews on some. We do not so just a request again to everybody. You know please try to sign up for document, reviews and I know it's sometimes easier to sign up for them than actually doing them.

C

So you will, you will hear from us. You know when we don't get the reviews, and you know please just try to make it a priority, because we can't really get through all these documents. If we don't get some amount of your view on them at least agenda, we have two items basically on the agenda right now: I usual status updates and then a couple of ice, related issues and I. Think Colin. You had a couple of comments on that yeah.

D

On the bundle draft I just like to spend, insert a couple minutes somewhere near the beginning of this meeting to try and talk about a proposal to move some text from vase up into bundle to clarify some stuff. So a few minutes on that. It's okay with people.

C

Alright sounds good. Anybody have any concerns with that.

E

C

That is a good question is Krista here.

C

Okay, that could be a problem.

D

C

Well, we're going to need them here anyway, we can run through the status stuff- and hopefully Krista will be here by that, but um I assume he's okay with that agenda change as well. So we'll take that as yes, any other comments on the agenda.

C

Martin yeah.

F

So I won't just want to make sure that we closed the deal on 4572 like put a nail in it barriers. Yes, that's everything! So is everyone happy with the plan that we had it's on the list? Okay, the plan is to rather than try to monkey patch 4572 with an update, we will produce a new version of 4572 that contains exactly the changes that we agreed on in list, but it's now a document that obsoletes 4572, it's just complete as opposed to apache thing. I think Collins already done the work for that.

C

We have it on the status lights as well, so well we can revisit their. Let me get to that any other comments on the agenda.

C

All right in terms of the working group status. We don't have any published RFC. Yet we have a couple of jabs still in the RFC editors q, 23 26 bits has now broken 2000 a mile stone, so in all 48, a thousand and seven days to be exact. At this point.

G

Yet so this minus mr. I at least can say 20 to 50. Others has said it's okay to be published, so we're actually just waiting for the remaining trees to sign off on a document all.

C

Right so there's hope, there's hope.

C

Yeah exactly and I guess the second one.

C

And I guess seen in that light, when you look at our cheese peanut, which stands out, I think, is 850 for days is not all that bad actually right. Well,.

H

There you go our this been. Are those authors hear any of them? Darn, ok,.

G

I

Were yeah, they responded before you did the massive update right: they they approved it and then Magnus you what to document and then so we're just waiting for them to approve it again. Yeah.

C

But I think we just got a ping them in I. Don't know you need us to ping them actually or you want thing in themselves or yard. See editor yours.

I

Yet, are you ping them, please, huh can you guys bring them? Will.

C

Do that yeah thanks all right, don't take her. You can take an action for the chairs, please alright msid and mocks exclusive RFC out of the queue you know somewhat short a time. So that's a good thing. Mark's attributes in ITF last call, and we don't have anything in I used to you, review application requested right now. We thought we would have a couple actually, but that turn out not to be the case. Yeah.

C

We had two drafts that are waiting for write-ups right now. Sctp sdp, I, just reviewed that finished this morning looks pretty good Pablo editorial updates are needed, but we should be able to move forward with that. One sdp, neck I, think is in pretty decent shape as well. So we'll expect that one to proceed pretty shortly as well. The red draft is Adam here.

C

Okay, we we tried getting a response from adam s that this one is actually ready to move forward or not and haven't heard back from him. So if anybody sees Adam, um try to prod him as well or it can go, glass call a bundle. We just you know can't seem to get this one finalized. You know formally, it is still in working group. Last call sorry Colin. What's there, no okay, formerly we are still in working group last call.

C

We are really hesitant to reissue a work last call on this draft, so we just want to try to finish up with the comments that are there. I guess, you're, not calling you have a couple of additional comments in there, but really want to just try to you know appeal to people, try to just focus on you know. Let's get this document wrapped up, you know be a little bit flexible and, let's just you know, move ahead with this one.

C

It's just been way too long at this point and I think you offer feels the same way too. So, let's try to focus on this one.

H

Then Adam faith and am just now said: uh yes, I- need to make sure that's done, I'll get them an entry before lunch today. Alright.

C

Cool, thank you. ah Simulcast I think there were a couple of comments on that. One.

J

As well yeah, we have a few more edits that we need to make as as a result of a few comments that we received. Otherwise, it should be good. Alright,.

C

So 4572 update as Martin was just talking about before, so they always thought we already. Then there were some changes, comments, etc. So there is a substantially updated version of the document, at least in terms of the text. That's in there because the overall approach it will now obsolete the old 4570 to update, given that we feel from the chairs point of view that we should reissue a working group last call more formally. So that's what we're planning on doing.

C

Assuming we don't get any comments, then we can just move forward with that one again, dzl ssdp again, we thought we were ready to ask a publication. Did a final review on that one found a couple of issues talk to the authors about that one.

