[01:09:43] Christian Huitema leaves the room: Disconnected: closed
[01:09:43] Christian Huitema joins the room
[01:12:36] Christian Huitema leaves the room: Disconnected: closed
[01:16:02] undefined joins the room
[03:00:08] zulipbot joins the room
[04:40:44] Christian Huitema joins the room
[08:37:29] Meetecho joins the room
[08:37:59] Christian Huitema leaves the room
[08:45:02] masque joins the room
[08:50:03] Dallas McCall joins the room
[08:50:03] Nicolas Kuhn joins the room
[08:50:03] Mihail Zverev joins the room
[08:50:03] Brendan Moran joins the room
[08:50:03] Daniel Havey joins the room
[08:50:03] Lucas Pardue joins the room
[08:50:03] Chunshan Xiong joins the room
[08:50:03] Tobia Castaldi joins the room
[08:50:03] Xavier de Foy joins the room
[08:50:03] Maxime Piraux joins the room
[08:50:03] Alex Chernyakhovsky joins the room
[08:50:03] Paolo Saviano joins the room
[08:50:04] Loren McIntyre joins the room
[08:51:12] Hirochika Asai joins the room
[08:51:35] Paolo Saviano leaves the room
[08:51:42] Christopher Wood joins the room
[08:51:45] Markus Amend joins the room
[08:51:55] Ted Hardie joins the room
[08:52:04] Sean Turner joins the room
[08:52:10] Kenneth Murchison joins the room
[08:52:28] Maxime Piraux leaves the room
[08:52:32] Maxime Piraux joins the room
[08:53:14] Martin Duke joins the room
[08:53:29] <Martin Duke> test
[08:53:53] Mike English joins the room
[08:54:06] Robin Marx joins the room
[08:54:41] Mark McFadden joins the room
[08:54:47] <Lucas Pardue> test timed out after 30 seconds
[08:54:52] Nancy Cam-Winget joins the room
[08:55:16] <Robin Marx> Are there any scribes yet?
[08:55:27] David Schinazi joins the room
[08:56:00] John Border joins the room
[08:56:02] Massimo Nilo joins the room
[08:56:04] Jonathan Lennox joins the room
[08:56:12] Martin Thomson joins the room
[08:56:14] Adam Wiethuechter joins the room
[08:56:16] Tommy Pauly joins the room
[08:56:46] <David Schinazi> Hi everyone!
[08:56:54] spencerdawkins joins the room
[08:57:14] Massimo Nilo leaves the room
[08:57:23] Massimo Nilo joins the room
[08:58:01] Philip Eardley joins the room
[08:58:01] Jörg Deutschmann joins the room
[08:58:19] Magnus Westerlund joins the room
[08:58:20] Massimo Nilo leaves the room
[08:58:30] Massimo Nilo joins the room
[08:58:34] Benjamin Schwartz joins the room
[08:58:38] <Lucas Pardue> the silence is killing me, where is Magnus and his Marillion?
[08:58:40] Stefan Santesson joins the room
[08:58:50] <David Schinazi> Right! We need some musak
[08:59:03] Alessandro Ghedini joins the room
[08:59:09] <David Schinazi> I did that at WEBTRANS at 108 and folks didn't like it though
[08:59:13] <Lucas Pardue> are you calling Marillion musak? Harsh!
[08:59:17] <Martin Thomson> I wonder if Byron managed to fix my ICE woes.
[08:59:23] Mark Nottingham joins the room
[08:59:27] <David Schinazi> Oh no that's not what I meant Lucas
[08:59:30] Eric Kinnear joins the room
[08:59:36] DanielHavey joins the room
[08:59:42] Yanmei Liu joins the room
[08:59:46] <Martin Thomson> Fantastic!
[08:59:58] <Lucas Pardue> the rumor mill doesn't care what you meant :P
[09:00:00] Marcus Ihlar joins the room
[09:00:03] <David Schinazi> :)
[09:00:04] <Jonathan Lennox> MT: If you know, what was the fix?
[09:00:04] <Martin Thomson> Now I have to work out how to properly thank Byron.
[09:00:05] Alan Frindell joins the room
[09:00:10] Chi-Jiun Su joins the room
[09:00:14] Brian Trammell joins the room
[09:00:23] Philipp Tiesel joins the room
[09:00:25] Mirja Kühlewind joins the room
[09:00:32] Gorry Fairhurst joins the room
[09:00:38] Marten Seemann joins the room
[09:00:42] <Martin Thomson> Jonathan: it's a corker.  Basically, there was a race condition and we were failing all of ICE when the first candidate failed if that race went the wrong way.
[09:00:44] <Martin Duke> @MT we'll be sure to change Meetecho enough to break your audio again before 110
[09:00:45] Eric Kinnear leaves the room
[09:00:48] Eric Kinnear joins the room
[09:00:49] Mike Bishop joins the room
[09:00:59] Bron Gondwana joins the room
[09:01:05] Eric Kinnear leaves the room
[09:01:08] <Martin Thomson> I got to learn a lot about their deployment model.
[09:01:08] Eric Kinnear joins the room
[09:01:09] Hiroyuki Goto joins the room
[09:01:11] David Oliver joins the room
[09:01:19] Zaheduzzaman Sarker joins the room
[09:01:24] Valery Smyslov joins the room
[09:01:26] Roman Danyliw joins the room
[09:01:27] <Jonathan Lennox> Race in FF or in MeetEcho?
[09:01:29] Hiroyuki Goto leaves the room
[09:01:29] Dan McArdle joins the room
[09:01:31] Hiroyuki Goto joins the room
[09:01:34] <Martin Thomson> Race in FF.
[09:01:35] mnot joins the room
[09:01:35] Jana Iyengar joins the room
[09:01:46] <Martin Thomson> Probably down to a peculiarity of the way that Meetecho was trickling candidates.
[09:01:51] Gorry Fairhurst leaves the room
[09:01:52] <Brian Trammell> sliiiiiiides
[09:01:52] Eric Rescorla joins the room
[09:01:53] <Ted Hardie> Big slides!
[09:02:14] <Adam Wiethuechter> I was not 100% paying attention and all that didn't sound like what it actually was....
[09:02:23] <Meetecho> Martin Thomson: glad the issue is fixed! I shared some more info on trickling via mail
[09:02:24] Alissa Cooper joins the room
[09:02:31] Toerless Eckert joins the room
[09:02:32] <Adam Wiethuechter> 4am does weird things to my brain and how it processes conversations
[09:02:49] Pete Resnick joins the room
[09:02:54] Alessandro Ghedini leaves the room
[09:02:57] Alessandro Ghedini joins the room
[09:03:13] ihlar joins the room
[09:03:25] Dragana Damjanovic joins the room
[09:04:16] Victor Vasiliev joins the room
[09:04:21] Victor Vasiliev leaves the room
[09:04:25] Victor Vasiliev joins the room
[09:04:25] <Robin Marx> I can scribe
[09:04:37] <Robin Marx> would be useful to have someone who actually knows thing about proxying to read through
[09:04:39] <Robin Marx> notetaker
[09:04:43] Jiao Kang joins the room
[09:04:45] <Eric Kinnear> &lt;3
[09:05:13] Magnus Westerlund too joins the room
[09:05:44] Kazuho Oku joins the room
[09:05:45] Igor Gashinsky joins the room
[09:05:54] <David Schinazi> I can't hear Tommy
[09:05:55] <Mirja Kühlewind> Tommy you are breaking up
[09:06:01] <Martin Thomson> A d o is u s
[09:06:02] <mnot>   ommy is ard t ear
[09:06:14] <Mirja Kühlewind> what was the proposal now?
