Re: [MMUSIC] New Version Notification for draft-reddy-mmusic-ice-happy-eyeballs-02.txt

["Tirumaleswar Reddy (tireddy)" <[e-mail]>]

Hi Ari,

Please see inline 

> -----Original Message-----
> From: Ari Keränen [mailto:[e-mail]]
> Sent: Thursday, November 07, 2013 6:42 AM
> To: Tirumaleswar Reddy (tireddy)
> Cc: [e-mail]
> Subject: Re: [MMUSIC] New Version Notification for draft-reddy-mmusic-ice-
> happy-eyeballs-02.txt
> 
> Hi Tiru,
> 
> On Nov 6, 2013, at 1:56 AM, Tirumaleswar Reddy (tireddy) wrote:
> >> Have you done an exercise (or better, implementation?) on how the candidate
> >> pairs would then look like? Since ICE uses MAX/MIN sorting of the pairs
> (see
> >> http://tools.ietf.org/html/rfc5245#appendix-B.5) simply having high
> priority
> >> on the other side may not give the results you want.
> >
> > MAX/MIN logic used for calculating candidate-pair priority ensures that for
> a particular agent, a lower-priority candidate is never used until all higher-
> priority candidates have been tried. So even if the remote peer does not
> support Happy Eyeballs Extension algorithm, the ICE agent would start ICE
> connectivity checks based on the local prioritization and if IPv4 ICE
> connectivity is received by the remote peer then it would cause Triggered
> Checks for IPv4 candidate-pair. Hence IPv4 candidate-pair will be tested
> earlier even if remote peer does not Happy Eyeballs Extension algorithm. So
> effectively even if just one of the end points supports the algorithm,
> connectivity checks are optimized and should see better results.
> 
> I'm afraid that still wouldn't work like that. As you said, "a lower-priority
> candidate is never used until all higher-priority candidates have been tried".
> You would not start checks based on a local high priority for an IPv4
> candidate since a *pair* with that candidate would not have high priority (due
> to the "2^32*MIN(G,D)" part of the formula -- it's the smaller priority of the
> candidates that counts).

You are right, there is no benefit if only one endpoint supports Happy Eyeballs Extension algorithm. we will update the draft to high-light this point.

Thanks and Regards,
-Tiru.

> 
> That said, I remember getting this wrong in my own implementation the first
> time so as I said you may want to do this exercise with real numbers and see
> where you end up.
> 
> 
> Cheers,
> Ari