C

They have updated the document since then just reviewed that again there's a couple of fixes that have to be done in there, but given that there were some non editorial changes done as a result of the review, we think that we probably need to do another workbook last call, but just on the changes there again, it's not the entire document. I think the document is in pretty good shape, but again I want to make sure that people in the group are comfortable with what's in there and then I sip sdp and trickle.

C

I sip those are both on the agenda today. So we'll get back to those in terms of new work items. We don't have anything there. Other working group tasks not in the agenda right now or already covered else ora we have already covered MSRP use its data channel I think we are getting ready to move forward with that one. Now that we have the stp neck draft well on its way, 4566 bits.

C

We have just not put any emphasis on this one, because it's not holding up progress on anything else and you have plenty of other documents to focus on that are dependencies, especially on the RTC website. So once we get more of that stuff finalised will will pick up the pace: 145, 6, 26 bits again, SCP, Directorate and designated expert reviews. We have a couple of those as well. Thank you to Dan wing for reviewing a couple of those, and we had a couple of additional ones for their youngest designated Expo to.

C

Last slide here is just basically recap: Jonathan was area equals.

K

So somewhat to the previous slides number to this slide, I believe to Cullen earlier for the a66. This is currently actually a web RTC dependency, some wording reference. Is it what r dr 66 biz yes I mean? Maybe it shouldn't, maybe something? Maybe there something should hear everything perfectly sticks itself, but currently I think Cullen's tree walking algorithm does find whatever this success. Okay,.

C

Because I thought we had covered that previously and we had agreed that that was not the case.

D

So I, ok, so just to make sure I understand, we think that probably web RTC shouldn't depend on 4566 biz, it's just beyond 4566. Is that sort of like that.

C

Yes, that is the chairs assumption right now now. Is there anything in 45-60 six bits that is not in 4566 that you need, and hence we got in this forward, always suggest that you were trying to stay current with the reference I.

D

Imagine it's an accidental staying current with the references that we can go back and look at unless trickle ice in some way needs it. That's the only one I can pick up. That would be okay,.

C

So calling can we ask you to take an action to verify that.

D

Try and figure out where that is by the end of the meeting yeah, whether.

C

You guys, in fact, do you have a dependency on it or not, because if you do.

L

C

We obviously need to put more emphasis on it right now wish. Yes, you don't all right. So for the note taker you gotta! Thank you just.

M

Just on the topic Christian groves, there might be something in the SDP negotiation for with artistic data channels, because there's that new level we're introducing a new level for the SDP attributes, so there might be something with the iron, a registration stuff so and will be some dependency there. Okay,.

D

Christian had exactly right, it is that document that refers to 4566 is that that is our normative. This again is through that data channel document, so I think we'd have to look at that document and figure out. If the clips there was a problem. Okay,.

C

What Colin you will still get back to us with more details around exactly what it is that you need, so we can understand whether we need to push this one ahead or not. Okay, Kristen! Yes,.

N

There's, of course, a lot of documents that way that we very reference you know, bundle and everything which does something with stp, but I, don't remember if we could reference base. So maybe we don't. So maybe it's not an issue, but my point is that in RTC web, if they reference one document which reference 4566 based and you're going to have an implicit reference anyway and then, if you have somewhere 4566 and under place 4566 base, maybe it doesn't matter because I guess there are no major changes in those on major deviations.

N

C

We looked at the documents if we go to the next slide when we look at the documents previously, at least in the last meeting, that we have on this list, which were identified as RT C Webb dependencies at the time, we did not see 4566 this ending up being a dependency as a result of these documents. Now we have not gone through all the RTC web documents to verify that that isn't heat also the case over there.

C

Understood understood so, but at least here we haven't seen that and no I have not heard of anybody else either, but I agree mean ya: gotta trace through that. Of course,.

C

Really normatively.

B

C

O

Yeah max crafter, first 4566 bliss because of the iono update that we did to see how do we for the new attributes, what the procedure we need to follow and 4566. This explains: how could it with the new attribute or operating division category.

B

That's okay, too,.

E

P

Generally speaking with my co-chair behind me, going Ted comes up to suggest it down rough exception and he's exactly right, given that the change here is actually Justin iono procedures, I think a Down rough is eminently sensible and would help us avoid pain and suffering since I'm not in favor of pain and suffering. Let's do get yeah.

C

That sounds very appealing to me. Any 80 comments, objections, disagreements, I.

H

Do not disagree all.

C

I

That document has already been approved, so, yes, you're, gonna, add a Down up now.

P

Ted hearty I'm, not that down rep king but I'm a troll in the bottom of the down roughs dungeon.

P

It is entirely possible to send a message to the IETF mailing list, rerunning the last call for the sole purpose of the down rep, but that that's all you you, you open right. You say we're not taking comments on anything else, but this down rev all other matters are closed. It works pretty well I.

H

Think the after I know the 80s need to look at this and confer with the chairs. If you want to do. Okay,.