[09:06:29] <Lucas Pardue> I am your Jabber scribe for this session. If you would like me to take a question or comment to the microphone please prepend it with "MIC" or @ me
[09:06:33] <Tommy Pauly> Proposal was to group documents by topic
[09:06:34] <Mirja Kühlewind> that makes total sense!
[09:06:35] <Eric Rescorla> I'd actually prefer to discuss the WG items together
[09:06:39] Gorry Fairhurst joins the room
[09:06:46] <Mirja Kühlewind> good catch tommy
[09:06:54] <Eric Rescorla> And the proposed items after
[09:06:55] <Jana Iyengar> chairs, decide, and move on
[09:07:01] <Mike English> Live demonstration of unreliable transport
[09:07:02] Stephan Emile joins the room
[09:07:03] <Mirja Kühlewind> impressive how awake you are
[09:07:04] Kirsty P joins the room
[09:07:12] <Martin Thomson> If we are going to make progress on WG items, let's do that as much as we can.
[09:07:12] Yanmei Liu leaves the room
[09:07:19] Yanmei Liu joins the room
[09:07:44] Kazuho Oku leaves the room
[09:07:46] Kazuho Oku joins the room
[09:07:55] Igor Gashinsky leaves the room
[09:08:06] Frédéric Fieau joins the room
[09:08:26] <Mirja Kühlewind> @lucas I thought this was your chair's hat?
[09:08:44] Alessandro Amirante joins the room
[09:08:53] Paolo Saviano joins the room
[09:09:03] Toerless Eckert leaves the room
[09:09:43] <mnot> that looks familiar. Can't put my finger on it...
[09:09:53] <David Schinazi> @mnot :D
[09:09:58] Paolo Saviano leaves the room
[09:11:31] <mnot> I'd like a way to visualise the difference between support for major features and corner cases.
[09:11:45] <David Schinazi> That's a good diea
[09:11:46] <Alan Frindell> impressive first interop
[09:11:49] <David Schinazi> idea*
[09:12:28] <Tommy Pauly> Obscurity: the easiest security
[09:12:35] Dirk Kutscher joins the room
[09:13:23] Benjamin Kaduk joins the room
[09:13:57] Joerg Ott joins the room
[09:14:58] Frédéric Fieau leaves the room
[09:15:15] Harald Alvestrand joins the room
[09:15:15] <David Schinazi> To be fair, the fact that you need to implement a draft is enough to discourage attackers :P
[09:15:26] <Sean Turner> hahah
[09:15:32] <Brian Trammell> :D
[09:15:39] <Brian Trammell> security by internet-draft
[09:16:06] <Tommy Pauly> That's an interesting model for encouraging implementations... run an open proxy and see how long it takes people to write an implementation that works
[09:16:10] Michael Scharf joins the room
[09:16:24] Jonathan Lennox leaves the room
[09:16:33] Igor Gashinsky joins the room
[09:16:42] <Marcus Ihlar> We have an implementaion based on aioquic. Only tested locally, but would ready to try interop soon as well.
[09:16:43] Nitin Batta joins the room
[09:16:58] Joerg Ott leaves the room
[09:18:15] Jonathan Lennox joins the room
[09:18:34] <Lucas Pardue> I sort of indicated when I spoke, I like the grid but it kinda gets lots in trees. For hackathon/interop it might be nice to have a headline I got protocol foo over MASQUE speaking
[09:18:44] <Lucas Pardue> ROFL
[09:18:53] <DanielHavey> You did say, "Next slide please."
[09:18:53] <Sean Turner> I almost bought that first slide :)
[09:19:01] <Kirsty P> way funnier  than the tweet slide
[09:19:13] <Igor Gashinsky> Well done :)
[09:19:18] <Brian Trammell> the second slide is actually a pretty good logo for Masque
[09:19:19] <mnot> +1 - I approve of chair trolling
[09:19:34] <Brian Trammell> (in the tight, twisty tunnel sense)
[09:20:02] <Pete Resnick> Would have been perfect if Rick Astley was sliding down the slide.
[09:20:04] <mnot> Now I just want to see Schinazi reactions
[09:20:16] <DanielHavey> David has slides on the brain now.
[09:20:49] <Lucas Pardue> also worth highlighting: the QUIC WG has typically relied on HTTP/0.9 over QUIC for exercising the transport features. Carrying that over MASQUE is a bit limited / boring
[09:21:05] <Martin Thomson> who is going to own the h3 datagram shim?  httpbis?
[09:21:17] <mnot> I believe he's getting to that...
[09:21:25] <Tommy Pauly> Indeed. Chairs have discussed, though.
[09:21:28] <Lucas Pardue> sframe
[09:21:29] <Martin Thomson> liaison statement time~!
[09:21:45] <Martin Thomson> lucas, if you weren't on the other side of the planet with a travel ban...
[09:21:48] <Tommy Pauly> Oh yeah, we'll have 3GPP define it...
[09:21:49] Benjamin Schwartz leaves the room
[09:21:52] Benjamin Schwartz joins the room
[09:21:59] Benjamin Schwartz leaves the room
[09:22:14] <mnot> our need meeting is at MWC
[09:22:18] <mnot> next
[09:22:25] <Eric Rescorla> yeah, we can do this at the next plenary
[09:22:30] <Eric Rescorla> in 2 years
[09:23:16] <Eric Rescorla> This seems like it's kind of a non-issue
[09:23:29] <Eric Rescorla> I mean isn't it like "you get to two bytes slightly sooner
[09:23:39] <Mike Bishop> You could take the tactic that MAX_STREAMS uses and drop the last two bits, if you really care about the encoding length that much.
[09:23:50] <Martin Thomson> I think that the choice David has is perfectly fine.  Has anyone suggested otherwise?
[09:23:53] <Eric Rescorla> Good idea.
[09:24:01] <Mike Bishop> The real argument right here is whether N:1 or 1:N is potentially useful.
[09:24:05] <Eric Rescorla> Well, it just seems like poorly motivated.
[09:24:20] <Lucas Pardue> the main motivation is not space saving
[09:24:44] kaduk@jabber.org/barnowl joins the room
[09:24:47] <Ted Hardie> I think that's the kind of "benefit" that tends to bite you hard later.
[09:24:51] <Mike Bishop> Or perhaps more relevant, if you think datagrams might be used for something else as well.
[09:25:23] <Martin Thomson> Lucas, that assumes that you are only using the connection for MASQUE
[09:25:43] <Alan Frindell> wouldn't the second one be stream id=4, flow-id=1
[09:25:54] <Martin Thomson> Alan, yes.
[09:26:19] Julian Reschke joins the room
[09:26:23] <Martin Thomson> stream id 8, flow id 2
[09:26:45] <Alan Frindell> then mix in some other, non connect-udp streams
[09:26:58] <Martin Thomson> Right, or other users of datagram
[09:27:02] <Ted Hardie> So the re-use of flow IDs is on different stream IDs, not within a stream?
