Re: [MMUSIC] trickle-ice review

[Peter Saint-Andre - &yet <[e-mail]>]

On 7/7/15 1:53 AM, Bernard Aboba wrote:
> On Jul 6, 2015, at 9:16 PM, Peter Saint-Andre - &yet
> <[e-mail]> wrote:
>> As we have discovered recently (see the research the Philipp Hancke
>> has been performing on deployed WebRTC audio/video services), the
>> best call setup time and thus user experience is found by sending
>> only relayed candidates first.
>
> [BA] This result has been known for quite some time - and provides an
> argument that the recommended priority formula is sub-optimal.
> However, the question is what to do if relay candidates are not yet
> available, but host candidates are. Are you saying it is optimal to
> wait and trickle relay candidates??

The current text doesn't cover any of those nuances:

    For optimal performance, it is RECOMMENDED that an initial offer
    contains host candidates only.  This would allow both agents to start
    gathering server reflexive, relayed and other non-host candidates
    simultaneously, and it would also enable them to begin connectivity
    checks.

But we know that, if relay candidates *are* available, it is not optimal 
for the initial offer to contain host candidates only. So the 
categorical text in the document isn't right, I think.

One approach would be to cover the nuances (if you have a relayed 
candidate, send just that in the initial offer - but if not then send 
only host candidates).

Another approach would be to say "immediately send whatever candidates 
you already know about in the initial offer because this enables both 
agents to start gathering additional candidates and performing 
connectivity checks". That seems to be what the current text is actually 
trying to say. Then leave the issue of host vs. relayed out of the mix 
here because I think that's a separate issue and not specific to trickle 
ICE.

Peter

-- 
Peter Saint-Andre
https://andyet.com/