D

I just gonna say that the procedural troll will work in the salt mines and bring back an answer to the working group. Alright, so.

C

For the note taker you got that captured as well that we, the chairs, will confer further with the 80s on how to proceed whether we can replace the mocks attributes reference to 4566 bits with 45 626, all right, christer.

N

Yes, I think the 4572 update should now be based on that it did. You know, change with it bit with the discussions with Cullinan and Thomas and everything. That's no gonna be a direct reference. I think I see you.

C

N

C

I think gossamer.

N

Dtls SDP but I don't know, but but at least 45 70 to update oh.

C

Because it's listed as a indirect you're saying that should be a direct, yes agreed.

C

All right any other comments on the OTC web dependencies in the music.

E

D

May not matter once we get rid of 4566, but it has a weird reference to a char set draft from long long ago and it seemed that that one could be converted to the actual RFC I was going to send an email to ally to see if we could get this fixed, but there's another weird dependency problem.

D

Since we discussing all this basically there's a char set document, the draft updated hasn't been updated in like a year now, I think and when we actually looked at the draft in 4566 biz, it looked like it could easily just reference the original RFC. It didn't need to reference the new draft, but right now it has a normative reference to that. That draft.

C

Okay, good, could you send a mail on that just so we don't yeah.

D

Can we capture on actions to or whatever cope with it yeah an HR set reference in 45-60 six bits? Okay,.

C

Alright, thanks come any other comments.

C

All right, I think that is it for the status updates, so Colin. You wanted to talk about bundle, real, quick.

D

So one of the most recent we just sent the J sub draft to working group last call and had one new big block of text about how you take RTP, packets and figure out what happens to the next like how they eventually end up with the right codec, and this involves the how you look at the mids, the rids, the SSR sees and that type of information and figure out what happens now. The the text we have their has some problems with it and a few ideas do it now.

D

What I want to propose doing and I had a good conversation conversations different people about particularly Magnuson, Colin and Jonathan this morning and I think I can put together a proposal a to pull requests to fix this up in a way that I hope everyone's okay with and I want to give people a heads up on that here. So it would take the bulk of that text.

D

That's in section 6 of the J sub draft right now, and it would move it over to the bundle draft on the algorithms that this, the resulting algorithm that you would write if you were implementing a browser, would not change very much a couple. Little fine print on that coming back in a second, but just for working purposes. It doesn't change for a second we'd rewrite the terminology a little bit to term this more in terms of an RTP stream.

D

Explain how the you know you look at the SSRC and from the SSRC number in the RTP packet. That is which stream it's in, and then we write some of the other text to be in that terminology that wood line it up more consistent with the RTP documents.

D

The one change that would be actually made to the algorithm would be to make it so right now we have a line that if you have a payload, if the only information, it's just a summarized algorithm in some cases it can use payload type to determine where which, in section that this would go to- and it only does that in cases where you don't have ribs or mids or SSRC signaling zor, some better way of doing it and one of the things that we have than that is that if there was multiple that it could go to the same, payload type occurred on multiple in sections.

D

The current drought says it goes to the first one and Magnus pointed out that that's just that that's nuts, that the this should never happen like you could only be able to use this hint if they were all unique. So it would be better in the case that they were if the payload types are not unique.

D

The same payload type occurs on multiple infections, and you have no other way to figure out which one of these in sections it goes to it, discards, the packets, that's the only technical change, the algorithm, ah so I expect to do a pull request to bundle, pull requests to to Jason you'll, still be a bunch of texts in jessup that talks about how how this maps up to the names of the objects that we use over there right, like you, know, I'm not going to be in the bundle draft, we're not going to be talking about RTP senders and receivers and transceivers we're going to be talking about em sections and then adjacent trap will say about in section lines up with the names we use that, but the bulk of the algorithm actually moves over to the bundle draft is what I'm, proposing so I put together, pull requests.

D

I will put together pull request for bundle and forge a set at the same time, put them on both the lists, and then people will need to review them. I would like to move relatively quickly on this. People have thoughts or comments with that plan on understand. People have to see the actual text to agree to it, but that that plan sound like a reasonable approach to people.

D

I mean the whole algorithm moving to the bundle to start with really.

Q

Adam, okay, that sounds good to me: okay, okay,.

D

That's it. Thank you.

R

All right, Krista.

J

Europe, which one you guys talking this.

S

S

N

N

Okay, this is for a few issues for, for the ICPC p draft I'm, not the main editor of this draft. Unfortunately, I didn't have time to go through these slides with, and they know author, but these issues that I'm going to raise there have been raised us on the list. There's been some comments on it. Actually, the most of them came to my mind when I was talking with the BF CP people about helping them to fight, try to finalize their text.

N

But it's not really BFC be specific. It's it's! It's stp over answer specific. So so, let's take a look at those. So next slide please- and this is just to make this is not. This is eyes hdb specific! That's why we talk about it here and not in in in dice working good. Okay. So first is the default candidate in sdp answer.