[09:27:15] <Martin Thomson> The reuse thing is weird.
[09:27:32] <masque> I thought the flow IDs are at the QUIC connection level, not at the stream level?
[09:27:34] <Eric Rescorla> so one nice property of vv's proposal is it prevents you from saying incoherent things
[09:27:37] Eckard Bogenfeld joins the room
[09:27:41] <Eric Rescorla> @masque: who are you?
[09:27:43] Luca Niccolini joins the room
[09:27:48] <Lucas Pardue> even masque itself is confused!
[09:27:58] <Martin Thomson> flow IDs in tommy's proposal are mapped to connection IDs
[09:28:03] <Martin Thomson> From memory.
[09:28:27] <Tommy Pauly> @MT, no, flow IDs are not mapped to connection IDs
[09:28:31] <masque> (I'm Alex Chernyakhovsky, my client appears to have somehow put the room as my name, very confused as to how tthat happened, trying to fix it)
[09:28:36] <Eric Rescorla> thx
[09:29:00] <Tommy Pauly> @MT this allows you to have multiple requests that use the same flow ID
[09:29:17] masque leaves the room
[09:29:19] Markus Amend leaves the room
[09:29:39] <Martin Thomson> Oh, right the whole connection ID addition thing.  I really don't like that part in your draft.
[09:29:54] Eckard Bogenfeld leaves the room
[09:29:55] Luca Niccolini leaves the room
[09:29:58] Luca Niccolini joins the room
[09:30:57] Eckard Bogenfeld joins the room
[09:31:02] Markus Amend joins the room
[09:31:25] Magnus Westerlund leaves the room
[09:31:36] achernya joins the room
[09:31:36] Stefan Santesson leaves the room
[09:32:03] Magnus Westerlund joins the room
[09:32:25] <Eric Rescorla> Well, I think push IDs need their own namespae for practical reasons
[09:32:33] <Alan Frindell> I'm trying to reconcile this with the suggestion from webtrans to either make that more masque-like or use masque directly.  The flow-id strikes me as analogous to Http3Transport session ids?  It's a way of grouping things going to the same place.
[09:32:36] <Martin Thomson> Can a server advertise a limit on the number of flow IDs?
[09:32:55] <Christopher Wood> @Alan the webtrans relationship will be discussed in the next slide, I think.
[09:33:04] <Martin Thomson> I agree with Victor on that last point.
[09:33:15] <Alan Frindell> cool
[09:33:30] <Mike Bishop> I could hear Tommy just fine, FWIW.
[09:33:35] <Mirja Kühlewind> not sure about the many-to-1 mapping but we need to further specify this because deciding
[09:33:59] <Mirja Kühlewind> s/because/before/
[09:34:43] <Eric Rescorla> So it seems like the next set of slides demonstrate that this header is nontrivial complexity
[09:34:54] Jiao Kang leaves the room
[09:35:13] Markus Amend leaves the room
[09:35:13] Eckard Bogenfeld leaves the room
[09:35:16] Markus Amend joins the room
[09:35:16] Eckard Bogenfeld joins the room
[09:35:19] Dirk Kutscher leaves the room
[09:35:27] <Martin Thomson> What Tommy is hitting on is relevant.  if we can imagine a different model for this, then having the indirection is valuable.
[09:35:34] <kaduk@jabber.org/barnowl> I could hear Tommy intermittently
[09:35:41] <kaduk@jabber.org/barnowl> (the first time)
[09:36:16] <Martin Thomson> I don't like the idea that you need a request in order to use datagrams.  Because there are uses that might involve other than 1:1 between streams and datagram IDs.
[09:36:39] <Mike Bishop> Imagine that we rev to send some control frames using datagrams instead of having HOLB on the control stream.
[09:36:41] Eckard Bogenfeld leaves the room
[09:36:56] <Victor Vasiliev> @MT: do you have any specific examples in mind?
[09:37:42] <Jana Iyengar> @MIke Bishop: I like that idea. Of course, you do need reliability there though.
[09:38:06] <Martin Thomson> In this model, you can imagine having adaptive header compression for different datagram uses.  If you were sending the same thing in every DATAGRAM frame, or the same thing in most DATAGRAM frames, then there might be value in saying "this flow ID should have these fixed bytes prepended".  That would be very useful for QUIC tunneling, to erase connection ID overhead.
[09:38:18] <Mike Bishop> Yes, but things like MAX_PUSH_ID where you only need the most recent version, not all of them you've ever sent.
[09:38:18] Jay Iyer joins the room
[09:38:22] <Martin Thomson> You can do that differently of course, it's just an example.
[09:38:33] <Jana Iyengar> @Mike: Yup, agreed
[09:39:32] <Lucas Pardue> I though no one needs MAX_PUSH_ID
[09:39:49] <Martin Thomson> The cynicism is strong today.
[09:39:50] <Martin Thomson> :)
[09:40:00] <Alan Frindell> maybe we can have a dedicate wg to hate on push
[09:40:07] Alessandro Amirante leaves the room
[09:40:11] <Mike Bishop> Don't we already?
[09:40:20] <Martin Thomson> I thought it was all of them.
[09:40:23] Benjamin Kaduk leaves the room
[09:40:24] <Alan Frindell> I'm really unsure which wg I should hate in
[09:40:52] <Jana Iyengar> You can hate on push anywhere. You will only find support. It's one of the few things that happens to be universally hated.
[09:40:59] <Harald Alvestrand> PUSH needs pushing out?
[09:41:06] <Martin Thomson> I don't think that ECN-CE would be signaled that way.  I think that we want some way to send control messages in the request stream.
[09:41:18] <mnot> Yup.
[09:41:19] <Mike Bishop> @Jana, it's not universally hated.  That's why there's discussion.
[09:41:26] Gorry Fairhurst leaves the room
[09:41:42] <Lucas Pardue> control messages in request stream works as long as the payload is framed in DATA
[09:41:49] <Mirja Kühlewind> MT that also a way to support ECN. I think we need more discussion on that.
[09:42:11] <Martin Thomson> Mirja, I agree.
[09:42:26] <Martin Thomson> One thing that concerns me about using DATAGRAM for something like that is signal loss.
[09:42:31] <Mirja Kühlewind> and we need more discussion on ECN now because even if it is an extension (which would be sad) we need to make sure we design the right extension mechanism for it
[09:42:32] <Brian Trammell> yep
[09:42:59] <Martin Duke> We should be ready for L4S, where CE marks may be happening at high volume
[09:43:07] <Eric Rescorla> I'd like to propose that we don't cut David on time. This is one of our two adopted WG items and the one that is clearly the highest priority from the charter. The last two presentations fall into AOB
[09:43:10] <Martin Duke> but signal loss is OK
[09:43:11] <Lucas Pardue> the best extension is an optional extension
[09:43:11] Brendan Moran leaves the room
[09:43:19] <mnot> +1 to EKR
[09:43:25] <Eric Rescorla> And therefore we should use as much time as needed for this topic.
