Re: [Dart] ICE might send your traffic over TCP

On Jul 24, 2014, at 9:25 AM, Black, David <david.black@emc.com> wrote:

>> I don't think different drop precedences would be useful in respect to TCP.
> 
> Clarification: "in respect to TCP" -> "within a single TCP connection".
> 
> IMHO, Ruediger's larger overall point is valid:
> 
>>> It may make sense to have different TCP flows marked by different drop
>> precedences. The less important TCP flows get throttled in the case of
>> congestion.

Yes, agreed.  Sorry, my sloppy wording.

>> -----Original Message-----
>> From: Dart [mailto:dart-bounces@ietf.org] On Behalf Of Jonathan Lennox
>> Sent: Thursday, July 24, 2014 9:01 AM
>> To: Ruediger.Geib@telekom.de
>> Cc: dart@ietf.org
>> Subject: Re: [Dart] ICE might send your traffic over TCP
>> 
>> I don't think different drop precedences would be useful in respect to TCP.
>> 
>> The problem is that the way the WebRTC APIs are structured, it's not
>> necessarily clear to an application that its media traffic is indeed going
>> over TCP, so it might make precedence requests that the browser implementing
>> the APIs can't (and shouldn't try to) satisfy.
>> 
>> The question is how the browser should behave when that happens.  I spoke to
>> Cullen about this, offline, after the DART session, and I believe he had a
>> simple proposal.
>> 
>> On Jul 24, 2014, at 3:54 AM, Ruediger.Geib@telekom.de wrote:
>> 
>>> Jonathan,
>>> 
>>> Let's briefly discuss your point below: Different drop precedences are
>> required if congestion within a PHB group may occur. TCP is designed for
>> reliable transport. In case of a loss, TCP will reduce bandwidth and
>> retransmit the dropped information. I'm not an RTP expert. The few
>> applications known to me don't seem to benefit from such a behavior, if a
>> single application flow uses different drop precedence levels in combination
>> with TCP.
>>> 
>>> It may make sense to have different TCP flows marked by different drop
>> precedences. The less important TCP flows get throttled in the case of
>> congestion. But is there a need to avoid re-ordering across different drop
>> precedences, which is an AF PHB group feature? Again, I'm not an RTP expert -
>> but that doesn't sound like making a lot of sense too.
>>> 
>>> In general, I'd be interested to learn, if some applications were
>> benefitting from ECN (or a congestion indication by marking rather than by
>> dropping). Independent from the question whether they use TCP or RTP
>> transport.
>>> 
>>> Regards,
>>> 
>>> Ruediger
>>> 
>>> 
>>> 
>>> 
>>> -----Message-----
>>> From: Dart [mailto:dart-bounces@ietf.org] On behalf of Jonathan Lennox
>>> Subject: [Dart] ICE might send your traffic over TCP
>>> 
>>> [snip]
>>> 
>>> How (say) a WebRTC implementation should handle API requests for multiple
>> drop precedences when the underlying ICE channel is TCP is unclear to me.
>>> 
>>> Jonathan Lennox
>>> 
>>> 
>> 
>> _______________________________________________
>> Dart mailing list
>> Dart@ietf.org
>> https://www.ietf.org/mailman/listinfo/dart