N

First couple of background in according to offer answer, even though, as I said there, it's not actually specified but I think it's what most people as you is, the m-line in the offer announcer must match so I did the proto the transport in the m-line in the offer. An answer must not sure if you put transport X in the offering the same line in the corresponding game line in the answer you need to have protocol X.

N

Also, what with what the draft I seep stp says, is that the c & M line, I, guess it in the boat offer? An answer must contain the same default candidate.

N

The reason why why it's required in in offer is because we don't know whether it is going to be used in the first place, so you need to put something in the c & M line that works, even if the, even if the answer doesn't support the one to use sighs. The problem is that you also have the same requirement for the answer and there is really no justification for it and the problem that can occur. If you go to the next slide. Please it's what it's offer. An answer have different.

N

The offer has one default candidate and the answer has another one and they use different transport. Now we say that you must put the default in the CNM line, but if the answer has a different transport, it can't put it in the CNN line because of the tree 264 rules. So the only thing you tag can do is basically reject the whole em line, which is kind of stupid, because I mean it wants to do eyes and there are candidates with transport that it does support. So it is able to establish a session.

N

So this is just a 32 64 restriction issue. So basically my suggestion is there that we basically remove the requirement to insert default candidate in the comm line on the stp answer. The answer knows: ok, we're going to use eyes so so it puts whatever their that matches their mining offer and and we're done and then he puts the candidates because the candidates are going to be used in which actually establish the transporter. They see an inline doesn't really matter, but it has to be there because the 32 64. So that's my suggestion.

N

T

I approve this message.

E

Allen fourth I mean I mention it's on the list as well, and yes, I definitely support making this change would be a good thing. We just echo what in the offer and ignore it.

K

John likes I mean the one issue here is the other thing this does. Is it's also for detecting ice mismatch? Okay, so you've got an spc, that's mucking, with your CLM lines. It doesn't understand ice. That's for detecting this I'm worried that'll break that mechanism. um I, don't know it was I assume was put in ice for a reason.

U

K

Know but I mean I, think that you know, and yes spc shouldn't, be passing a equals lines. They don't understand, but they do so.

K

You know, I think, that's that's why you know put the ice mismatch. Stuff was in here and I'm afraid, it'll break that now mind you. I think that the candidate lists here I mean so. The worry you have I guess still early for the PFC p case is that the answer doesn't have any candidate that matches this transport right because I mean it pretty.

K

If you're doing, if you know you're doing guys what candidate you use doesn't matter as long as there's some candidate that match the transport, that's okay, but I guess the where you have is you know: I offer UDP and TCP PFC p, defaulting TCP. Does that only does UTPB FCP? Actually now now.

N

Now, Rica that yeah.

L

N

Case where the other endpoint doesn't support a protocol to begin with, that's my next issue, so so.

H

N

So it will come back to this one right.

H

N

Basically, regarding the mandate to put the default candidate in the m-line on.

K

The alligator it's the ice mismatches, the case you care about their if.

N

You care about that and I'm not sure we do, but maybe we do and based actually, but maybe I should have taken other issue first because, depending on how we move forward with the next issue, maybe this issue is will big will actually go away. So so I mean we yeah.

L

So the issue is it's actually not the default candidate, which is in the answer. It's the selected candidate, it's the selected candidate which goes into the answer. So slight correction terminology, I, think the original explanation why this thing was needed is: if there are anything on the line which monitors the signal in exchange, it doesn't understand ice. It needs to know which transport which r.I.p import the being used for communication. So it can get this information from and CNM line, and if we remove this requirement, they no longer will need this.

L

And so that's why the requirement is there and basically, what happens is if we, as I said, if we're switching from UDP to tcp, then I will have different transport protocols in am line which causes the mismatch which breaks the 32 64 rules. So that's the issue and again I know that requirement that the CNM line is supposed to match, what's actually being used for communications, for some external monitoring tools, then raced multiple times on the list, but I'm not sure if it actually affecting anyone.

V

Yeah I mean I other existence. Anybody is doing anything useful with this. Any of this information I mean the bottom line. Is that the instructions in arson 5245 tell you to put in like the like, at least what I thought was actually the least likely um you know set of parameters, namely the relay parameters, and so like, like these mismatches are constant, and you know so on. You know that the like I strongly suspect putting chicken chicken chicken were working here and, if not putting chicken chicken that I'm fairly confident putting like exactly that's it.

V

You offer put in here or work um and like I, think it's really incumbent anybody thinks it like this stuff is doing anything useful to like produce some evidence. That's the case and.

N

I think just I mean in most cases I don't think there's. This is anything even going to be an issue because, for example, RTP I assume that people are going to put the transport in them and I think it's going to be very rare that you put, for example, tcp in a transport. The same thing goes for the data channel. Actually in the draft a ctp DTLS, we say now that you should use UDP and tcp should just be the last.