[09:43:28] <Lucas Pardue> +1
[09:43:30] <Eric Kinnear> Ack Ekr
[09:43:35] <Mike Bishop> +1
[09:43:35] <Mirja Kühlewind> my presentation is actually input to the reqs deaft
[09:43:36] <Martin Thomson> Martin Duke: that is a good point.  High rate markings can probably have loss.
[09:43:40] <Mirja Kühlewind> draft i mean
[09:43:47] <Martin Duke> I endorse ekr's idea
[09:44:21] <Martin Thomson> Can't use connection.
[09:44:23] Brendan Moran joins the room
[09:44:31] <Eric Rescorla> I don't see how you get into this hole
[09:44:38] <Tommy Pauly> Indeed Connection is not allowed
[09:44:42] <mnot> Connection doesn't exist in H2/H3
[09:45:03] Benjamin Kaduk joins the room
[09:45:04] <Martin Thomson> This is the wrong hole.
[09:45:06] <mnot> Make it a SETTING
[09:45:10] <mnot> :)
[09:45:15] <Martin Thomson> The real hole is MASQUE over h2, then an h3 backend.
[09:45:17] Brendan Moran leaves the room
[09:45:20] <Martin Duke> any signal that is preserves the binding to an individual packet is already robust because QUIC and TCP have to be robust to packet loss. If we make it an OOB signal, then we have to worry about these problems.
[09:45:22] Brendan Moran joins the room
[09:45:26] <Tommy Pauly> There already is a SETTING for datagram
[09:45:28] <Martin Thomson> mnot is right about the setting.
[09:45:41] <Martin Thomson> Use the setting to indicate that you support connect-udp and we're done.
[09:45:44] <Alan Frindell> How does the proxy know that backend supports Datagram at SETTINGS time
[09:45:55] <Martin Thomson> It won't work on HTTP/1.1, but that's totally fine.
[09:45:58] <Alan Frindell> Doesn't know which backend
[09:46:00] <Lucas Pardue> disagree. A setting implies that the you already know the upstream's capablity
[09:46:00] <mnot> For better or worse, SETTING replaced the function of hop-by-hop headers in H2. Of course, it's not per-message.
[09:46:06] <Martin Thomson> Or just use Connection in HTTP/1.1
[09:46:25] <Lucas Pardue> a masque client might have different proxies for different hosts
[09:46:41] <Harald Alvestrand> "just say no" is a perfectly fine answer
[09:46:42] <mnot> OF COURSE, a h2/h3 extension could re-introduce Connection; it is defined in Semantics, not Messaging, after all.
[09:47:01] <Eric Rescorla> @mnot, let's name it Conexxion
[09:47:07] <mnot> that's already consumed
[09:47:10] <mnot> (seriously)
[09:47:16] <Martin Duke> I love that Bishop's background looks like a Zoom background, but isn't
[09:47:19] <Eric Rescorla> Really? Why?
[09:47:29] <Lucas Pardue> Quonection
[09:47:31] <mnot> some intermediaries re-spell Connection on the wire to preserve message length
[09:47:39] <Eric Rescorla> Wow.
[09:47:45] <Mirja Kühlewind> what?
[09:47:46] <Eric Rescorla> Which spellings are gone?
[09:47:47] <mnot> .. and that's one of the common ones
[09:47:50] <Alan Frindell> By just overwriting a couple bytes in the original string
[09:47:52] <mnot> I have a list somewhere
[09:48:48] <Eric Rescorla> I want to answer MT with "don't do that then" as well :)
[09:49:08] <mnot> Of course, Connection isn't well-implemented: https://cache-tests.fyi/?id=headers-omit-headers-listed-in-Connection#
[09:49:36] Yanmei Liu leaves the room
[09:49:42] Yanmei Liu joins the room
[09:49:46] <Eric Rescorla> Connexxxxxion?
[09:50:43] <Martin Thomson> I would not try this in HTTP/1.1
[09:50:47] Rohit Abhishek joins the room
[09:50:53] Matthew joins the room
[09:50:55] <Martin Thomson> I was just suggesting that it was - in theory - a valid approach.
[09:50:59] Rohit Abhishek leaves the room
[09:51:02] Rohit Abhishek joins the room
[09:51:19] <Martin Thomson> No.  I think we want a new setting for this.
[09:51:53] <Lucas Pardue> if you have a new setting, can we just do extended CONNECT?
[09:52:06] <mnot> The common ones are "nncoection", "cneonction", "yyyyyyyyyy", "xxxxxxxxxx", "x_cnection", "_onnection"
[09:52:41] <Martin Thomson> The first h3 server passes the request along.  The h2 server passes the request along.  The next h3 server understands the requests.
[09:52:53] <Martin Duke> @Mirja are you interested in writing a MASQUE ECN draft? It would be helpful to sketch out a design to make sure we're not blowing it in adopted base design.
[09:52:59] <Martin Thomson> Then you get datagrams flowing from the h3 server to the h2 server, but going nowhere.
[09:53:03] <Martin Duke> ECN would definitely be in charter scope
[09:53:05] <Julian Reschke> mnot: wondering whether we should have a set of reserved field names for that purpose...
[09:53:30] <Martin Thomson> It's a contrived example, the h2 server shouldn't claim support for datagram, but it might also support h3 for others and it might want to support datagrams.
[09:53:39] <Alan Frindell> I don't see how a setting fixes it?
[09:53:40] <Mirja Kühlewind> @Martin let's discuss
[09:53:52] <Martin Duke> OK
[09:54:09] <Martin Duke> That question is actually to everyone, not just Mirja
[09:54:14] <Mike Bishop> The first intermediary supports H3 datagrams for some other use-case, but doesn't implement CONNECT_UDP.
[09:55:08] <Martin Thomson> This is not a method in the sense we understand.
[09:55:11] <Mike Bishop> We already have that paradigm for CONNECT-to-protocol.
[09:55:15] Hannu Flinck joins the room
[09:55:29] <Mike Bishop> Yes, settings in H3 are constant.
[09:55:35] <Lucas Pardue> I actually think we need this setting, not for MASQUE but to avoid squatting on H£_DATAGRAMS
[09:55:37] <Eric Kinnear> +1 Mike, we've got a "supports extended connect" setting already
[09:55:57] <Kazuho Oku> MT's suggestion makes sense; use of UDP is hop-by-hop; SETTINGS is the mechanism that H3 provide for hop-by-hop negotiation
[09:55:58] <Martin Thomson> I floated this a while back, I think.
[09:56:30] <Kazuho Oku> It also follows what we do for websocket for h2
[09:57:05] <Martin Thomson> There are two provisional registrations for udp://
[09:57:07] <Martin Thomson> "BitTorrent tracker protocol based on UDP. "
[09:57:16] <Martin Thomson> "MPEG Transport Stream over UDP "
[09:57:23] <Martin Thomson> Seems awkward.
[09:57:29] <Lucas Pardue> third time lucky
[09:57:51] <kaduk@jabber.org/barnowl> When did those registrations get made?
[09:58:02] <mnot> The protocol being spoken is not UDP; it's a new protocol over UDP
[09:58:05] <Lucas Pardue> how about mt://
[09:58:08] <Ted Hardie> It's provisionally registered in the IANA registry because of that.
