Re: [Tsv-art] Tsvart telechat review of draft-ietf-rum-rue-09

[Brian Rosen <[e-mail]>]

Thank you for this review.  It was fortunate for us that you got selected to review it, because your background in WebRTC will let us strengthen the document.

Before getting into the specifics, I wanted to explain the model better:
There is a user who is sitting in front of some client
There is a provider who has implemented the RUE server interface.

We want to allow the user to be able to use a WebRTC client.  When they do, there us a third piece: a gateway that has a WebRTC server on one side and a RUE client interface on the other.  Unlike a real WebRTC situation, the other end isn’t necessarily a WebRTC client, and in particular, may be just a SIP client.

A provider is free to implement a direct WebRTC interface to their service.  That would not conform to the RUE interface, and our document is not intended cover that case.



> On Nov 29, 2021, at 11:08 PM, Bernard Aboba via Datatracker <[e-mail]> wrote:
> 
> Reviewer: Bernard Aboba
> Review result: Ready with Issues
> 
> This document has been reviewed as part of the transport area review team's
> ongoing effort to review key IETF documents. These comments were written
> primarily for the transport area directors, but are copied to the document's
> authors and WG to allow them to address any issues raised and also to the IETF
> discussion list for information.
> 
> When done at the time of IETF Last Call, the authors should consider this
> review as part of the last-call comments they receive. Please always CC
> [e-mail] if you reply to or forward this review.
> 
> Reviewer: Bernard Aboba
> Document: draft-ietf-rum-rue-09
> Result: Ready with Issues
> 
> Overall, found several issues with this document:
> 
> 1. The specification appears to contradict some of its normative references
> with respect to support of IPv6.
See below
> 
> 2. Although the document does include language within the provider
> configuration, it does not mention language negotiation [RFC 8373], which
> included relay services as a use case.  So I'm curious as to how users with
> different communications requirements might be accommodated. For example, a
> user who is speech impaired but not hearing impaired, and so would like to
> communicate via ASL to the interpreter (or realtime text in English if video
> fails) but can receive spoken English.
As a practical matter, providers have a queue per language they support.  The URI the RUE device uses would be specific to the language the user prefers.  The Dial Around service is also provisioned with multiple language URIs.  

> 
> 3. Maintainability. This specification does not appear implementable using the
> WebRTC 1.0 API, and as a result, it cannot run in a browser.   Major changes
> would be required to WebRTC code bases such as libwebrtc and pion to allow for
> native implementation.  The separate fork that would be needed would become
> increasingly difficult to maintain over time.
See above on model and below for details.

> 
> ---------
> 1. IPv6 support
> 
> Section 4 says:
> 
> "   Implementations MUST support IPv4 and IPv6.  Dual stack support is
>   NOT required and provider implementations MAY support separate
>   interfaces for IPv4 and IPv6 by having more than one server in the
>   appropriate SRV record where there is either an A or AAAA record in
>   each server DNS record but not both.  The same version of IP MUST be
>   used for both signaling and media of a call unless ICE ([RFC8445]) is
>   used, in which case candidates may explicitly offer IPv4, IPv6 or
>   both for any media stream."
> 
> [BA] This document requires that RFC 8835 be supported, but this paragraph
> conflicts with that document in multiple ways. RFC 8835 indicates ICE as both
> required to implement and to use.  When it says that "Dual stack support is NOT
> required" is it referring to the client or the provider? RFC 8835 seems to
> conflict in its IPv6 requirements as well.

We want to allow ICE to be implemented, but not necessarily used.  Would it be acceptable to note that in the exceptions for 8835 conformance?

I don’t see anything in 8835 that requires dual stack.

Dual Stack is not required at either the client or the server.  What other changes do you think we need to harmonize with 8835?