N

You know the worst case, but this again came up with BF CP because, as far as I understand, maybe FCP UDP and tcp has equal weight. So to say there is no one which is more, you know, and then the other. So I think that's where, where this issue could could come up and as far as I know, in bisbee FCP endpoints don't have to support both tcp and UDP. They could support just one of them, but then we come to them no more today soon in my next issue.

N

So if you don't support the protocol to begin with, so for the transport.

Q

Adam Roche, going back to what Donovan was saying, I think that he like gave an answer to the problem, but then kind of breezed past. It I think if you just say we're changing mismatch detection for answers, so that if it doesn't match any of the candidates, if the mismatch as opposed to it, doesn't match the default candidate is a mismatch and then everything works. I think that makes Jonathan's problem go away now.

K

Now the this CNM line is what defines what the default candidate is. This matches, when the CNM mine don't match any candidate, and that means you've gotta be WA and between its mucking with your stp, but leaving a equals caturday didn't without understanding it one. It's understood your issue well,.

N

I mean, I think the CNMI should obviously match some candidate I mean otherwise. Yes,.

L

N

We can, of course, yeah we have. We cannot take saying that there I mean it can say that the CNN line must must match a candidate. We cannot.

K

Say that's what it says when did the problem, like you said, is the BF CP case where you can't you know, is the cnmi I mean. If you say see them on, must match some candidate and must use the same transporters in the offer problem. The problem you want to do is the DFC pk's. Yes,.

N

So maybe we should go to the next issue. We can come back here because I think I described that that that's the next issue, where you don't support the transport begin with. So do you want to comment on this or you wanna make okay, so some next slide, please okay, so basically did is the.

C

Dress turn you know so, for the note taker sounds like there's no clear conclusion on this. One is ok.

I

N

Got finally okay, so the basic the background is for the same thing here, then the transport in the arson in the answer has to match the offers an excellent please.

N

So this is basically what if the answer does not support a transport in the a.m. Line of the offer to begin wit- and I guess here we're mostly talking about the BFC- be the BF CP- contains an offer in the CM line. You have, for example, UDP, and then you have a candidate for tcp, so the offer does support boat, UDP and tcp, but in the end line it puts the UDP and then it has a tcp candidate. Now, let's say that that answer he only supports tcp and basically yeah.

N

That's fine, though, for sent me a tcp candidate. So we can, you know, establish, but again you have this rule that the answer damn line in the answer must have a same transport. So basically the answer has to put UDP in the ACM line of the answer, even if it doesn't support UDP begin with so can I can I gesture, you know so so so now it's a next slide. Please so basically I had two alternatives of how we can solve this.

N

First, we could define that every protocol supporting eyes must define a mandatory to support transport again for RTP I think this is the factoria you. If you support, if you do, are to be used for two DP, you may sue participe. If you do the data channel, you support you defeat, but again with BF CP. You cannot assume that all endpoints are going to support both UDP and TCP.

N

So basically, that's it. We define a mass support, transport and but the alternatives to is basically we allow the answer to put whatever in the CM line and because again it knows that ice is going to be used, so you're actually going to use the candidates then to actually establish the connection.

N

So I know this is a little hacky. You put something in the CM. I mean this is the thing you know we shouldn't use 2 cm long to begin with when we use eyes, but but but you know, this is what we have so so yeah I, don't.

K

Exa me know, the third possibility is, as you said, there's you know, you're just inferring, that the transports have to match the offer and the answer I mean evil anything break. If I mean. If we want to say that if you're using eyes, they don't, you know they have to semantically match, but you know they don't have to actually match, because the default candidates might be different. Yeah.

N

As I said, there is nothing in 32 64 and please correct me if I'm wrong. That say that the transporting offer also has too much I was looking and I'm right.

K

So that there is a wealthiest about.

N

K

So so hey, let's say it doesn't have to match and.

N

Well sure, if the community can agree on that, but I think that that's something we need to agree on, because a lot of people are talk to you are assuming yet the Transport has too much in the offer answer. Not only the transport but I hold the whole proto proto value. So so so, if we want to room, you know, maybe that's true. Maybe we should update to 26 for them to clarify that, but but that's a separate issue, but but II.

N

If we make that assumption, we should have let's get it into the nose I. Think of this meeting.

T

Charles a call so I, I agree with you.

T

I kind of have that assumption that the what you put in the inline the transport needed to match, I think, a lot of people and to have that built in so I think it's safer to remain with that, but relaxing the rule checking that you do is to like when isis, in effect, I think, is okay, and I thought that was the way it was that's why ice is written right and then, if you happen to like, if you say, UDP in your m line, but the ice candidate you end up using is TCP I thought you had to send re-invite yeah.

N

T

N

Think today you don't have to send entry in white, but I mean that's after the real fur answer. First operons, the transaction is done right. That's when you basically update to to write another middle.

