Re: RTCDataChannel characteristics and failures -API description -

On 2014-08-26 23:53, Martin Thomson wrote:
> On 26 August 2014 14:42, Michael Tuexen
> <Michael.Tuexen@lurchi.franken.de> wrote:
>>> These 10 retries will then be spread over about 500 seconds, but in a really bad network the watchdog will usually abort the Association well before that, right?
>> As far as I understand, ICE will kick in earlier...
> About 470 seconds earlier.  Though that only applies to a single path.
> There is always a chance to create a new path between the hosts and
> resume the association.
Yes, after failure it is possible to try to reopen data channels, and 
that is my main point in why I propose text about failure handling in 
both WebRTC and rtcweb. Currently WebRTC API says that sending in 
reliable channels is guaranteed through retransmissions. But there is 
always a risk of failure.
I have seen WebRTC developers in discussion groups being astonished that 
the data channels are aborted sometimes, and needing information that 
the channel can break and then a new open is needed to be able to 
continue. ( and it will sometimes be impossible to know if requested 
transmissions have reached the receiver. )

Various implementations and network environments may result in different 
times before the association and all channels are aborted. It will not 
be possible to provide exact information about that in the API spec. But 
it should be clear that failures can happen and what happens then.

I tried to compose text proposals in this thread, and hope that they can 
be adjusted and included in the API spec, and when needed coordinated 
with text in the rtcweb channel specifications.

Regards

Gunnar

Received on Wednesday, 27 August 2014 09:58:12 UTC