> 
> ---------
> 3. WebRTC usage
> 
> Section 5.5 says:
> 
>   Implementations MUST conform to [RFC8835] except for its guidance on
>   the WebRTC data channel, which this specification does not use.  See
>   Section 6.2 for how RUE supports real-time text without the data
>   channel.
> 
> [BA]  The WebRTC 1.0 API does not allow data to be sent except via the data
> channel API.  So when it says that the WebRTC data channel is not used, by what
> mechanism is real-time text to be sent?  This is not audio/video data so that
> it cannot be sent using the RTCRtpSender API.   The only way I can think of
> that this might be implemented is to add back support for the RTP data channel
> so that real-time text could be implemented on top of the RTCDataChannel send()
> method. However that in turn would violate aspects of the WebRTC security
> model, such as the deprecation of SDES.
As above, there would need to be a gateway between a WebRTC client (which would be using the data channel to send RTT) and the RUE client interface.  That gateway would need to unwrap the RTT from the data channel, wrap it in RTP per RFC4103 and vice versa.  We recognize the ugliness of this solution.  I was not involved in the development of WebRTC and I don’t understand why it does why it uses the data channel, but I have been dealing with the consequences of that in several venues, and it’s a big problem.    There is increasing deployment and use of RTT and it all uses RFC4013.  But it is what it is, and the gateway will have to interwork.  It has been one of the largest challenges in the experimental implementation that is underway for RUE.  Is there some text change we need?


> 
> If that is the intent here (it's the only practical way that RTT can be
> supported other than via the gateway model), then I would prefer that it be
> stated explicitly.
> 
> Other places in the document conflict with the RFC 8835 requirement.
> 
> For example, RFC 8835 Section 3.4 says:
> 
>   The primary mechanism for dealing with middleboxes is ICE, which is
>   an appropriate way to deal with NAT boxes and firewalls that accept
>   traffic from the inside, but only from the outside if it is in
>   response to inside traffic (simple stateful firewalls).
> 
>   ICE [RFC8445] MUST be supported.  The implementation MUST be a full
>   ICE implementation, not ICE-Lite.  A full ICE implementation allows
>   interworking with both ICE and ICE-Lite implementations when they are
>   deployed appropriately.
If we left this alone, it would mean that ICE-lite couldn’t be used with a non WebRTC client. We probably have to allow a RUE client to use ICE-light, but the server can’t assume it.  I will address that.  Both sides have to support full ICE, but we don’t require that is be used.  Is there another problem I’m not seeing?

> 
> RFC 8835 also has requirements for support of IPv4 and IPv6 that seem to be in
> conflict with the statements in Section 4.   For example, RFC 8835 Section 3.2
> says:
> 
> 3.2.  Ability to Use IPv4 and IPv6
> 
>   Web applications running in a WebRTC browser MUST be able to utilize
>   both IPv4 and IPv6 where available -- that is, when two peers have
>   only IPv4 connectivity to each other, or they have only IPv6
>   connectivity to each other, applications running in the WebRTC
>   browser MUST be able to communicate.
> 
>   When TURN is used, and the TURN server has IPv4 or IPv6 connectivity
>   to the peer or the peer's TURN server, candidates of the appropriate
>   types MUST be supported.  The "Happy Eyeballs" specification for ICE
>   [RFC8421] SHOULD be supported.
None of that seems to require dual stack.
> 
> Section 6 of the document says:
> 
>   This specification adopts the media specifications for WebRTC
>   ([RFC8825]).  Where WebRTC defines how interactive media
>   communications may be established using a browser as a client, this
>   specification assumes a normal SIP call.  The RTP, RTCP, SDP and
>   specific media requirements specified for WebRTC are adopted for this
>   document.  The RUE is a WebRTC "non-browser" endpoint, except as
>   noted expressly below.
> 
> [BA] Is RUE is really a WebRTC "non-browser" endpoint?  If the goal is to allow
> RUE to be easily built on top of native WebRTC libraries such as libwebrtc or
> pion, then it should inherit WebRTC requirements such as ICE, dual stack
> support, etc.
There really isn’t an exact fit with non-browser endpoint, but it was the closest we could get.
So we need to highlight exceptions.   The server should be able to handle a client that offers full ICE.  
> 
> Section 6.1 says:
> 
>   Implementations MUST support [RFC8834] except that MediaStreamTracks
>   are not used.  Implementations MUST conform to Section 6.4 of
>   [RFC8827].
> 
> [BA] Since MediaStreamTracks are how audio/video is obtained from devices (or
> rendered), I don't understand how a WebRTC application (browser or non-browser)
> can function without them.  Elsewhere, the specification states that the data
> channel isn't used, now it seems to say that audio/video isn't used either.
I may have been too broad in what we were attempting to do.  We definitely are not using JSEP or Media Capture and Streams API.  So there isn’t any real manifestation of a MediaStreamTrack.  There is just the RTP and associated SDP.  Could you suggest how we might describe that better?