T

Boxers I thought yes,.

N

I think they issued, but by relaxing this because let's say that I offer you UDP I put that in a transport and then you send me back, for example, tcp in the end line of your transport. Let's say we don't do eyes, you know how do you know that I support tcp?

N

You do now sighs support UDP, because that's why I had in my not offer. But now you send tcp back because that, but so you don't know if I support it.

T

No, so what you put in the m-line, you keep, if you said, UD p and the m-line answer in the answer. You keep it as UTP, but you left. The ice negotiations still happen. Yeah.

N

So you mean yeah sure yeah. So that means that I put UDP a name line on the answer, even though I may not support it. To begin with, because I do support the tcp that you hadn't you in the candidate right.

E

I'll unfold on you have just to echo that number two seems the way to go here. Clarification here would be useful and yeah everything will be better.

K

Yeah, like a you know, did discard port known invalid address and you know priority 0 candidate.

N

Yeah well, that brings another question. If he put not tcp in the in the m-line, do you have to put a dummy candidate also pointed to TV I? Don't.

H

N

I, don't think yeah but yeah, but I. Don't think what why should we do? That I mean it's.

K

Because, at least in the 5245 roles, if it doesn't match you're, not it's, it's a mismatching you're, not very nice. The cni mine has to match one candidate.

N

Okay, but maybe we shouldn't remove that restriction to Italy in the answer. I I do get it in the offer, but but but yeah.

K

N

Mean again, we have to remember that for most uses, this is not going to be an issue. I know.

H

I agree for rtb these.

N

Diseases at this moment, I can only think of B FCP, where it is made, become an issue cause.

L

You want something edible, that's right! ah First of all, we're still having an issue where we have a backwards compatibility where, for instance, when you're sending did you fall when the remote party does not support ice and we send an offer with CCP and it only supports ugp. That means it will fail, even though we technically support tcp. So that's issue number one which, by still dealing with issue number or the data proposal which I had I'm not sure if it's mighty little bit radical, is to also add the ice transfer tag.

L

So when we basically don't want to deal with selecting, UDP or TCP just put I so RT dtls srtp and in the offer, and put the same in the answer and put a bogus IP address in there and don't deal with default candidate at all, because.

N

L

N

If you put the so called the ice transport in the m-line of the offer, then you don't have backward compatibility. If the answer doesn't support ice and the first thing you said: if, if the endpoints don't support the same transport to begin with, I mean obviously, then it's going to fail, there's nothing. We can do that, no matter if they do ice or not, I mean if they don't so.

N

If I only support you to be a new, only support disappear as a session I mean I'm not trying to fix that, because there is not fixable. No.

L

But I'm just do what I was saying is when the offer offers support both. What's a UDP and TCP, it makes a decision of putting UD be in the offer, and the answer only supports DCP. It doesn't support ice. That setup fails, but it's kind of a use case. One use case 2 is where you where you don't need any way. You actually don't only want to support ice and with that's what we do with WebRTC. We're with technically are putting a bogus IP address in the default candidate.

L

So this whole dance, which will have with the default, can do this kind of pointless. So if we can just indicate it clearly that there is no default candidate there and we have a different transport, we kind of bypass that problem for cases where no nice use case is not supported to begin with. So those are two independent, but the first one is an issue in the second one is a proposal for cases when I when I see is required. I.

N

Agree- and you know in the bar t see, another usage is so this is fine to mandate, but I don't think in our documents we want to in this generic ice. Stp I. Don't think we want to assume that the other endpoint supports eyes. I mean we want to have this backward compatibility case, but then.

L

Essentially, what you probably want to do is, if you end point support both you GP in TCP. You want to put to em lines, one which would be a TCP and other, which would be u GP and have some way of basically the answer, picking one: that's the that's the backwards compatibility case that you're trying to achieve because the be backwards.

N

Compatible, yes, that's all stp kapnek I mean it's.

L

Anderson but we're still kind of try we're trying to solve the problem which doesn't necessarily souls itself quite quite cleanly right now, because, even if, with the issue that you raised is basically we are generating an offer. We need to pick the protocol, but we don't know which one they answer will support and it doesn't work for the oil and we kind of have sold it when they answer just happens to support ice.

C

Crystal just a time check, you got 15 minutes left total. So you decide how you want to spend that time.

S

But anyway, so I'll.

E

Just make a very quick answer to Roman's point in depth that problem it is today you have to make a guess at the surface. It's a way it's done yeah. Maybe you should have done to em lines, but that's not the way it's done. So we find it. R is over.

N

Yeah so I mean we discussed this multiple m lines in the past many years ago for all different kind of reasons. If you, for example, a if you want to offer about a vp and sav p and all this kind of stuff, but but and obviously you have bundled, but that's not really fit because you don't really doing multiplexing here, you differ you're, giving different alternatives and the mechanism we define for that was captaining. So so.

