Re: Consensus call for adding support for non-ack eliciting DATAGRAMs

[Tommy Pauly <[e-mail]>]

> On Sep 8, 2021, at 2:03 PM, Vidhi Goel <vidhi_goel=[e-mail]> wrote:
> 
>> A small question to the editors, if this targets the general Internet - you probably have an answer, there are various possibilities -  how will this transport spec detect congestion, and what method will be used for congestion control?

Is this question for the editors of the ack-frequency draft (as opposed to the datagram draft)? Vidhi’s right for datagram as it stands—it acts like any other frame. The impact of changing ack frequency seems to be more interesting here, and is one of the reasons I believe this particular work belongs in that document, and not the datagram extension.

Thanks,
Tommy
> 
> Datagrams are ack-eliciting and would use the same (ACK-clocking) congestion control as other reliable frames in QUIC. In other words, the congestion control is a shared state for datagrams + other frames.
> 
> Thanks,
> Vidhi
> 
>> On Sep 8, 2021, at 10:26 AM, Gorry Fairhurst <[e-mail] <mailto:[e-mail]>> wrote:
>> 
>> A small question to the editors, if this targets the general Internet - you probably have an answer, there are various possibilities -  how will this transport spec detect congestion, and what method will be used for congestion control?
>> 
>> Gorry
>> 
>>> On 8 Sep 2021, at 17:41, Ryan Hamilton <rch=[e-mail] <mailto:rch=[e-mail]>> wrote:
>>> 
>>> 
>>> Well said, Ian and Martin. I agree that no change is the right outcome here.
>>> 
>>> On Wed, Sep 8, 2021 at 8:53 AM Ian Swett <ianswett=[e-mail] <mailto:[e-mail]>> wrote:
>>> Agreed, if we're going to do this, I'd like to address it in the ack frequency draft and not in datagram.  I also think there are valid use cases to not ACK stream data as well, such as Media over QUIC, where frames may not fit into a single QUIC packet.
>>> 
>>> On Wed, Sep 8, 2021 at 8:22 AM Martin Thomson <[e-mail] <mailto:[e-mail]>> wrote:
>>> No change is good.  It's nothing we can't fix trivially later if we find that was the wrong outcome.  And getting this right, even if it were needed, would be tricky. It's also not all that useful when you consider that ack frequency exists as a way to manage the cost and overhead of acknowledgments.
>>> 
>>> On Wed, Sep 8, 2021, at 21:31, Lucas Pardue wrote:
>>> > Hello QUIC WG,
>>> > 
>>> > This is a consensus call for datagram issue #42 [1] - Allow a Sender to 
>>> > Control Datagram ACKs. The proposed resolution is to close this issue 
>>> > with no action.
>>> > 
>>> > If you object to the proposal, please do so on the issue or in response 
>>> > to this message. 
>>> > 
>>> > The call will run for one week, closing at end of day on September 15 
>>> > 2021, anywhere on earth.
>>> > 
>>> > [1] https://github.com/quicwg/datagram/issues/42 <https://github.com/quicwg/datagram/issues/42>
>>> 
>