Hi,

> On Dec 9, 2022, at 8:48 AM, Shihang(Vincent) <[email protected]> wrote:
> 
> I guess you are referring to the lifetime message property (see 
> https://www.ietf.org/archive/id/draft-ietf-taps-interface-18.html#section-9.1.3.1).
>  We are improving the deadline delivery by deadline-aware scheduling and 
> redundancy(FEC).

Yes.


> The scheduling is doable on top of QUIC (may involve a lot of timer 
> overhead). But the FEC requires the knowledge of packet boundary which can 
> not work on top of QUIC, at least not QUIC stream.  

That seems orthogonal to my statement about TAPS: my comment was an add-on to 
the mention of the TTL parameter in the email from Francois below. I just meant 
to say that TAPS offer such a parameter too.

Cheers,
Michael


> 
> -----Original Message-----
> From: QUIC <[email protected]> On Behalf Of Michael Welzl
> Sent: Thursday, December 8, 2022 6:35 PM
> To: François Michel <[email protected]>
> Cc: Ian Swett <[email protected]>; Chuan Ma <[email protected]>; 
> [email protected]; Christian Huitema <[email protected]>; [email protected]; 
> [email protected]
> Subject: Re: Add deadline awareness to QUIC
> 
> FWIW, the TAPS API has deadline-awareness, and QUIC can operate underneath it…
> 
> 
> Sent from my iPhone
> 
>> On 8 Dec 2022, at 11:26, François Michel <[email protected]> 
>> wrote:
>> 
>> Hi,
>> 
>> I wasn't aware of this ttl parameter in Chrome. We also studied quite a bit 
>> the addition of deadline-awareness to QUIC a while ago on a ToN article 
>> right there, see Section VIII:
>> https://dial.uclouvain.be/pr/boreal/object/boreal%3A264704/datastream/PDF_01/view
>> 
>> In short, we added an API to our QUIC implementation to be able to send 
>> "messages" with deadlines attached to it, each message being sent on its 
>> dedicated QUIC stream. Attaching a deadline to a message means that its 
>> stream must be totally delivered within that time, so I guess the ttl field 
>> of Chrome behaves similarly. This allowed us to schedule FEC in a more 
>> clever way, and it might be the case for multipath scheduling too. For 
>> instance with FEC we could spare a lot of bandwidth by protecting several 
>> video frames at once using the deadline and framerate information (Figure 16 
>> in the article).
>> 
>> That being said, we did not make any specific protocol modification to use 
>> deadlines, we just modified the API to be able to attach deadlines to 
>> streams, like the ttl feature of Chrome. That affected the stream scheduling 
>> and FEC scheduling, but no specific frame was required on our side so the 
>> QUIC specification stayed unmodified.
>> 
>> Looking at MoQ and Webtransport, it could be useful to already provide 
>> deadline information from the WebTransport API and forward this info to QUIC 
>> to be able to do clever streams/fec/multipath/whatever scheduling.
>> 
>> Right now I am working on adding deadline awareness on our new FEC implem on 
>> top of quiche-cf, so do not hesitate to reach out to me if you have any 
>> question or comment.
>> 
>> Cheers,
>> 
>> François
>> 
>>> On 21/11/22 12:16, Ian Swett wrote:
>>> This looks like it might be conflating a few features, some of which are 
>>> very media specific and some which are more general, such as prioritization 
>>> and deadline awareness.
>>> I may have missed it, but I'm not sure I understand why the deadline 
>>> awareness requires a new frame.  For low latency media, the sender 
>>> typically knows what the deadline is, so it has no need to communicate it 
>>> to the peer, unless this design has relays in mind.  If so, Media over QUIC 
>>> is even more the right venue for this.
>>> For example, the Chromium QUIC code has had a ttl feature for a few years 
>>> that we've used for various use cases:
>>> https://source.chromium.org/chromium/chromium/src/+/main:net/third_party/quiche/src/quiche/quic/core/quic_stream.cc;l=1384
>>>  
>>> <https://source.chromium.org/chromium/chromium/src/+/main:net/third_party/quiche/src/quiche/quic/core/quic_stream.cc;l=1384>
>>> Ian
>>> On Tue, Nov 15, 2022 at 11:03 AM Chuan Ma <[email protected] 
>>> <mailto:[email protected]>> wrote:
>>>   Hello Huitema and Sarikaya, thanks for the reply.
>>>   You are correct that some of our designs are similar to MoQ answers.