[09:58:11] <Lucas Pardue> masque tunnel
[09:58:13] <Eric Rescorla> maybe we can do _onnection://
[09:58:19] <Mike Bishop> Both of those are referring to a UDP endpoint, so it seems consistent.
[09:58:23] <kaduk@jabber.org/barnowl> https://www.iana.org/assignments/uri-schemes/uri-schemes.xhtml just
lists Dave Thaler...
[09:58:39] <Brian Trammell> heh
[09:58:40] <kaduk@jabber.org/barnowl> I am not sure I believe wikipedia for this one...
[09:58:50] <Ted Hardie> Dave did a bunch of the provisionals to help people avoid collisions.
[09:58:59] <mnot> It's interesting that we've got this far without a URI scheme for HTTP-based tunnels / proxy connections.
[09:59:06] <Brian Trammell> "is that hard"
[09:59:13] <Ted Hardie> I can help if you need it.
[09:59:19] <Victor Vasiliev> I'm confused
[09:59:30] <Eric Rescorla> @ted: can we have udp#?
[09:59:32] <Victor Vasiliev> I definitely got provisional IANA registration, and it wasn't hard
[09:59:50] <mnot> EKR: GOAWAY
[09:59:51] <Martin Thomson> This is a good point.
[09:59:51] <Ted Hardie> @Eric What an appealing notion.
[09:59:55] <Martin Thomson> masque://
[10:00:01] <Lucas Pardue> https://tools.ietf.org/html/rfc8589
[10:00:04] Mihail Zverev leaves the room
[10:00:09] <Jana Iyengar> +1 to masque://
[10:00:11] <mnot> masquie:// ++
[10:00:14] <Mirja Kühlewind> or we go without one as connect did so far...
[10:00:19] <Martin Duke> mt = "Masque target"?
[10:00:20] <Brendan Moran> +1 to masque://
[10:00:24] <Eric Rescorla> I would have thought that masque:// referred to the server
[10:00:25] <Brian Trammell> masque:// seems obvious
[10:00:28] <Martin Thomson> masque://tcp:10.0.0.1/ ?
[10:00:31] <Kazuho Oku> so the method would be CONNECT?
[10:00:38] <kaduk@jabber.org/barnowl> # is too passe; gotta go for ?
[10:00:40] <Brian Trammell> does the *server* also need a scheme?
[10:00:51] <Kazuho Oku> GET masque://tcp:10.0.1/?
[10:00:54] <Martin Thomson> masque://tcp@10.0.0.1:443 ?
[10:01:03] <mnot> masque+tcp://
[10:01:18] <Martin Thomson> Not sure how to encode a token alongside the authority.  But I think that it has to be in the authority component.
[10:01:29] <Jonathan Lennox> masque:10.0.0.1:443;transport=tcp
[10:01:30] <Mirja Kühlewind> and back to mt:// (masque+tcp)
[10:02:11] <kaduk@jabber.org/barnowl> SYN+ACK just gets you the initial congestion window, not "huge piles
of data"...
[10:02:12] Chunshan Xiong leaves the room
[10:02:33] <Kazuho Oku> servers willing to do this can use stream-level flow control?
[10:02:54] <Lucas Pardue> no flow control on DATAGRAMS
[10:03:17] <Kazuho Oku> ah right, but then server can drop the frames
[10:03:21] <Lucas Pardue> yes
[10:03:36] <Tommy Pauly> SPIN BIT
[10:03:38] <Kazuho Oku> it's the same as ordinary client sending many UDP datagrams
[10:03:39] <Mirja Kühlewind> there is still cc for datagrams
[10:04:14] <Martin Duke> @kazuho yes but the client can use the MASQUE proxy to hide its identity
[10:04:18] <Mike Bishop> We're already asking a server to send random packet contents to arbitrary endpoints.  There's a degree of trust and authorization in this relationship already.
[10:04:24] Alissa Cooper leaves the room
[10:04:51] <Kazuho Oku> @martin duke my point is that the proxy can discard DATAGRAM frames that it receives
[10:04:52] Benjamin Kaduk leaves the room
[10:05:11] <Martin Duke> i believe that's david's proposal?
[10:05:26] Benjamin Kaduk joins the room
[10:06:06] Benjamin Kaduk leaves the room
[10:06:09] <Kazuho Oku> yeah I think I'm not convinced that we need a recommendation when the masque proxy is essentially a router forwarding packets
[10:06:34] <Lucas Pardue> can we apply the anti-amplification recommendations/logic from QUIC to this case?
[10:07:05] <achernya> Wouldn't that require the udp-endpoint to do the work, not masque-server?
[10:07:16] <Eric Rescorla> @kazuho: I think the difference here is that the masque proxy is coalescine all the clients behind a single IP
[10:07:30] <Jonathan Lennox> Is the assumption that CONNECT-UDP is only for QUIC?
[10:07:34] Julian Reschke leaves the room
[10:07:35] <Lucas Pardue> no
[10:07:38] <Tommy Pauly> No, it's not just for that
[10:07:39] <Eric Rescorla> and so the usual DoS-traffic pushback mechanisms don't work
[10:07:44] Benjamin Kaduk joins the room
[10:07:48] <Kazuho Oku> @EKR I think it's no different than a VPN?
[10:07:58] Nancy Cam-Winget leaves the room
[10:08:02] Nancy Cam-Winget joins the room
[10:08:07] <Kazuho Oku> I do not think VPN restricts the nunmber of packets it forwards until it receives somethnig from a server
[10:08:07] <achernya> VPNs tend to assign 1 IP per client, but could NAT it
[10:08:11] Julian Reschke joins the room
[10:08:12] <Tommy Pauly> Well there are proxy status codes we can send back
[10:08:16] <achernya> Connect-UDP will naturally have a small (possibly singular) IP pool
[10:08:23] <Eric Rescorla> @Kazuho: well, the VPN case you can see RSTs etc.
[10:08:48] <Tommy Pauly> Agreed with Kazuho here
[10:08:51] <Eric Rescorla> And the VPN provider can in principle look for TCP logic violations like ignoring RST or running ahead of the ACK
[10:08:51] <Kazuho Oku> If the server receives RST, I believe the proxy terminates the connection?
[10:09:01] <Tommy Pauly> QUIC goes over VPNs too
[10:09:19] <Eric Rescorla> @Tommy: yes, well this may be an issue there too :)
[10:09:42] <Martin Thomson> circuit breakers work to the extent that you can observe throughput and the protocol in use matches some expectations
[10:09:52] <Eric Rescorla> In any case, isn't the conclusion that the SYN/ACK thing isn't needed.
[10:09:59] <Tommy Pauly> And a lot of other UDP traffic has gone over VPNs for a long time. It's not all just TCP traffic.
[10:10:17] <Martin Thomson> If every UDP protocol had a means to independently ask for permission to communicate before proceeding, that might be good.
[10:10:23] <Jonathan Lennox> You don't want the circuit breakers to ossify congestion control algorithm behavior, though
[10:10:47] Dmitriy Moskvitin joins the room
[10:11:23] <Harald Alvestrand> WebRTC tried to use ICE for that....
[10:11:37] <Kazuho Oku> maybe the desired action for a circuit breaker would be discard packets rather than close the connection
[10:11:57] <Lucas Pardue> to the minute takers, the Jabber logs for this session are available at https://www.ietf.org/jabber/logs/masque/2020-11-20.html
[10:12:05] <Jana Iyengar> +1 Kazuho. That's what the simplest circuit breaker would do.
