[clue] Notes from CLUE WG Design Team

[John Leslie <[e-mail]>]

Attending: John Leslie, Allyn, Paul Kyzivat, Mark Duckworth, Marshall Eubanks, Mary Barnes, Rob Hansen, Roni Even, Spencer Dawkins, Stephan Wenger

1005 CTO
Mary: trying to get host role
Mary: email thread on signaling options (trying to share)
1008 clue archives up
Mary: lost my WebEx screen... (archive showing again)
Mary: A1... A2... B... C... (variation on B)
Paul: intent to merely capture what we discussed at meeting; did I succeed? just a starting point for the discussion
Allyn: I thought it was good
Roni: I tried to do a bit of phrasing... we had some drafts in transport from Stephan; we're talking about three ways to convey the full media description; first issue is clue protocol... three options... tied to all of them, what part of the media description is in the SDP; this info relevant to all of them, where you carry the information about the media itself
Rob: inaudible... email on the 27th
Mary: focus of these, SDP as a tool, you did higher level
Paul: I don't disagree with what Roni said, different way of looking, we were talking essentially use-cases, Roni talking signaling, both are useful to talk about; my thought was we're trying to see how it plays out
Roni: on whiteboard... how do you end up setting all the channels eventually, do you offer 1 audio, 1 video, and later negotiate the rest? I tried to say what topics need to be discussed; missing in A) clue as part of the SDP, A1 in Paul's description
Rob: (hard to hear) that was the one we agreed we should avoid if possible
Mary: are we agreed we want to avoid that
Roni: that's OK
Mary: A1 is off the table
Mary: Roni, yours is more complete
Roni: break into topics... analyze what topics being discussed
Paul: we need both to express pieces of mechanism and call flow; offer-answer, symmetric pair... at end this has to impact offer-answer, one, two, or more... if you give up capability exchange you can probably make it work in one; your vanilla SIP device might have trouble with one offer-answer
Roni: may have problems sending grouping
Paul: whether we use bundle or not, one part might be more complex, we don't know yet
Roni: depends how you want to negotiate all the media channels... proposal using RTCP, (looking up draft) draft-lennox-clue-rtpusage
Allyn: which document that Roni wrote is he talking about?
Roni: similar to what bundle is doing... need to try to extend all those options, and code using those options... how to carry clue info, what does initial offer-answer carry, orthogonal; how many messages you need to set up a call, minimum everything in invite, 3264 response; each side has to describe what it can offer; for three modes of negotiation, work out number of messages
Paul: maybe we can figure out the worst case... initial invite, exchange 3 messages from each end, followed by another offer-answer, 12 messages; lots of optimizations possible; is 12 messages too many?
Roni: it's not 12 messages
Paul: possible it could be more... if you use info to do clue messages
Marshall: how long would it take, perhaps 800 msec
Roni: assumption after first exchange you have at least one audio; user experience, once you get an answer you expect to be able to talk (otherwise you hang up), important for the user experience
Marshall: how long it takes is more important than the number of messages
Paul: initial media offer structured to be benign (not what you really want to use), will two ends have something useful to exchange while they work on longer negotiation
Roni: you start out not knowing the other side talks CLUE... an application issue
Marshall: you don't need to receive media to have receiver display "wait for CLUE negotiation"
Paul: are we doing anything here that forces a bad user experience?
Stephan: one key problem, which screen is left vs. right; I don't call it a good user experience if things flicker up on the wrong screen; sufficient to bring up audio; shouldn't put up any video before the CLUE negotiation, it's normal to come up in mono audio
Roni: doesn't cause any problems, receiver will receive only one audio stream, it's normal for the quality of it to change
Paul: is the number of messages going to lead to excluding one or more alternatives? do we have to pick one now?
Stephan: I'd rather one RT more than limit future capabilities; IMHO it would be foolish not to allow capability messages; we're not talking about cell-phone, these things today take 10-20 seconds to come up
Paul: I'm hearing any of these techniques is potentially acceptable
Roni: preference for optimized if everything else equal
Marshall: RTT is useful to optimize, #messages is not
Paul: certain amount of piggybacking... at least two RTTs, may or may not be able to multiplex offer-answers on those, probably talking 3 to 4 RTTs minimum, need to work out the details; you can multiplex CLUE on offer-answer for some but not others; I'm hearing all these options are within an acceptable range
Roni: all of them can work, if all work, I prefer less RTTs; we strike out CLUE running in RTP, we need to describe all of them, how many RTTs, then make a decision; another issue is whether all negotiation should be on the signaling channel
Paul: I don't think we've gotten far enough to decide that -- can you get all your messages in one PDU
Roni: we're talking about end-to-end, not through an intermediary; we have problems opening TCP in those use-cases because of NAT traversal
Rob: RTCWEB considering that
Paul: we may be forced into ?? over UDP... hand-waving over fragmentation
Spencer: that's not the only scenario
Paul: convince yourself messages are small enough
Mary: let's try to close on this -- work out details of the options -- draft deadline for -00 is Monday
Roni: Stephan has transport draft, we'll try to update that to include all three options
Stephan: this draft intended to provide an analysis of available options, reasonable to discard not-usable options and provide details on the useful ones
1052
Marshall: I posted XML for Stephan's transport draft
Mary: we have two sessions; we may not need another call next Tuesday
Marshall: I'm co-chairing a BoF which conflicts with one of them
Mary: think we're done
1055 adjourn

--
John Leslie <[e-mail]>