>>>   For example, we both use partial delivery to improve real-time app
>>>   delivery. Actually, we sent the draft to the MoQ mailing list and
>>>   discussed the detailed design and which layer to place the module.
>>>   But our draft focuses on providing general deadline-aware transport
>>>   service as a QUIC extension which is out of the scope of MoQ. MoQ
>>>   builds on top of QUIC and is explicitly tailored to media delivery.
>>>   We would like to hear the opinion of the QUIC community to see the
>>>   need for such a *general deadline-aware transport service*.
>>>   On Tue, Nov 15, 2022 at 7:12 AM Behcet Sarikaya
>>>   <[email protected] <mailto:[email protected]>> wrote:
>>>       On Mon, Nov 14, 2022 at 1:05 PM Christian Huitema
>>>       <[email protected] <mailto:[email protected]>> wrote:
>>>           Hello Chuan Ma,
>>>           Your draft aligns very much with some of the options
>>>           investigated for
>>>           "media over QUIC". Have you considered participating in that
>>>           working group?
>>>       I agree, this looks very much like media transmission material.
>>>       Behcet
>>>           -- Christian Huitema
>>>           On 11/14/2022 10:20 AM, Chuan Ma wrote:
>>>> Dear all:
>>>> 
>>>> I'm Chuan Ma from Tsinghua University. I want to discuss
>>>           the deadline
>>>> awareness of the current application and whether we
>>>           should add it to the
>>>> QUIC protocol.
>>>> 
>>>> Applications may have specific deadline requirements for
>>>           data transmission.
>>>> For instance, a video conference application may require
>>>           sending the data
>>>> with a deadline of 200ms to enable live interaction. The
>>>           application may
>>>> drop the data that misses the deadline, even if the data
>>>           has already
>>>> reached the other end. In this case, it is possible to
>>>           drop the data after
>>>> the given deadline from the sender to save bandwidth and
>>>           decrease queuing
>>>> time. Deadline requirement is also helpful to offer
>>>           deadline-aware
>>>> scheduling combined with QUIC stream priority. Such
>>>           scheduling methods can
>>>> increase the punctuality of QUIC and allow more data to
>>>           arrive on time. It
>>>> is possible to extend QUIC and offer deadline-aware
>>>           transport as a service.
>>>> 
>>>> Nowadays, deadline-aware data transmission is getting
>>>           more and more
>>>> popular. Applications that emphasize real-time
>>>           interaction, such as VR/AR,
>>>> gaming, and video conference, are drawing more and more
>>>           attention. So
>>>> deadline-aware transport has a lot of use cases. However,
>>>           currently, it is
>>>> the application that is responsible for realizing
>>>           deadline-aware transport.
>>>> This transport service should be offered by the transport
>>>           protocol but not
>>>> be left to the application. It is reasonable to provide
>>>           such transport
>>>> service in a new transport protocol, and QUIC is a good base.
>>>> 
>>>> We wrote a draft to show this idea in
>>>> https://datatracker.ietf.org/doc/draft-shi-quic-dtp/
>>>           <https://datatracker.ietf.org/doc/draft-shi-quic-dtp/> and
>>>           implemented a
>>>> protocol based on QUIC called DTP (Deadline-aware
>>>           Transport Protocol) in
>>>> https://github.com/STAR-Tsinghua/DTP.git
>>>           <https://github.com/STAR-Tsinghua/DTP.git>. We also built a
>>>           live stream
>>>> prototype application to compare the performance between
>>>           DTP and QUIC (
>>>> https://github.com/STAR-Tsinghua/LiveEvaluationSystem.git
>>>           <https://github.com/STAR-Tsinghua/LiveEvaluationSystem.git>). We 
>>> found that
>>>> DTP outperformed QUIC in improving data transmission
>>>           punctuality and saving
>>>> bandwidth resources. We published the paper in ICNP22 and
>>>           built several
>>>> systems like proxy and tunnel based on DTP. It would be a
>>>           good idea to give
>>>> this method a try.
>>>> 
>>>> We'd like to know what you think about this topic. Please
>>>           let us know if
>>>> you have any comments.
>>>> 
>>>> Best regards.
>>>> 
>>>> Chuan Ma
>>>> 
>> 
> 
> 

Reply via email to