[10:12:06] <Martin Thomson> Harald, that only works to the extent that the TURN server can see the ICE, but as it doesn't have the password, this is hard for a TURN server to observe
[10:12:40] <Martin Thomson> Kazuho, I can imagine a second level of response for extreme abuse
[10:13:14] Pete Resnick leaves the room
[10:13:20] <Jana Iyengar> ICMP Source Quench!
[10:13:26] <Harald Alvestrand> mt, in the webrtc case we expected the TURN client (webrtc endpoint) to be the one who obeys the refusal to talk. As far as I can tell a TURN server has no protection.
[10:13:29] <Kazuho Oku> Yeah I think it ends up indifferent from a ordinary QUIC client sending too many packets.
[10:13:46] Steffen Klassert joins the room
[10:13:54] <Jonathan Lennox> Not sure what a TURN server can do about a misbehaving client.
[10:13:57] <Jonathan Lennox> Other than the authentication
[10:14:50] <kaduk@jabber.org/barnowl> (obligatory "OAuth is an authorization protocol, not an authentication
protocol")
[10:16:10] Zhuangyan joins the room
[10:16:50] <Sean Turner> it's a framework ;)
[10:16:50] <Magnus Westerlund too> I don't think TURN does anything against missbehaving UDP flows. It is a basic forwarder and behaves like a router. There is one packet in and one out. Thus, it would need an explicit upstream signal that will request a particular stream to be killed or some other protection determination for what is missbehavior. I don't think there is any easy signal that one can determine that.
[10:16:57] <Tommy Pauly> It should be *possible* to cryptographically authenticate a client, but not required
[10:17:14] <David Schinazi> @Tommy that's the intent
[10:17:17] <Tommy Pauly> I think it's as simple as adding "should be able to"
[10:17:21] <Tommy Pauly> Not "should"
[10:17:47] Brendan Moran leaves the room
[10:18:06] Brendan Moran joins the room
[10:18:12] <Eric Rescorla> @Tommy: so you're going to say that TLS client auth fulfills that?
[10:18:26] <Eric Rescorla> If so, that seems reasonable
[10:18:29] <Tommy Pauly> Yes
[10:18:57] <Tommy Pauly> I'm saying that we already have what we need for various use cases with TLS as well as HTTP proxy auth
[10:19:31] englishm joins the room
[10:20:09] englishm leaves the room
[10:20:33] <Magnus Westerlund too> Are it proxying or tunneling?
[10:21:00] <Jonathan Lennox> Does "in their unmodified entirety" mean that the proxy can't have a NAT in it?
[10:21:37] <Tommy Pauly> Yes to being extensible. But the requirements should not say it is *only* entire IP packets
[10:22:22] <Victor Vasiliev> Compression as an extension makes more sense layering-wise
[10:22:45] Gorry Fairhurst joins the room
[10:22:46] <Adam Wiethuechter> +1 on using it as an extension
[10:22:52] <Jonathan Lennox> Is CONNECT-UDP a compression mechanism over CONNECT-IP?
[10:23:05] <David Schinazi> @Jonathan you can think of it as such
[10:23:18] Bhavit Shah joins the room
[10:23:21] <Tommy Pauly> Not exactly—because a compressed CONNECT-IP would allow the client to control IP addresses in a way that CONNECT-UDP wouldn't
[10:23:23] <David Schinazi> But that's not really how it came about
[10:23:27] Rohit Abhishek leaves the room
[10:23:41] <Tommy Pauly> There is no source IP in CONNECT or CONNECT-UDP
[10:24:00] Kazuho Oku leaves the room
[10:24:06] <Victor Vasiliev> Header compression is ~never easier, because the kernel produces a packet with IP headers
[10:24:09] Kazuho Oku joins the room
[10:24:12] <Jana Iyengar> Right, what @Tommy said is what I see as a key difference. And it's not just Source IP, it's all the IP fields, including IPID
[10:24:20] englishm joins the room
[10:25:04] englishm leaves the room
[10:25:22] <Kazuho Oku> use case++
[10:25:59] Mauro Cociglio joins the room
[10:26:13] <Mirja Kühlewind> but IP is hop to hop and not e2e
[10:26:31] Fabio Bulgarella joins the room
[10:26:43] Nicolas Kuhn leaves the room
[10:26:46] <Philipp Tiesel> Evil question: why don't we do PPPoverMASQUE?
[10:27:13] <Martin Thomson> PPPoQ don't you mean?  or PPPoH3
[10:27:31] <Eric Rescorla> I agree with what Tommy just said
[10:27:42] <Zaheduzzaman Sarker> +1 to Tommy
[10:27:48] <Victor Vasiliev> If you run it over RTCDataChannel, you can call it PPPoP2P
[10:27:48] <Mirja Kühlewind> yes that's what I'm aiming for
[10:28:01] <Philipp Tiesel> @martin: right … would be PPPoH3 though
[10:28:05] <Harald Alvestrand> MPLS over MASQUE?
[10:28:21] <Gorry Fairhurst> +1 I think there should be an extensible way to compress in some well-defined way.
[10:28:24] <Lucas Pardue> 3Po3
[10:28:35] <Martin Thomson> P3PoH3
[10:28:40] Qin Wu joins the room
[10:28:51] englishm joins the room
[10:28:59] <Lucas Pardue> using p4 for network slicing
[10:29:15] <Martin Thomson> Can we add a network slicing requirement?
[10:29:16] <Philipp Tiesel> because re discussed many things now PPP already does
[10:29:24] Nitin Batta leaves the room
[10:29:24] Rohit Abhishek joins the room
[10:29:34] <Lucas Pardue> MUST NOT mention netwrok slicing
[10:29:47] <Martin Thomson> How about an extension for newIP?
[10:29:52] englishm leaves the room
[10:29:55] Rohit Abhishek leaves the room
[10:29:55] <Victor Vasiliev> Network slicing?
[10:29:58] Rohit Abhishek joins the room
[10:30:05] <kaduk@jabber.org/barnowl> REALLY SHOULD NOT network slicing
[10:30:11] <Jana Iyengar> Let's please not forget ATMoH3
[10:30:16] <Sean Turner> @VictorV it's 5G :)
[10:30:19] <Lucas Pardue> it's the "cutting edge" Victor
[10:30:28] <Martin Thomson> Jana: 53 byte PMTU
[10:30:32] <Martin Thomson> Or rather 48
[10:30:45] <Zaheduzzaman Sarker> 5G is Old ... we are on to 6G now ;-)
[10:30:45] Rohit Abhishek leaves the room
[10:30:45] Dmitriy Moskvitin leaves the room
[10:30:48] Dmitriy Moskvitin joins the room
[10:30:51] Rohit Abhishek joins the room
[10:31:16] Nitin Batta joins the room
[10:31:30] <Martin Thomson> I really don't get the load balancing requirement now that I read it
[10:31:31] <Victor Vasiliev> Is that a Generation Z name for 802.1Q
[10:31:37] <Martin Thomson> Why would this even USE streams?
