| Network Working Group | R. Raymond |
| Internet-Draft | E. Lagerway |
| Intended status: Informational | Hookflash |
| Expires: December 28, 2013 | I. Baz Castillo |
| Versatica | |
| R. Shpount | |
| TurboBridge | |
| June 26, 2013 |
WebRTC JavaScript Object API Rationale
draft-raymond-rtcweb-webrtc-js-obj-api-rationale-00
This document describes the reasons why a JavaScript Object API approach is a far better solution than using SDP [RFC4566] as a surface API for interfacing with WebRTC. The document outlines the issues and pitfalls as well as use cases that are difficult (or impossible) with SDP with offer / answer [RFC3264], and explains the benefits and goals of an alternative JavaScript Object API approach.
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."
This Internet-Draft will expire on December 28, 2013.
Copyright (c) 2013 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
While the IETF RTCWEB WG is not specifically tasked with providing an API by the W3C, the group has effectively defined a surface API with the mandate to use SDP [RFC4566] with offer / answer [RFC3264].
SDP is a condensed text based format that typically describes real-time media streams, networking properties, codecs, media state and media attributes. SDP is completely extensible and can be used to describe absolutely anything so long as it is formatted correctly within the few limited constraints.
The points for mandating SDP with an offer / answer API typically boil down to:
This document will explain why these reasons are insufficient to continue with a SDP with offer / answer mandate approach given strong logical arguments and reasons with real world scenarios where this approach fails and due in no small part to its lasting consequences (including negative consequences for SIP).
The document highlights the benefits and goals for a different “JavaScript Object API” approach, which satisfies the RTCWEB charter’s requirements, yields greater compatibility and offer a road-map where future potential extensions can be readily added without breaking existing implementations.
A “JavaScript shim” is described including details on how it can offer a wrapped API around a core WebRTC JavaScript Object API. This Shim will provide the same level of “ease of use” as experienced with the current SDP WebRTC API. However, this Shim is not mandatory to use for those who do not require an “SDP with offer / answer” model.
The issue with SDP is not the expressiveness of the format but its usage as an arbitrary universal format and an API surface instead of providing JavaScript developers an object model they can readily understand. JavaScript could be used to control the plumbing of media objects using familiar JavaScript expressive concepts enshrined with methods, properties and events. Today, in many real-world use cases, controlling WebRTC requires modifying SDP directly.
Requiring JavaScript developers serialize their API control requests into a text format (via modifications of SDP existing blobs) is only one aspect of the many issues the SDP approach creates for developers. Needlessly, an offer / answer state machine is imposed on JavaScript developers as well.
While the currently mandated SDP based API allows developers to quickly implement basic calling demos and interoperability with some SIP networks, it has many issues that will be explored and explained in this document and include (but not limited to):
An alternative to all the issues caused by a universal format is described later in the document. This alternative allows JavaScript to control the behavior of the media engine’s plumbing while providing extensible and modifiable shims written entirely in JavaScript that produce consistent signaling and exchange formats specific to the network where those formats operate.
While the primary goal of WebRTC is to enable browser to browser communication, the definition of a “browser” is ever expanding. Beyond just traditional hand-held applications, hybrid applications that are part HTML-5 and part native code exist. Servers will become as much as part of the WebRTC infrastructure as browsers. Minimizing the requirements to the basic wire compatibility necessary to achieve RTC is essential for maximum compatibility, flexibility and varying usage scenarios.
The mandate for the RTCWEB charter is to simply provide basic “on-the-wire” compatibility and for any basic security requirements (such as enforcing ICE connection agreements). The RTCWEB charter goals have been exceeded by going well beyond that scope by mandating an API that works fine for simple SIP interoperability demos but does not provide easy compatibility to the basic constructs needed as outlined from the charter for use with other on-the-wire signaling protocols (other than SIP). If SIP is the only end goal of the WG, then that goal must be specifically stated rather than effectively mandated by making alternative signaling approaches unreasonably difficult to achieve.
The current SDP approach requires an offer / answer state machine. Mandating an offer / answer state machine implies that:
This offer / answer model is defined as required with the current implementation. Not only do the browser vendors have to enforce the rules, all JavaScript authors must also adhere to these rules of signaling. While WebRTC does not dictate the signaling mechanism between browsers, effectively it is imposing this signaling state machine on all implementations (which need not be mandated as part of the RTCWEB Working Group).
