Re: [MMUSIC] immortal candidates in ICE nombis

[Bernard Aboba <[e-mail]>]

On Mar 29, 2015, at 4:29 PM, Christer Holmberg <[e-mail]> wrote:
>> 
> Shouldn't it be enough to send consent requests on *one* of the candidate pairs associated a virtual connection - instead of having to send on each associated candidate pair? Then, if an interface is taken out of service, you simply switch to another candidate pair. Similar, if you don't receive consent responses from the remote endpoint (e.g. because it has taken the interface out of service) you simply try another candidate pair.
> (Keep-alives of course need to be sent on each candidate pair, in order to maintain NAT bindings.)

[BA] I sent a similar comment earlier.  While testing all pairs you want to keep alive ensures consent, it is also expensive in terms of energy consumption.  The question is what is required to keep NAT bindings open, assure consent, and verify connectivity prior to a switch if all desired live pairs are not tested. 

You cannot just switch to another candidate pair if there is no indication the remote candidate is still live (e.g. No recent response to a consent check from any local candidate, no indication via a timeout attribute that it is still alive, etc.)  At a minimum re-trickling the local candidates you want to re-use would be needed, to restart pair formation and testing after the candidate pair was pruned.