[10:31:46] <Jana Iyengar> @mt: No, that's the message size... we are likely to have a number of those DATAGRAM frames within a UDP datagram that might be 1350 bytes... which will eventually go over ATM and be broken into 53 bytes
[10:31:55] <Tommy Pauly> MT agreed that this is confusing about streams...
[10:32:32] <Zaheduzzaman Sarker> I am also not sure about the load balancing requirement
[10:32:59] <Gorry Fairhurst> +1 to Zahed.
[10:33:15] Kazuho Oku leaves the room
[10:33:21] Kazuho Oku joins the room
[10:33:51] <Kazuho Oku> If you can forward arbitrary IP packets, then you are essentially a router. Then, why do we need multiple routers?
[10:33:52] <Mirja Kühlewind> I'm not sure I understand the scenario...
[10:34:09] <Magnus Westerlund too> So I don't see that we can declare a solution for keeping e2e IP flows in order out of scope. If you do multiple QUIC connections to proxy and then split the e2e flows over those connections you are going to have a reordering issue.
[10:34:10] <Mirja Kühlewind> can someone draw an image?
[10:34:11] <Tommy Pauly> Agreed, this sounds like something for the draft
[10:34:14] <Tommy Pauly> Not the requirements
[10:34:30] <Zaheduzzaman Sarker> load balancing where?
[10:34:39] <Mirja Kühlewind> this seems more like a wish list that actual (hard) requirements
[10:35:26] <Martin Thomson> The claim was that there were performance issues if data was confined to a single stream.
[10:35:39] <Martin Thomson> I wonder if it wasn't down to flow control or HOLB issues.
[10:35:51] <Martin Thomson> Streams are not the right vehicle for this anyway.
[10:36:40] <Mirja Kühlewind> i don't like require documents...
[10:36:49] <Mirja Kühlewind> requirement documents
[10:36:50] <Martin Thomson> Mirja: yeah, writing requirements is really hard.
[10:36:51] <Mirja Kühlewind> i mean
[10:37:00] <Jana Iyengar> DGRAMs are an exact match to this. I agree that STREAMs are super weird here
[10:37:14] <Martin Thomson> I spent 8 years doing requirements full time and I still always got the level wrong.
[10:37:45] <Mirja Kühlewind> I think there can be use cases where you want additional reliability by the tunnel connection e.g. on a high loss link
[10:37:56] Nitin Batta leaves the room
[10:38:10] <Martin Thomson> Mirja: that might be out of scope for this group.
[10:38:45] Dan McArdle leaves the room
[10:38:53] <Eric Rescorla> I don't think I understand this "Addressing Architecture" point.
[10:38:59] <Sean Turner> I also like the idea of writing a req draft without 2119/8174 language
[10:39:01] <Eric Rescorla> I thought you needed a mechanism for the server to assign addresses
[10:39:10] <Eric Rescorla> to the client
[10:39:15] <Eric Rescorla> So maybe this is just writing again
[10:39:30] <Martin Thomson> The server gives the client IP addresses and theclient has no say in that?
[10:39:42] <Martin Thomson> That seems like what I would do.
[10:40:27] <Eric Rescorla> Getting out of the queue to see the examples.
[10:40:30] <Martin Thomson> time check: 20min left
[10:40:35] Zhuangyan leaves the room
[10:40:38] <Eric Rescorla> because I think I also misunderstand but it may soon become clear
[10:40:54] <Bron Gondwana> there's nothing afterwards, we could go all night! :p
[10:41:11] <kaduk@jabber.org/barnowl> m e e t e c h o will cut us off at T+5min, no?
[10:41:20] <Eric Rescorla> T-5 one hopes!
[10:41:31] <Lucas Pardue> and gather.town is being turned off at 11:00 UTC
[10:41:38] Jana Iyengar leaves the room
[10:41:54] <Eric Kinnear> My read is that enumerating the conversations we need to solve to gain consensus here is fairly high priority, but feel free to correct that. And of course, let's keep things moving folks...
[10:42:04] <Bron Gondwana> until we go wander prague.oldtown in March
[10:42:05] <Martin Thomson> I'm not sure that you need NAT capabilities here.  As long as you are able to get new IP addresses, or multiple addresses, the linkage problem isn't that bad.
[10:42:24] <Lucas Pardue> lol bron
[10:42:45] <Martin Thomson> If a client wants to be able to get a stable address, that seems like a missing requirement.
[10:42:50] <Eric Rescorla> @MT:  so you need some way in which the client's apparent address is X but it's true address is Y.
[10:42:51] <Robin Marx> Lucas: is that jabber log URL perpetual or should I copy raw text into minutes?
[10:43:01] <Mirja Kühlewind> session is not cut off
[10:43:04] <kaduk@jabber.org/barnowl> Jabber log is perpetual
[10:43:05] <Eric Rescorla> So that either means NAT or telling the client what its IP is so it can stuff it into the header
[10:43:27] <Lucas Pardue> @Robin: perpetual but for convenience I think duplicating the text is preferred
[10:43:28] <Martin Thomson> Ekr: I'm thinking the latter.
[10:43:54] <Eric Rescorla> I don't know how standard VPNs work
[10:43:55] <Martin Thomson> The former is possible via the latter: the server can give out 1918 addresses and perform NAT
[10:44:19] <Lucas Pardue> if you're a machocist then you can splice the jabeer comments into the minutes at the correct time
[10:44:30] <Martin Thomson> It just might perform NAT immediately after the packet leaves the proxy
[10:44:51] <Jonathan Lennox> Where "leaves the proxy" might be entirely conceptual
[10:45:05] <Martin Thomson> right
[10:45:39] Yanmei Liu leaves the room
[10:45:57] <Eric Rescorla> @MT: true
[10:46:12] <Eric Rescorla> @David how does an IPsec VPN work?
[10:46:36] <Martin Thomson> I would hope that this design is informed by VPN design.
[10:47:04] <Kazuho Oku> So this is essentially IPCP of PPP?
[10:47:05] <Martin Thomson> How does this manage consent to communicate?
[10:47:07] Chi-Jiun Su leaves the room
[10:47:08] Yanmei Liu joins the room
[10:47:13] Chi-Jiun Su joins the room
[10:47:15] <Eric Rescorla> right, but I think perhaps this last design
[10:47:35] <Martin Thomson> symmetric reachability = hmm
[10:48:04] <Martin Thomson> does the server then advertise those routes in BGP?
[10:48:27] <Martin Thomson> that would need AS numbers rather than ranges
[10:48:44] <David Schinazi> @MT It could do that, but that's somewhat orthogonal
[10:49:05] <David Schinazi> Routing protocols are out of scope, we're just saying "hey I can reach this FYI"
[10:49:30] <Robin Marx> (I'm having serious trouble keeping up with this, if someone can jump in, would be nice)
[10:49:54] <Martin Thomson> Robin: you can record chicken chicken chicken at this point
[10:50:07] <Valery Smyslov> All this looks like re-inventing IPsec
[10:50:30] <Valery Smyslov> In ipsecme all these issues were discussed long ago...
[10:50:46] <David Schinazi> @Valery it's a different way to proxy IP, but yes it has similarities to IPsec
[10:50:46] <Lucas Pardue> @Bron IETF overflow session link is https://objection.lol/courtroom/9sk8ef
[10:50:56] <Valery Smyslov> Very similar.