There are other models for signaling other than offer / answer. For example, one-sided constraints based negotiation is an alternative model. This type of negotiation requires each side to determine what it wants to receive independent of the other. This signaling is akin to saying “if you plan to send anything, make sure it conforms to the following”. Changes to the media may occur without agreement from the remote party where each side decides what is acceptable to receive without agreement from the other. The remote side can decide if it wants to send within those constraints or not. There is no round trip offer / answer required in this model.
Offer / answer introduces the unnecessary asynchronism to the API and JavaScript implementations. For example, changing the list of codecs expecting to receive or the current sending codec can be done immediately without the need for asynchronous calls.
Offer / answer is not required to achieve RTC wire compatibility but it is currently mandated when alternatives could exist.
The offer / answer SDP state machine is already violated in WebRTC. Trickle ICE precludes offer / answer round trips and other proposed standards like NoPlan [I-D.ivov-rtcweb-noplan] suggest relaxing the offer / answer model even more. The rules of what offer / answer at this point is undefined and in clear violation of the strict previous rules without clear direction on what exactly constitutes offer / answer anymore and where it should and should not be used.
A new state for offer / answer called PRANSWER is now defined, which did not exist as part of the standard offer / answer state machine. Offer rollback is not adequately defined either should an offer / answer conflict occur.
Currently, switching codecs requires an SDP offer / answer round trip even though it is not technically need for an RTC engine to change codecs. Should this be another exception to the offer / answer state machine?
SDP is a flexible format many alternative methods to express the same meanings. The smallest change can alter the SDP’s meaning.
This creates a parsing and SDP generation compatibility issues. If the SDP is packaged by JavaScript and delivered to the remote browser then each browser must support every single possible variation of SDP for every browser version and platform in existence. They must do this without failure. They also do not know exactly the format that is expected by the remote party in advance in order to generate compatible SDP (despite not having sufficient knowledge about the remote party to provide the correct SDP).
Since WebRTC does not mandate the format on the wire for signaling, one supported use case for WebRTC must be allowing the browser generated SDP to be converted into alternative on-the-wire formats. This SDP conversion may be performed by JavaScript in the browser or by an intermediate gateway. In either case, the converter must be entirely aware of all flavors and variations to the SDP possible from every browser platform and version, despite browser version detection being heavily frowned upon by industry best practices. Likewise, the JavaScript or gateway must know how to generate the correct SDP for all browsers and versions before passing the serialize SDP blob into the browser. Generating compatible SDP may be impossible unless the exact formats and restrictions are unquestionably clear by all implementers of the specification (which is anything but clearly described in the current WebRTC SDP based API that developers are mandated to use).
The current SDP based API is limited to placing a call and answering a call and adding media. To perform common edge cases or to utilize RTC features beyond the basic API requires SDP mangling.
Many of the operations from JavaScript to control or fetch properties from RTC will be through serialization to / from the SDP instead of a developer using familiar JavaScript language constructs (e.g. object methods, properties and events). The JavaScript developer must learn an entirely new protocol called “SDP” and be able to parse and generate not only basic SDP but any SDP extensions without introducing a single compatibility issue.
Examples; A JavaScript developer wants to hold / un-hold media streams. The developer must use a widely adopted but hidden feature to parse the SDP from the browser, change it to add the appropriate “hold” state, send that hold state to the remote side, wait for the “answer” to accept the hold, parse the result on the return to see if the hold was accepted and feed the result to the browser.
Worse, a flood of extensions to SDP for WebRTC are being written to “enhance” and “extend” the functionality of the browser with new features. Many basic things are not defined in the current SDP based API, for example, changing non-negotiated codec parameters, such as codec bandwidth.
There is no facility for JavaScript to detect what SDP the browser is currently using or capable of delivering. The developer has no idea of the extensions available, or what SDP will be produced, or what SDP is compatible. The developer’s JavaScript code must be able to handle everything generated by the browser for any use case beyond basic call, answer and hang-up. This is a heavy burden to place on a JavaScript developer who is not familiar with the details of RTC concepts as expressed in SDP, and is a challenge even for those who are familiar.
Effective APIs are meant to be contracts between a producer and consumer, this SDP methodology offer little in the form of any such contract.