> 
> RFC 8827 is the security architecture of WebRTC.  Is it really true that RUE
> implementations only need to support Section 6.4??  If so, this allows major
> deviations from WebRTC to the point where interoperability could be severely
> impaired.
There clearly is an error here.  I tried to track back to see where/how I created it.  The reference should be to Section 5, SDP Identity.  We can’t use most of the rest of 8827 because it’s about WebRTC signaling and consequences of that.  All we can pull in is the Identity SDP exchange.  We have a lot of the meat of what is in 8827 in our text (we require DTLS-SRTP, we require TLS on SIP, ICE, etc).

> 
> 6.2.  Text-Based Communication
> 
>   Implementations MUST support real-time text ([RFC4102] and [RFC4103])
>   via T.140 media.  One original and two redundant generations MUST be
>   transmitted and supported, with a 300 ms transmission interval.
>   Implementations MUST support [RFC9071] especially for emergency
>   calls.  Note that RFC4103 is not how real-time text is transmitted in
>   WebRTC and some form of transcoder would be required to interwork
>   real-time text in the data channel of WebRTC to RFC4103 real-time
>   text.
> 
>   Transport of T.140 real-time text in WebRTC is specified in
>   [RFC8865], using the WebRTC data chanel.  RFC 8865 also has some
>   advice on how gateways between RFC 4103 and RFC 8865 should operate.
>   It is RECOMMENDED that RFC 8865 including multiparty support is used
>   for communication with browser-based WebRTC implementations.
>   Implementations MUST support [RFC9071].
> 
> [BA] The reason why RFC 8865 was developed was that there was no practical way
> to support RFC 4103 except via the RTCDataChannel API. Requiring RFC 4103 to be
> supported doesn't make the practical problems go away - you need a way to
> provide the RTT data to be sent.
As I said, I wasn’t there, and I don’t know what the practical problem of sending T.140 in RTP were discussed.  I know we’re implemented it in mobile networks, in emergency service networks and in Video Relay services.  It’s required (regulatory) in some jurisdictions.  We can’t make that go away.   It’s deployed all over and is being used to replace TTY.

> 
> When you say that implementations MUST support RFC 9071, how is this supposed
> to work in a WebRTC implementation?
Gateway does it.

> 
> 6.3.  Video
> 
>   Implementations MUST conform to [RFC7742] with following exceptions:
>   only H.264, as specified in [RFC7742], is Mandatory to Implement, and
>   VP8 support is OPTIONAL at both the device and providers.  This is
>   because backwards compatibility is desirable, and older devices do
>   not support VP8.
> 
> [BA]  The reality is that H.264 is not very widely used in WebRTC applications
> (less than 1 percent of calls use it), and as a result, implementations are
> quite buggy.  I do not believe that it is serving RUE users well to make VP8
> optional.   Even if you're going to stick with H.264, you might consider going
> beyond only requiring support only for constrained baseline profile.  Some
> implementations now support constrained high profile, for example.

Every document has some compromises, and this is one of the ones we made. 

> 
> Section 6.8
> 
>   For backwards compatibility with calling devices that do not support
>   the foregoing methods, implementations MUST implement SIP INFO
>   messages to send and receive XML encoded Picture Fast Update messages
>   according to [RFC5168].
> 
> [BA] Really? Earlier in the document is says that backwards compatibility is
> not a strict requirement.  Any widely used WebRTC code base will support NACK,
> FIR and PLI.  seems
> 
Another compromise.  It’s pretty ugly to interwork these.  Possible, but ugly.