Re: [clue] [MMUSIC] [Ice] [art] ICE, ICE-bis, and Cluster 238

[Justin Uberti <[e-mail]>]

On Fri, Oct 12, 2018 at 8:14 AM Peter Thatcher <pthatcher=
[e-mail]> wrote:

>
>
> On Thu, Sep 27, 2018 at 4:21 PM Justin Uberti <juberti=
> [e-mail]> wrote:
>
>> I agree with Peter. Chrome's implementation is already closer to 8445
>> than 5245, so I don't see any issues associated with snapping this cluster
>> to 8445 (aside from the work involved).
>>
>> On that topic, note that JSEP will need a few more changes than just the
>> addition of the 8445 reference and note; the examples will have to be
>> updated, as will the logic regarding generation of offers and answers and
>> their parsing (to deal with the new ice-option). These changes will be
>> modest but probably will need to be done by the authors.
>>
>>
> I've updated the references to 8445 (and to draft-ietf-mmusic-ice-sip-sdp)
> in this PR: https://github.com/rtcweb-wg/jsep/pull/851
>
> I've also added the ice2 processing and added ice2 to the examples in the
> same PR.
>

Thanks for doing this, Peter.

>
> What did you mean by "and note"?  Is there something more needed?
>

The note in question is the text that Adam mentioned in the original email:

*While this specification formally relies on [RFC8445], at the time of its
publication, the majority of WebRTC implementations support the version of
ICE described in [RFC5245], and use a pre-standard version of the trickle
ice mechanism described in [RFCXXXX]. The use of the "ice2" attribute
defined in [RFC8445] can be used to detect the version in use by a remote
endpoint and to provide a smooth transition from the older specification to
the newer one. *

I do think we need text similar to this to be added to S 5.11 (processing
an answer) in JSEP.