If SDP is to become standardized for use with WebRTC then JavaScript developers must learn SDP to use RTC’s available features and build new features.
The choice must be made that SDP may be modified or not. If modifications are the only way to achieve RTC features available then what is allowed to be modified must be clearly defined in exact detail and the expected behavior of each feature (and modification of each feature), as expressed in SDP, must be defined. Anything short of of exact specifications will cause incompatibility. Again, the implication is that Web / JavaScript developers must learn SDP to utilize the available RTC features and they must learn the rules of modification equally well, which do not exist at all today.
If the choice is to not allow complete SDP modification at all, then the protocol becomes extremely tied to SDP based protocols like SIP. Yet, there is no mandate for SIP to be the standardized protocol in WebRTC, which presents the argument that SDP manipulation must be allowed.
The SDP mangling issue isn’t just an issue when the format is sent on-the-wire. If Browser A sends Browser B an SDP, the current philosophy is that the SDP is allowed to be modified. However, there is the possibility of modifying the SDP generated by Browser A and giving that modified SDP back to Browser A to change it’s options before it even creates the offer for Browser B (and likewise when Browser B responds with its SDP answer).
How much of the SDP is allowed to be modified before giving the SDP back to the local browser? SDP is a free-form format so anything can theoretically get changed, but should it be allowed? If not, what can and cannot be modified? CODECS? SSRC? SDES? Fingerprints? Transports? And so on…
This issue only gets compounded when extensions are factored in as well.
With the SDP baked into the browser binary, the only way SDP compatibility issues can be fixed is by releasing a new browser update and the JavaScript developers must support or work around flaws until the browser vendors deliver the fix and the user base upgrades their browsers.
While it could be argued that any bug must be worked around, SDP is a unique problem. SDP is a free-form format. Being compatible isn’t as easy as implemented a limited wire protocol for media transport or a API contract with defined features and attributes. The likelihood of free-form SDP containing errors is far greater than a typical well defined API due to SDPs many flavors, interpretations and lack of strong definition.
To illustrate a scenario: 1. Browser Vendor A has a bug 2. Browser Vendor B can’t work with A because of the bug so it implements a “work around” 3. Browser Vendor A fixes the bug but implements a work around to be compatible with Browser Vendor B’s “work around”
This situation is how browser bugs can become enshrined as there’s no way to update the SDP produces by the browser binary once it’s released until the next update release cycle occurs. This would not be true if JavaScript was used via a shim to produce SDP as JavaScript can be dynamically updated as needed at any time and a service provider can choose to update their JavaScript implementation to exacting expectations for their network regardless of the browser version.
The lower level RTC wire protocols that need to be mandated by the RTCWEB Working Group have limited scopes and well defined behaviors. Any mistakes are obvious, likely to present very rapidly, and easy to spot which party is doing something wrong and much easier to fix as a result. This is not true with a free form highly descriptive language for sessions. The combinations are limitless and every scenario is difficult to test, especially in concert with every other browser vendor with every version released. The session description will be the likely place of failure across the browsers when the session description is generated inside the browser’s binary.
One of the main arguments for using SDP with offer / answer was supposed to be easy of compatibility with existing signaling networks, like SIP. Instead, variations in the browser’s SDP will likely worsen SIP compatibility instead of enhance it.
A SIP provider must now be compatible with every browser’s SDP on every platform and version and the browser’s SDP must be compatible with every SDP from a SIP network. Alternatively, JavaScript or SBCs (Session Border Controller) must be used to re-write any incompatible SDP to be compatible. But this moves the problem from the browser to JavaScript or requires SBCs to “fix” the problem.
Had SDP been entirely generated by JavaScript rather than come from the browser engine, the JavaScript could create only SDPs compatible with a particular SIP provider under control of their own JavaScript and the SIP provider could chose which JavaScript SDP parsing / generation code to run, for maximum compatibility.
By mandating SDP, the requirement for compatibility with WebRTC are increased substantially with little benefit. Instead of just supporting basic media RTP [RFC3550], STUN/ICE/TURN [RFC5389]/[RFC5245]/[RFC5766], DTLS [RFC6347] and CODECS an additional bar must be passed, i.e. a browser or other WebRTC compliant API must support SDP with a full offer / answer state machine (or a state machine with additional rules to make it flexible for various scenarios).
