Re: [tcmtf] BoF feedback

[Spencer Dawkins <[e-mail]>]

On 8/5/2013 1:15 PM, Wesley Eddy wrote:
> On 8/5/2013 2:02 PM, Spencer Dawkins wrote:
>> So, Martin and I were talking this morning, and a question came up that
>> you folks might want to ponder, if you haven't already done so ...
>>
>> Re: UDP/IP flows, the theory here would be to say that UDP/IP flows are
>> more like RTP flows than TCP flows, so the concerns about TCP in TCM-TF
>> don't apply, but the nice people in RTCWeb are defining their data
>> channel over SCTP/DTLS/UDP, and my guess would be that their data
>> channel would behave exactly like TCP for bulk transfers.
>>
>> Have people already considered that?
>>
>
> I think the answer is that instead of searching for particular
> protocol layerings that could be harmed, the rules for invoking
> TCMTF on a flow would be such that it only operates on things
> that it is configured to (e.g. UDP to a game server IP and port),
> and that it would "fail off" and not do anything for types of
> flows that it isn't sure how to optimize, time release for, or
> otherwise not break.

Right. That might not have been clear, but that matches what I was 
thinking - opt-in.

So I think my next question if if people expect the spec to give 
guidance about how to decide what gets configured "in".

Thanks,

Spencer