N

Yeah but anyway, so so okay, can we move on I just like to go through everything? No.

R

I was always trying to get this to resolution. Okay, it's Nobel. Okay! Can we can we agree that this is all about an answer in answers? Behavior, okay, we have to I send points. We know what the offer will do that we're wondering how to answer okay, and we have basically two possibilities- a answer with a default candidate that doesn't exist or be use a candidate that is in the list, but relax the matching constraints think that those are the two choices and we can let vote that. Maybe hum.

T

Charles Echols I think we have a lot of examples from the past where, where the answer can be, do something that maybe isn't entirely correct by the standard when it has enough information from the offer and what it knows of its own capabilities to do the right thing and the way we've always defined, it is put the right thing in the offer for the m-line so that for backward compatibility, something that doesn't know this new thing, whether it be ice or feedback parameters or rtcp here or if it's using security that securing the RTP the right thing happens so, for example, putting a vp but letting someone answer with leaving the m-line as a vp, even though they put feedback parameters in there are sending a VPN and using security.

T

You know you should have put sav key, but you didn't because it might, the call might fail. That's that's. We've done that a lot in the past. It's always worked out, okay, it's too bad that we had to do it, but it seemed better than kath. I think this is just one more example. Take that same approach do the thing in the offer that's going to work if for backward compatibility, if the answer it doesn't support, the new thing doesn't support ice. Then you know.

T

If they do they let ice do its thing and and just echo back the inline what you got a need in the offer so and if we have to send a reinvite great, we do that that way, the middle boxes are happy. We've done that a lot boys.

J

Roman colleges we just plus our town here Roman, go ahead.

L

Again, just I think the better solution is just to have a must implement transport for the protocols right now. The only thing which is kind of out of that is be FCP and that solves all the problems with legacy it solves all. The problems with current stuff so might be easier to.

N

Do it that way and.

L

You would have to.

N

Say that you must use that as default candidate. Also because you.

L

Might basic you and you will say that you must use it as a default candidate to no for an answer, and once you pick something different for eyes, you do another. We invited to move to the next one and.

L

D

That would work fine as long as no one had shipped the FCP yet, but given people have shipped it for over 10 years, changing it in any way whatsoever is impossible for this case. So that's the bad news. This is the really good news for everything that does ice.

D

It absolutely doesn't matter what you put in this field right, because it ignores it so I write it well and so for what we should do that for so what we should put in here is the thing that will cause the least surprises for stuff that doesn't understand it, um which is just you know, it's exactly what Charles said is just like copy the data back in there, and I don't know like we've- we've discussed this, probably what every six months for the past four years and always reach the same conclusion: I, let's kill it I mean that's that that it, you know which choice which of these two you you nicely put what the choices were, which of the two choices we choose doesn't make any difference to someone who does ice.

D

So we're only talking about what happens, to things that are on path on the signaling path and don't understand ice and the thing that will cause them least surprises echoing the value in it's the conclusion we reach every time we discuss this so.

N

Basically, what you're saying here is alternative to. We can't make alternative one, because you know it's too late for BF c p at least alternative to is basically put whatever in the inline or does it.

D

Put in thief, if you're doing ice in the blank, a relevant field which no one reads copy in the data that was came from the other direction, that's on proposed. It.

C

All right so in the interest to try to close something here. Can we get a hum of people in favor of 2 and people opposed to two? So if you are in favor of option to hum now, if you are post option to Amell all right for the note-taker option through please so Krista, you got nine minutes what you want to yeah.

N

Just to the previous one, I assume this also means that we're going to remove the requirement to put the default candidate, the damn line, adults or because you know we just know said that yeah just put something there I just like to get that didn't entry node I mean that was my suggestion of the previous. If you go a couple of slides back.

N

No, that one more yeah.

N

There's a spelling error: they should be removed, but basically we're not going to say that that you have to put their default candidate in am I, not answer. Alright,.

C

So again, just asking real, quick anybody opposed to the proposed solution here. John.

K

Dies, I think that implementation is doing a strict 5245 implementation will have a problem with that. So I would propose instead add a it mean you have to put something there and it has to match a candidate. So in these weird transport mismatches cases, which is the only case that matters make it- you know, put a you know: priority 0 invalid IP address dummy candidate, as you know, to the list just to keep things like that happening. Keep.

N

In mind that we have this ice to option this ice option, which is going to be in the offer, so if you think there's something which is doesn't work backward, we can always make that you know this is something you can do. If you know okay, they offer.

H

N

They send that song.

D

Come on, Jensen I can easily live with that, but don't we didn't, we always previously put whatever you thought you were using was what you put in here, but.

N

I thought we just said in the earlier slide: we will be harmed for alternative to which base a different field.

N

No, this is what the support right, no I'm, the boat. Both issues are barred to transporting them line right.

D

Address partridge, okay, yep, it's copy across.

N

C

Could you restate the conclusion because I'm not sure I understood how Jonathan modified it.