With an alternative approach, this entire requirement for SDP could be removed without any loss of compatibility or increase in complexity while achieving greater compatibility via the JavaScript shim.
The current mandated SDP based API cannot be implemented as a standard by independent browser vendors in its current form. A list of subsequent behaviors regarding the usage, parsing, handling, extensions, behaviors, constraints and other such reference documents must be meticulously defined for SDP with the modified offer / answer state machine or no browser can ever claim to be “compliant”. The current definition process is far from complete.
The current WebRTC SDP based API is far from achieving that goal due to the inclusion of free-form SDP with offer / answer and it is grounds for removing it as it goes beyond the RTCWEB’s charter and limited scope.
Any incremental approach that does not remove the offer / answer model requirement yields a road block to achieving alternative WebRTC signaling protocols other than SIP.
An alternative JavaScript approach that does not require sessions to be described on the wire embedded inside the browser is being proposed as an alternative solution so the RTCWEB charter can complete its defined goals in a timely fashion.
There are many oddities in the SDP RFC [RFC4566] and the various related extensions.
For example; will rtp maps be required or not? They are not required for basic CODECs according to the RFC. However, with all the flavors of CODECs being offered, defining a mapping between payload an interpretation is critical to compatibility and not just a good idea.
Another example; should “t=0 0” be respected? Is that allowed to be changed? Do the browser vendors need to enforce the attribute, or should the JavaScript layer enforce it? Should the streams wait to start until the NTP time stamp and close when the NTP time completes?
These questions must all be completely addressed in detail. This could also cause a cascade of updated references and confusion as to which version is to be adhered and what each browser specifically support. Nominally referencing the SDP RFC will not be sufficient, and deltas from the established standards will need to be defined when the rules change.
At the time of authoring this document, three plans on how to handle large number of media streams in SDP have emerged currently under consideration from the IETF, referred to as PlanA [I-D.roach-rtcweb-plan-a], PlanB [I-D.uberti-rtcweb-plan] and NoPlan [I-D.ivov-rtcweb-noplan].
All three of these plans acknowledge that using SDP as it is historically defined in SIP is inefficient and problematic for large number of media streams, especially factoring in that each media line must have its own unique ports.
All three of these plans are a perfect example of why not to use SDP as the basis for WebRTC. SDP has some arbitrary limitations as a description protocol for multiple streams where as no such limitations exist at the lower layer transports themselves. RTP allows for multiplexing multiple SSRCs. In other words, the problem is SDP, not the real time transportation technologies.
These drafts illustrate the limitations of SDP and attempt to solve it by introducing either more complex descriptions around SDP and / or by “relaxation” of the offer answer model combined with altering the description language of SDP.
None of these drafts address most of the concerns outlined in this draft. If anything they further illustrate how divergent the SDP will become as more and more effort is put into working around problems inherent to the nature of utilizing SDP (or any universal format).
The issue that SDP implementers face should be isolated to those who require SDP for their signaling protocols (namely SIP) where they can choose the best practices for their networks for interoperability. These complex approaches do not have to be forced on other signaling protocols that do not have or require such limitations.
Certainly JavaScript programmers and the W3C should not be impacted by such limitations by introducing SDP (or any universal format) into the mix when it adds zero value and fails in its primary objectives, namely: interoperability with existing SIP vendors & networks.
This further illustrates why SDP baked into the browser binary is not beneficial for SIP vendors either. They will be forced to upgrade their SIP infrastructure to support SDP packets from browsers with these kinds of extensions or be forced to utilize a JavaScript SDP re-write of SDP approach to “fix” these incompatibilities.
With an object approach, newer signaling protocols could describe multiple media streams with ease and SIP providers could ensure they only generate SDP compatible with their networks and agree on their best practices and launch new features that incorporate approaches like as PlanA, PlanB or NoPlan in a manner they deem fit rather then when the browser vendors decide to upgrade the SDP arbitrarily.
The current SDP based API model does not allow for SIP forking even though the RTC engine can allow for demuxing a media stream. The current model does not allow for one offer to be transmitted but accept multiple answers, which is legal in SIP.
A JavaScript object API would not be affected if the upper shim supported SDP / SIP style forking in the negotiation or not, so long as the basic rules of the RTC media engine is respected.