>
>
>> On Wed, Sep 19, 2018 at 9:52 PM Peter Thatcher <pthatcher=
>> [e-mail]> wrote:
>>
>>> I'm late to the discussion, and reading through it, it seems that we
>>> have a lot of back and forth without addressing Cullen's root issue.  Let
>>> me see if I understand Cullen's root issue correctly.  I think it's
>>> something like:
>>>
>>> 1.  Cisco has existing code that it wants to call "WebRTC 1.0 compliant"
>>> without changing to be compliant with 8445.
>>>
>>> 2.  Cisco has existing code that it wants to continue to interoperate
>>> with endpoints, especially Chrome, even as they make changes to become 8445
>>> compliant.  And they don't want to have to test against old and new
>>> versions.
>>>
>>> Cullen, is that accurate?
>>>
>>>
>>>
>>> OK, so some of my thoughts:
>>>
>>> 1.  I don't think there is any interop risk here at all related to
>>> timings.  If you're worried about the drop in minimum check interval going
>>> from 20ms to 5ms, don't.  Just because the spec allows for going that low
>>> doesn't mean endpoints will.  And if they do, they'll do it carefully.
>>> Endpoints can and should still choose a value that works best regardless of
>>> the min in the spec.  For example, Chrome is still using an interval of
>>> 48ms (we're not in a rush to lower it, but we have non-browser endpoints
>>> that do go lower).  And if we roll out a lower value, it will be via
>>> experiments or opt-ins and carefully tracked to make sure connectivity
>>> rates don't drop.  If any problem were found in practice, it would be
>>> quickly reverted.
>>>
>>> 2.  I don't think there is any interop risk here related to nomination
>>> either.
>>> Chrome's default behavior has never been compliant to any spec anyway,
>>> and it's never been an issue.  And like with ping intervals, any changes to
>>> implementations will be done slowly and carefully.
>>>
>>> 3.  I don't think it really matters to major implementations what the
>>> dependency graph looks like.  Whether some point to 5245 and others to 8445
>>> or if all of them point to 8445, it doesn't matter, implementations will
>>> behave the same either way.  Chrome, for example will adjust timings as
>>> works well in practice (perhaps someday to below 20ms interval) regardless
>>> of which RFCs point to 8445 and which point to 5245.  If interop issues
>>> ever do come up, then they can be fixed.  And that has nothing to do with
>>> which RFCs point to 5245 and which point to 8445.
>>>
>>> 5.  You're going to need to test against different versions of different
>>> browser no matter what the RFC references are.  ICE timings and nominations
>>> seem like the least of your testing problems.  But on the flip side, Chrome
>>> (and I assume other browsers) have been very slow and careful when making
>>> changes to the ICE code.
>>>
>>> 6.  FlexICE should go a long way to putting the web app in control of
>>> the ICE behavior.  So if you are worried about what browsers will do with
>>> ICE, I suggest supporting the FlexICE effort.  In fact, it's the result of
>>> your proposal at TPAC in 2017 for wanting to have lower-level of control of
>>> ICE...  If we get that into all the browsers, you won't have to worry any
>>> more about any of this because you'll be in control (assuming you control
>>> the web app).
>>>
>>
>>> Altogether, I don't see any reason to not reference 8445 everywhere, at
>>> least not any related to interop risk and web browsers.
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>> On Fri, Sep 7, 2018 at 9:37 AM Cullen Jennings <[e-mail]> wrote:
>>>
>>>>
>>>> On Sep 7, 2018, at 1:25 AM, Christer Holmberg <
>>>> christer.holmberg@ericsson..com <[e-mail]>>
>>>> wrote:
>>>>
>>>> > Cisco has implemented stuff that is WebRTC 1.0 compliant without this
>>>> change.. These gratuitous changes, years after the implementation were
>>>> coded, with no real benefit will ensure that we are not
>>>> > and will not become compliant with the RFC. It's unlikely we will
>>>> upgrade to the new ICE until it has real befits.
>>>>
>>>> The main reason we did 8445 was because people had identified issues
>>>> with 5245. The work was driven mostly by the WebRTC community, including
>>>> yourself and the Chrome people (or, at least the Google people), and one of
>>>> the reason it took time to finalize 8445 was because you (among others)
>>>> wanted to make sure we get things right (by making network measurements
>>>> etc). Are you now saying all those changes bring no benefit? Did we all
>>>> waste our time?
>>>>
>>>>
>>>> Our testing, which we do not share, dig not indicate an improvement of
>>>> connectivity rates. I did not see results from others that did. Some of the
>>>> early test results from others that drove this work were not reproducible
>>>> in our testing. The one thing I think most people did find is that the more
>>>> out of sync the pacing of the two agents was, the worse the connectivity
>>>> was. But all of this is water under the bridge, we have old and new ice,
>>>> people can use either. What we are talking about here is what is the
>>>> minimum bar for WebRTC 1.0
>>>>
>>>>
>>>> > It is doubtful Justin will want to implement the 8445 mechanisms of
>>>> supporting both new and old ICE. Instead, we will move to say "works
>>>> with Browser X version Y or later." We have watched at W3C as it moved to
>>>> be that unless chrome does it, it rare that it becomes a standard.
>>>> > Right here I am watching how the stuff IETF defines will be less
>>>> relevant than the issue of what chrome implements.
>>>>
>>>> What exactly would Justin have to change?
>>>>
>>>>
>>>>
>>>> For us, the largest part is having to test for both old and new - it’s
>>>> not easy to do good automated testing for ICE.
>>>> _______________________________________________
>>>> mmusic mailing list
>>>> [e-mail]
>>>> https://www.ietf.org/mailman/listinfo/mmusic
>>>>
>>> _______________________________________________
>>
>>
>>> Ice mailing list
>>> [e-mail]
>>> https://www.ietf.org/mailman/listinfo/ice
>>>
>> _______________________________________________
>> mmusic mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/mmusic
>>
>