K

So my suggestion to avoid ice mismatch so I mean so my suggestion was that you know, meaning you have to pick a candidate. Doesn't you know either it doesn't matter which or columns? You know the one you think you'll be using if you're in this case, where you have the transport mismatch, include a dummy. You know, you know zero priority dummy candidate of the correct transport. You know with lower priority than any of your of your other ones. I think.

N

So you're basically okay with this, but you want to put the stomach and atomic.

K

A today and just satisfies mismatched issues.

C

Okay, yeah, it.

V

Just seems a particular enormous ounce of Hoops to fix something which is yet to be established. If your problem, like, if you're nice, miss match the connection will fail, and you will notice that and so early lore early warning on that does not seem like a particular important function. Is the ethicist at? Was anybody actually? Does that like, like again I, think it's really incumbent on you like demonstrate? This is like something we actually is were thinking any accommodation for whatsoever.

C

Alright, so I think we still don't have a clear conclusion on this one and if you want to spend you, can you spend your last five minutes trying to get to a conclusion, I.

N

Like to move on to the next one, I I think we have I, think we more or less agree. I think your jonathans issue to make sure that that that maybe we need something for backward compatibility, but again we can use that. Maybe we can say you know if you if there is, if, if this ice to option is not there, you have to put this dummy candidate. Otherwise we can remove that requirement also, but but I think we uh we.

H

Have a movie part.

N

Forward I, don't have text here, so we need to do pull request and so on and see that so yeah. Let's.

C

Get some text- and you know, discuss that on the list, then you know Acker I hope that satisfies you know your concern as well.

N

Okay, so move on, because that I think that the last thing is important because it so three slides forward.

N

One more one, more one, more yeah: this is actually. This is a text that we put together with with Roman when we were writing the AC TPS DP draft. But then we realize that this is generic ice generica ssdp, so it should be put into that document.

N

So, basically that some background again, you have two endpoints that negotiate this media session. You know they may have UDP TCP base candidates, and you know before nomination you based on removal of addressing of all that so before the net nomination, that the the media can be sent on any valid pair and and you can switch between candidate pairs before you have done denomination.

N

You know if you trickle, it may take a long time before you do nomination and so on and as you switch between candidates, obviously the transport may switch, because if you have both tcp and UDP candidates, so next slide. Please.

N

Ok, now you did you go backward or is this.

N

Yeah go yeah, one more okay, so that no, no one okay! So so this is basically rafin here. A couple of our I don't want to change what what I just had their. This is just a clarification text, so, first we we think that it's good to clarify that when, when the endpoints, which is transport, you don't need to send an offer on to offer to reflect that, because that means during nomination or before nomination.

N

If you switch between UDP and candidates, you would have to send a lot of updates before the call hasn't even been. You know accept it. So we want just I. Don't think anyone would do that, but I think we need to clarify that to make sure that you don't need to send a new offer. The m-line is what it is, and you know be. While you switch, you can switch, try the transport, but then, when an offer is sent for another reason, and this could off.

N

This could of course be, for example, once you've done the ice nomination, but maybe you're sending an offer for a cone ice related issue. Then, within that M line. Should always match the currently used transport, then not denominated transport I guess we should call it. So if you, if you sent the initial offer, for example, using UDP, but then you're going to end up using a TCP candidate, so next time you send an offer, you should actually have tcp and a month.

K

Drop lines I think your question isn't possible because I mean, by definition what something's nominated you know what candidate parrot uses, which is using a certain can't transport right, so I think your question could only happen if you're setting an update before nomination.

N

Yeah, but is it mandatory to send update once nomination, is done no and then at all I mean mate I.

D

Don't know and then now we're coming.

N

Back to the I know, you have a shoes way to expect with there's been talking about these never doing the nomination. I guess.

K

N

Discussion doesn't get.

K

Also so I think is your bullet point question under to their only applies before nomination after nomination, by definition, are using the same same transport, because domination has sure.

N

But I think it would be good to say that off denomination dead am line always have to match to nominated yeah.

D

N

D

N

Maybe some because I just want to avoid a case that someone sends UDP in the initial offer and maybe they nominated TCP than them. But that's.

H

N

It keeps their the UDP and then.

K

I'm gonna should avoid.

N

That so this is just a.

K

Terrifying, you thank you Sam that should match after nominations, and you said that okay.

N

C

So is no more comments. Sorry, we actually open time already is their way a clip quickly reaching a conclusion on this one or.

C

Yes, somebody objects, no, sorry! Alright! We are out of time. Unfortunately, so we will have to continue on the list.

C

And I guess we need to ask for more time next time.

Q

Maybe one final administrative thing I signed up to do a view of this I. Did it it's taking a long time to type it up, because it's generated a lot of comments, but I hope to have something on the mailing list by the end of this week or maybe sometime next week, all right. Thank.

J

You do we have the blue sheets.