If we don’t get WebRTC 1.0 correct, fixing the API in WebRTC 2.0 may become even more difficult.
At this stage, prototypes are underway but to our knowledge there are no major commercial services deployed by more that one major vendor using the current WebRTC API. Yet, the argument to even consider an alternative is that ‘it’s too late’. Imagine trying to arguing fixing it after major networks are reliant upon specific browser implementation. Having a good but simple API architecture from the start could alleviate a lot of pressure fix a broken 1.0 in a 2.0 release before APIs become entrenched.
The cost of fixing the API issues today may pale in comparison to the cost of compatibility problems spread across entire sets of industries where constant fixes and work around may be required.
Even SIP providers and vendors have started to realize that baking SDP into the browser is not necessarily in their best interests, but they do have an interest in a simple API to use since they aren’t specialized JavaScript developers but SIP integrators.
If an alternative approach provides them SIP providers a simple JavaScript API they desire and achieves greater interoperability because of predictable, controllable and tailored SDP for their network, would they not prefer such a model over the current “baked in the browser” approach?
If the current WebRTC specification was ever rebooted, the current mandated SDP based API would undoubtedly be scrapped in favor of a better approach without its inherent design and use case flaws with negative long term compatibility consequences.
One argument put forward, to keep the current SDP model, proposes the current WebRTC SDP-based API must be completed soon and incremental improvement approach can be used to gradually move away from these obvious problems.
The trouble with an incremental approach is that may increase incompatibility further. Not all browser vendors will match the incremental improvements in unison nor will all customers upgrade simultaneously. This puts the onus on JavaScript developers to support multiple versions of the WebRTC API and increase the number of APIs they must learn and maintain. The JavaScript developers must still perform all the workarounds required for the current API and limit using any additional APIs until all browsers universally support the incremental improvements. This will slow innovation and adoption of future improvements.
This will also create a situation where browser vendors cannot easily achieve compliance because they too must support the existing API and incremental improvements along the way.
Having a good solid simple foundation is key to ensuring basic compatibility while allowing for innovation to occur for those developers who are willing to give new APIs a trial without needing to supporting multiple sets of equivalent but incompatible APIs simultaneously.
An alternative JavaScript model, other than the model advocated in this draft, would be to create a JavaScript session description format construction API. Such an API would use JavaScript objects to construct the session description format rather than allowing direct control of how media should be plumbed together from JavaScript.
While using SDP as the chosen format for WebRTC highlights the issues described in this draft particularly well, using an alternative format like JSON instead of SDP does not remove many of the issues presented in this draft. The issues expressed are not solely caused by the lack of expressiveness of the SDP format but the nature of creating a universal format to supersede all formats and the difficulty in producing and implementing any standardized format.
A few years ago there was an attempt to create a new “SDP 2.0” format with a draft named Session Description and Capability Negotiation [I-D.ietf-mmusic-sdpng]. This effort to create the “ultimate” SDP format in XML was ultimately abandoned, in no small part because of the difficulties in coming up with a single solution that works for all scenarios.
Given the difficulty in creating a universal format that works for all scenarios, the idea that creating a JavaScript based API that constructs a similar flexible but well defined universal session description format using JavaScript objects is highly suspect. The reality is such an effort is complex.
Even if successful, this format is not necessarily the format that will be sent on-the-wire, especially for existing signaling protocols. As such, the format will still need to be transformed into a alternative formats by JavaScript (or by a proxy). If the format must be parsed or interpreted by an intermediate then the format becomes an interaction point to the browser no matter how clever the JavaScript session description construction API implementation. Whatever format is selected, each browser or alternative protocol format will have to decide how to convert and interpret the output and generate new compatible inputs and deal with the variations that will undoubtedly arrive from browser to browser and from version to version.
Even if JavaScript APIs are made available to simplify the construction or interpretation of a defined format, this format would still becomes a do-everything serialization access point for the browser and the defined exchange point the local and remote browser. Therefor the format itself must described in meticulous detail.
The standardization requirements for such an approach would increase substantially over the WebRTC JavaScript Object API advocated by this draft since not only would such a JavaScript format construction API have to be standardized (as any JavaScript API would) but the format it outputs from the API has to become standardized in detail as well.