[10:51:18] <Valery Smyslov> And I think looking in how TS are negotiared in IKEv2 may be usefull
[10:51:46] <David Schinazi> Yes, the goal was to have something similar to TS because there's no point reinventing the wheel
[10:51:48] <kaduk@jabber.org/barnowl> Might be worth saying that at the mic.
(I am in another meeting room at the moment and can't do it myself)
[10:52:08] <Valery Smyslov> OK
[10:52:45] Maxime Piraux leaves the room
[10:52:45] <Lucas Pardue> @valery if you want me to say something on your behalf please prefix a comment with MIC. OR feel free to jump in queu yourself
[10:52:50] Maxime Piraux joins the room
[10:52:58] <Mirja Kühlewind> I think this has already a very specific solution in mind
[10:53:20] <Martin Thomson> IP address selection might work really well here with HTTP-layer redirects.
[10:53:35] Mark Nottingham leaves the room
[10:53:36] <Martin Thomson> No extension required.
[10:53:41] <Mirja Kühlewind> requirements are probably more specific than needed
[10:54:18] <Mirja Kühlewind> also most of the requirements probably also apply to CONNECT-UDP; so I think only the IP specific ones are really interesting for discussion
[10:55:09] katerega micheal joins the room
[10:55:26] <Kazuho Oku> mirja++
[10:57:25] Alissa Cooper joins the room
[10:57:27] katerega micheal leaves the room
[10:57:53] Alissa Cooper leaves the room
[10:57:58] Alissa Cooper joins the room
[10:58:07] Julien Maisonneuve joins the room
[10:58:25] katerega micheal joins the room
[10:58:27] mnot leaves the room
[10:58:47] Jana Iyengar joins the room
[10:59:20] John Border leaves the room
[10:59:38] Mark McFadden leaves the room
[10:59:59] Brendan Moran leaves the room
[11:01:13] <Lucas Pardue> thanks all, fun session
[11:01:21] <Mirja Kühlewind> thanks all!
[11:01:29] <Jana Iyengar> @Lucas: what on god's earth is that objection site?
[11:01:33] <Brian Trammell> NEXT SLIDE!
[11:01:33] <Mike English> party words
[11:01:34] Valery Smyslov leaves the room
[11:01:36] <Zaheduzzaman Sarker> cheers
[11:01:38] Chi-Jiun Su leaves the room
[11:01:38] <Tommy Pauly> NEXT SLIDE
[11:01:39] Ted Hardie leaves the room
[11:01:40] Christopher Wood leaves the room
[11:01:41] Adam Wiethuechter leaves the room
[11:01:41] Alan Frindell leaves the room
[11:01:42] Sean Turner leaves the room
[11:01:43] Hannu Flinck leaves the room
[11:01:43] Dallas McCall leaves the room
[11:01:43] Harald Alvestrand leaves the room
[11:01:44] Hiroyuki Goto leaves the room
[11:01:44] Dragana Damjanovic leaves the room
[11:01:44] <Martin Thomson> NEXT SLIDE
[11:01:44] Bron Gondwana leaves the room
[11:01:44] Qin Wu leaves the room
[11:01:44] Philipp Tiesel leaves the room
[11:01:45] Victor Vasiliev leaves the room
[11:01:45] Daniel Havey leaves the room
[11:01:45] Brian Trammell leaves the room
[11:01:45] Julian Reschke leaves the room
[11:01:47] Eric Rescorla leaves the room
[11:01:47] Mike Bishop leaves the room
[11:01:48] Igor Gashinsky leaves the room
[11:01:48] Alex Chernyakhovsky leaves the room
[11:01:48] Martin Duke leaves the room
[11:01:48] Yanmei Liu leaves the room
[11:01:49] Bhavit Shah leaves the room
[11:01:50] Jay Iyer leaves the room
[11:01:54] Eric Kinnear leaves the room
[11:01:54] Nancy Cam-Winget leaves the room
[11:01:55] Tommy Pauly leaves the room
[11:01:56] Alessandro Ghedini leaves the room
[11:01:56] Marcus Ihlar leaves the room
[11:01:57] Kazuho Oku leaves the room
[11:02:00] Jana Iyengar leaves the room
[11:02:01] Gorry Fairhurst leaves the room
[11:02:01] Hirochika Asai leaves the room
[11:02:02] Kirsty P leaves the room
[11:02:03] Philip Eardley leaves the room
[11:02:04] Rohit Abhishek leaves the room
[11:02:08] Michael Scharf leaves the room
[11:02:12] <DanielHavey> I'm going to slide off to sleep now. Zzzzz...
[11:02:16] Xavier de Foy leaves the room
[11:02:21] <kaduk@jabber.org/barnowl> ELECTRIC SLIDE?
[11:02:21] katerega micheal leaves the room
[11:02:22] Massimo Nilo leaves the room
[11:02:29] katerega micheal joins the room
[11:02:44] Julien Maisonneuve leaves the room
[11:03:07] katerega micheal leaves the room
[11:03:16] <Jonathan Lennox> Slide trombone
[11:03:16] katerega micheal joins the room
[11:03:22] Jörg Deutschmann leaves the room
[11:03:44] achernya leaves the room: Disconnected: BOSH client silent for over 60 seconds
[11:03:58] Magnus Westerlund leaves the room
[11:04:27] Jonathan Lennox leaves the room
[11:04:40] Dmitriy Moskvitin leaves the room
[11:04:52] Dmitriy Moskvitin joins the room
[11:04:53] Dmitriy Moskvitin leaves the room
[11:04:54] Alissa Cooper leaves the room
[11:05:19] Benjamin Kaduk leaves the room
[11:05:19] Stephan Emile leaves the room
[11:05:19] David Schinazi leaves the room
[11:05:19] Tobia Castaldi leaves the room
[11:05:19] Marten Seemann leaves the room
[11:05:19] Luca Niccolini leaves the room
[11:05:19] Kenneth Murchison leaves the room
[11:05:19] Fabio Bulgarella leaves the room
[11:05:19] Mirja Kühlewind leaves the room
[11:05:19] Maxime Piraux leaves the room
[11:05:19] katerega micheal leaves the room
[11:05:19] Zaheduzzaman Sarker leaves the room
[11:05:19] David Oliver leaves the room
[11:05:20] Markus Amend leaves the room
[11:05:20] Roman Danyliw leaves the room
[11:05:20] Martin Thomson leaves the room
[11:05:20] Mauro Cociglio leaves the room
[11:05:20] Loren McIntyre leaves the room
[11:05:20] Robin Marx leaves the room
[11:05:20] Lucas Pardue leaves the room
[11:05:20] Mike English leaves the room
[11:05:20] Steffen Klassert leaves the room
[11:06:02] DanielHavey leaves the room: Disconnected: BOSH client silent for over 60 seconds
[11:06:07] Meetecho leaves the room
[11:06:32] Magnus Westerlund too leaves the room
[13:25:24] kaduk@jabber.org/barnowl leaves the room
[15:24:56] ihlar leaves the room
[18:29:59] ihlar joins the room
[22:04:30] ihlar leaves the room
[23:03:51] spencerdawkins leaves the room