Re: RTX/FEC

On 12/15/2018 11:15 AM, Varun Singh wrote:
> Is there really a reason to specify when or how to use a particular
> resilience scheme as long as endpoints implement the ones that have
> been mandated in the rtcwen-RTT-usage draft are implemented?
>
> I believe that how and when a resilience scheme is used is basically
> an app behavior (mobile apps may do something different from the
> common baseline behavior, and apps that are more content aware might
> choose to do something all together different as well.) 
>
> I think apps should be able to freely use the mechanisms available in
> the toolbox. 

two different approaches are possible, which both might be appropriate
under certain conditions:

- the app is the browser, and makes the best use of the mechanisms available
- the app is the Web app, and commands the browser to do the Right Thing
(but the browser needs to do the Right Thing when the app doesn't
specify a preference)

If Web apps should be able to command the browser about which mechanisms
to use, we need API to do so.

If browsers' behavior without Web app input is to be predictable across
browsers, there needs to be guidelines to say what to do when.

Whether the browsers' behavior needs to be predictable - I don't know.
It might be an area that can be safely left to product differentiation.

>
> On Sat, 15 Dec 2018 at 3.31, Bernard Aboba
> <Bernard.Aboba@microsoft.com <mailto:Bernard.Aboba@microsoft.com>> wrote:
>
>     We should probably rexamine WebRTC robustness more systematically,
>     because we can do quite a bit better than what is there now.
>
>     IMHO, for Opus, RED applied judiciously outperforms Opus FEC
>     (which cannot handle burst loss).
>
>     For video, differential protection (e.g. RTX or FEC on base layer
>     only) is more efficient than applying FEC or RTX to all bits.
>
>     > On Dec 14, 2018, at 6:25 PM, Cullen Jennings <fluffy@iii.ca
>     <mailto:fluffy@iii.ca>> wrote:
>     >
>     >
>     > I don’t know why we ever put RTX in WebRTC - it was clearly so
>     suboptimal in RTP in general that we created much better things to
>     replace it - like ULPFEC. I’d be in favor of removing it.
>
> -- 
> Founder, CEO, callstats.io <http://callstats.io>
> http://www.callstats.io
>
> Interested in networking, media quality, and diagnostics.
> We are hiring!: www.callstats.io/jobs/ <http://www.callstats.io/jobs/>


-- 
Surveillance is pervasive. Go Dark.