Every combination of this format possible would needs to be outlined. Any slight variations would likely cause JavaScript developers or other browsers to break their implementations. Obtaining 100% stability in such an output equally across all browsers, on all platforms with all versions is highly doubtful.
While a JavaScript format construction API is merely hypothetical at the time of writing this draft, any proposal will need to be vetted to see if it address all the concerns and issues brought up in this draft.
This hypothetical JavaScript session description construction API still puts the emphasis in driving the developer towards building up a media signaling exchange format rather than in the logic of how the media should be controlled and pipelined.
The JavaScript Object API model is being proposed as the alternative. In a follow-up to this draft the API will describe how the JavaScript developer gains control over the stream’s pipelining for the browser’s media/RTC engine and thus free the JavaScript developer to express signaling using whatever mechanism desired. A simplified shim implemented entirely in JavaScript will allow easier interpretation to any format desired by the JavaScript developer in a way that can be updated independent of a browser’s binary release. Should any changes be needed, a JavaScript shim generated format is strictly under the control of the network provider.
This is a typical scenario widely adopted SIP technique of an SDP attribute to place a stream on / off hold. This is the accepted methodology and performing alternative approaches would deviate from the expected practices for use with SIP and its manipulation of SDP. Although not officially documented, it is effectively supported in WebRTC implementations. This is a typical use case need by media application:
These are the steps that must be performed by a JavaScript developer:
This also implies that:
This is a very basic use case that is extremely complex for a JavaScript developer, but it is the only way to perform this particular action which is supported by the browsers, except only via the “SDP surface API”. Even if this particular use case ends up being an exposed JavaScript method to manipulate the SDP by the browser, there are countless other scenarios where tweaking a field to modify the behavior in the format will only be only available via SDP manipulation.
As WebRTC is a web API and not a SIP API, the API must be capable of allowing for alternative signaling methods without enforcing it’s own signaling aspects (other than basic principles like ensure ICE agreement has been achieved for security reasons).
Consider the following scenario: 1) Browser A and Browser B establish a connection 2) Browser A and Browser B use one-sided constraints negotiation where each party independently decides what “it expects to receive” 3) Browser A decides that it wishes to alter the properties of the video it expects to receive
With the current model: Browser A must be capable of independently modifying its expectations without waiting for an answer from the remote side (as that’s illegal by the nature of the offer / answer signaling), unless the rules are relaxed and special exceptions are made.
To achieve this a for one-sided negotiation: 1) Browser A’s JavaScript must respond to every SDP offer with an answer locally generated from JavaScript without a round trip, extracting out last known expectations from the remote SDP last received as part of the answer 2) The JavaScript must update the constraint signaling for the remote party 3) Browser B’s JavaScript sees the constraints have changed from Browser A thus it initiates a fake offer from the remote party (generating the intentions of the constraint and generating an SDP format) 4) Browser B’s JavaScript must examine the answer if any constraints of changed, and if so, it may trigger another reverse situation where step 1 is repeated, except with Browser A and B’s role reversed.
Is this really doable? Maybe, with a great deal of difficulty and SDP mangling but it is unquestionably a hack and a violation of offer / answer (and relaxed rules create exceptions and exceptions require additional logic to handle). The offer / answer rules are violated because no round trip was performed at the time when the constraints were changed.
This is also fragile because if Browser B failed to accept the fake offer there is no way to enforce the constraint nor can the JavaScript rollback the expected constraint. Likewise if the state machine in Browser A expected an offer to be generated before a new offer would be accepted, the conflict resolution process would be extremely difficult and messy.
This offer / answer state machine is not even required to fulfill the mandate of the RTCWEB Working Group charter but it is currently mandated because it supposedly makes producing “SIP interoperability” easier (which is highly suspect as shown herein).
A JavaScript shim approach with an object model and without offer / answer could achieve the same “SIP interoperability” without breaking other stateless negotiation models, such as one-sided negotiation.
Technically, there is no need for independent SDP offer / answer negotiation amongst all these peers to achieve a mesh scenario for this use case. Each client has enough information about the other clients to establish a peer connection. The current WebRTC SDP API imposes independent round trip negotiations that are not technically necessary.
A JavaScript Object API approach would allow for this kind of scenario without independent round trip negotiations for each WebRTC client in the mesh.
