- From: Harald Alvestrand <harald@alvestrand.no>
- Date: Tue, 14 Jun 2011 13:21:31 +0200
- To: "public-webrtc@w3.org" <public-webrtc@w3.org>
[Contributor hat on] I thought it might be useful to share some notes I made after last week's RTCWEB telephone call / interim meeting. This is NOT any kind of official minutes, might blatantly misrepresent some viewpoints, and may miss any number of things. Read for entertainment! ----------------------- The below is produced by copying the agenda and making comments on each point. ----------------------------------------------------------------------------------- Architecture: 30 minutes draft-alvestrand-rtcweb-overview (Harald covering all drafts - 15 min) draft-cbran-rtcweb-protocols draft-holmberg-rtcweb-ucreqs draft-rosenberg-rtcweb-framework discussion 15 min focus on identifying what are the big architecture decisions that are needed more than actually debating what they choices we should make. It was almost too successful in that - in that there were nearly no questions/controversies, and no really important ones. Use Cases & Solution Requirements: 45 minutes draft-holmberg-rtcweb-ucreqs (Christer - 15 min) discussion (30 min) Holmberg went through the requirement scenarios one by one, and the chairs said "if you don't want this to be a scenario we should support, shout". Important discussions: - The "multiparty video communication" scenario (no centralized server) gathered commentary. People pointed out, among other things, that this required mixing of audio streams in the client, and reasonably advanced control of audio (since the scenario suggested placing people's audio streams in stereo space by panning). - The "multiparty game" scenario (also without central server) also requires mixing of audio streams with locally generated audio. Even more control. - The "telephony terminal" scenario is a driver for supporting DTMF tone signalling as a separate payload type. Given that there are ~3 different ways to do DTMF signalling over SIP and RTP, this could turn expensive. No scenario was rejected outright. I think we can assume that they're all in the set we need to design for support of. One interesting input was Keith (Keith Drage?) who wondered if WebRTC was going to throw away the IETF's years of work on supporting centralized conferences without even thinking about it (XCON and friends). I think it would be good if the WG concludes that we're not intending to preclude their use in any way. It's unlikely that they will be reasonable to use in all scenarios, of course. Security: 30 minutes draft-rescorla-rtcweb-security (Rescorla 15 min ) discussion 15 min This generated a *lot* of heat. Once we got down to what the heat was actually about, it seemed that: - Asymmetric exchanges, of whatever type, protect the keys against the server. This is a Good Thing, except where wiretap capability is required - and, as EKR pointed out, the server has a *lot* of ways to mount a MITM attack even with asymmetric key exchange if it knows it needs to wiretap when the call starts. - SDES (passing keys via SIP/SDP) compatibility was required for backwards compatibility/interoperation. So at least in some cases, the server will know the keys, and the keys have to be API-accessible in this case. Two mechanisms requires more support than one mechanism, so one needs to have a good use case for requiring both. - EKR was dismissive of the idea that people would use the verification mechanisms in ZRTP and friends in a secure way. Others (in particular Alan J, I think) argued that people who had reason to be paranoid would use them when available, and that this was "good enough security". ZRTP draft-johnston-rtcweb-media-privacy (Alan J - 5 min) - focus on topic of are we going to have the "ZRTP or not" debate 5 min discussion Just a continuation of the previous discussion. NAT Traversal: 30 minutes draft-kaufman-rtcweb-traversal (Kaufman 10 min) 20 min discussion This turned out to be on the idea that you could implement ICE with only a barebones STUN component in the browser itself, and leaving the rest to Javascript. The author was severely challenged on implementability, due to the need to do fairly exact timing in ICE, and was basically told to go away and come back with a demonstration that it could be done. (Discussion continued on the list) RTP Usage: 30 Minutes draft-perkins-rtcweb-rtp-usage (Colin - 15 min ) 15 min discussion Lots of heat here. Magnus, as part of the old school of RTP, was adamant that putting audio and video (and possibly other stuff) in the same RTP session was a Bad Idea. Several people claim that the cost of opening multiple UDP ports is significant, and wanted to avoid it. Discussion ensued, but little agreement; it seems, however, that it's clearly OK to use one RTP session for multiple video streams. Non audio/video data: 25 min frame issues (chairs 5 min ) - draft-holmberg-rtcweb-ucreqs - draft-cbran-rtcweb-protocols discussion 20 min * make sure we agree on requirements * are focused on unreliable data ? We'd run out of time at this point. The UDP/DCCP/DTLS stack seems to be the answer people feel is reasonable for the requirements. "We can build a reliable channel on top of an unreliable one if we need to" seemed to be a common mantra. Call ended after 4 hours and an odd number of minutes.
Received on Tuesday, 14 June 2011 11:22:00 UTC