Consider the following scenario:
Under this scenario, what should browser B do? To reject the offer means communication cannot occur. To accept the offer has ambiguous meaning because the answer might have misunderstood the extension’s intention and does not allow for the appropriate behavior.
The exact rules of what is allowed in SDP and what is not and how extensions are treated must be defined clearly and non ambiguously. The current SDP offer / answer API is ambiguous for extensions.
Assuming that a lack of response to an extension is non-agreement to use the extension is not acceptable. For example, if the extension was security related dictating some security precondition to opening a stream, the offer must be rejected as the precondition cannot be met. Ignoring the extension would mean the offer was accepted where it cannot be accepted.
Consider the following scenario:
This is not an unrealistic scenario by any stretch of the imagination. This currently happens in the SIP world, but at least in that world new devices are tested to ensure compatibility before roll outs occur on the network so issues can be addressed before the user’s experience is broken. Since the SIP provider and gateway vendor do not have control over the update cycle of the browsers, their users are much more prone to breakage by taking the SDP from the browser and sending to their network.
Whereas this is what happens with a JavaScript Object API model with SDP shim written in JavaScript-only:
The reason why the browser update does not affect the gateway is because the SDP is generated entirely in JavaScript and thus updates to the browser do not change the SDP generation logic. The SDP is entirely in control of SIP network provider. Any bugs with SDP compatibility can be addressed by the SIP provider without changes in the browser’s binary. Bugs, updates and improvements are completely within the boundary and control of the SIP network provider.
Consider the follow scenario:
Using the current WebRTC API, this would require an offer / answer round trip to perform the change and thus the quality would be updated until the answer was acknowledged. This round trip is unnecessary technically since the bit-rate can be dynamically adjusted without remote acknowledgment. Yet, the current SDP API model imposes a round trip (unless yet another exception to the SDP rules are defined).
The browser can expose simple object methods, properties and events representing the various RTC components at an abstracted level and provide a solid API for controlling how the media should be pipelined.
A JavaScript-only shim can be layered on top of this API providing easy SDP offer / answer capability for those who want a similar “simple” API for use with SIP but on top of the object model (but developer can chose not to use this shim if they do not need SDP). Likewise the object model could be used to produce alternative formats to SDP if the same simplicity is needed but in an alternative on-the-wire session description format.
The API described in the solution will be presented in a follow-up draft and referenced once available. This solution will allow for the RTCWEB Working Group to complete its chartered mandate without starting from scratch. Any of the drafts proposed to solve issues in expressing SDP for WebRTC can be moved to more appropriate working groups. For example, SDP for SIP issues can be moved to the appropriate SIP working groups and multi-party SDP to the MMUSIC (e.g. drafts like PlanA, PlanB or NoPlan).
By having a JavaScript object API, the exact inputs, outputs, properties and events can be well defined as an API and will be designed to be a specific contract between browser vendors and JavaScript developers.
New objects and methods can be added without breaking existing compatibility. Compliance can be verified with unit tests able to test each and every behavior across all browsers versions on every platform. JavaScript developers can expect their version of the API contract to remain fixed to expected behaviors and not break (unless through well planned deprecation).
Any extensions added to a JavaScript object model does not change the behavior expectation from JavaScript developers when using the current version of the API regardless of any extensions. This is unlike SDP where extensions could be silently added into the SDP produced by the browsers. Any component that consumes the SDP may be unaware what those additional feature behaviors imply or require as a result.
Signaling related bugs produced by the JavaScript shims can easily be fixed and updated at any time regardless of the browser’s release cycle. If a SIP provider discovers their SIP is not compatible with the JavaScript shim, the SIP provider can update the shim code to their own needs dynamically without waiting for a browser to be patched and updated.
With a JavaScript API, the features are well defined so the surface API is fixed to the agreed contract. Once agreed, a browser vendor only has to ensure their compatibility with well defined limited scope unit tests, and need not worry about some free-form protocol that may introduce untold compatibility issues after a browser is released in real-world scenarios. This is also true of any non-browsers that may with to implement and be compliant to the WebRTC API for JavaScript and provide their own JavaScript and WebRTC engines.
While SIP is not the main RTCWEB Working Group charter responsibility for WebRTC, SIP compatibility is highly desirable. By exclusively generating SDP from a JavaScript shim, the SDP produced will be identical across all platforms and all devices with every browser version and entire under the control of the SIP provider. This increases compatibility for SIP providers. The SDP produced from the shim can be custom tailored to a SIP network without affecting any other SIP vendor or harming compatibility with other utilizing WebRTC.
With a JavaScript shim approach on top of an object model, the information going over the wire can be transformed from the JavaScript object properties to alternative formats, including JSON, XML or SIP (or anything custom). As the JavaScript shim to use is under control of the service provider and identical regardless of the platform, the output from the JavaScript format generation is consistent and controllable thus ensuring maximum compatibility with the network and signaling utilized.
The party receiving this format can be sure the format is to exacting specification of their choosing rather than relying on whatever format is produced by whatever browser vendor.
The JavaScript developer should not need to understand the mechanics of RTC other than understanding how to plumb the objects together. Those whom need extended properties or events for finer control can obtain them with simple method access to an object, but those extended attributes should not be required for simple use cases.
The objects model should allow a simple method for collecting information that will be needed for various alternative negotiation models. One of the targets for negotiation must be SDP and SIP.
The proposed JavaScript object model should not require the offer / answer state machine but should not preclude this state machine being built in a layer above. The offer / answer state machine must be possible to implement as a JavaScript shim without any additional built-in browser services needing to be implemented.
Extending the object model for the expected common extension use cases without breaking the JavaScript API must be possible. Such possible extension use cases should include items like local mixing and data synchronization or extended properties, events or features.
As any design, there may be limitations but the design should hold up to various realistic scenarios that are likely to happen in the near future.
The API must be possible to describe specific API behavior sets to the browser vendors so they have the appropriate guidelines for implementation, including the mapping to on-the-wire to RTC protocols. The API presented in the follow up draft may be the input to a W3C efforts to define specific and exact expected behavior sets for an object based JavaScript API for an official WebRTC 1.0 release.
The proposed JavaScript Object model will have an API to define the basic JavaScript usage of the data channel.
The API must adhere to the expectations of the RTCWEB charter either directly, via extensions that can be defined by the working group on top of the object model or possibly via a JavaScript shim written to utilize the functionality of the object model but it must not preclude the RTCWEB charter from fulfilling its goals.
The goal of the object model is to allow for a JavaScript shim that provide a simple mechanism for parsing and generating SDP for basic compatibility with SIP networks (capable of supporting the WebRTC wire protocols).
The goal of this object based API is not to provide working JavaScript shim on top that is a 1-for-1 matching of the current WebRTC API as a shim, including all behaviors, features, bugs and expectations since the definition of the current API is not defined enough to be able to produce that level of compatibility. This would be an impossible goal as a result and would add little value.
Extensions are beyond the scope of the JavaScript shim, but it is possible for others to fork and modify the shim to their own needs specific to their own SIP/SDP network infrastructure.
Compatibility with the SDP used in all SIP networks is not a stated goal for the JavaScript shim since not even SIP providers can agree on a common agreed definitive standard set of RFCs and drafts.
A JavasScript object model would remove much of the need for cross IETF working group coordination, which has become common place with the current movement because of utilizing SDP. By limiting the RTCWEB technologies used to only those required for Real-Time Communication from the browser (e.g. RTP, ICE/STUN/TURN, DTLS), the RTCWEB Working Group is freed from tight couplings with other IETF working groups, each having their own charters, schedules, agendas and interests and thus ensures more rapid progress between RTCWEB Working Group the W3C and developers who are to use this technology.
While RTCWEB has it’s own security considerations for protocols, a JavaScript object model has no additional requirements other than those already established for use within RTCWEB, e.g. ICE connectivity permission check or DTLS fingerprint checks.
JavaScript as a browser language itself has security consideration but nothing inherent to using a JavaScript object model versus a JavaScript SDP API model. Any proposed implementations must have a JavaScript API. The specifics of any API must have their own security considerations should any exist.
Any specific issues for the proposed JavaScript object model will be outlined in the separated draft JavaScript Object API draft as needed and warranted.
| [RFC4566] | Handley, M., Jacobson, V. and C. Perkins, "SDP: Session Description Protocol", RFC 4566, July 2006. |
| [RFC3264] | Rosenberg, J. and H. Schulzrinne, "An Offer/Answer Model with Session Description Protocol (SDP)", RFC 3264, June 2002. |