Re: [Moq] Scope of MOQ and ULL Streaming

[Spencer Dawkins at IETF <[e-mail]>]

Hi, Nathan,

On Sat, Feb 11, 2023 at 3:49 AM Suhas Nandakumar <[e-mail]>
wrote:

>
> Hello Nathan
> Few responses inline under [Suhas]
>
> On Fri, Feb 10, 2023 at 11:10 AM Nathan Burr <[e-mail]>
> wrote:
>
>> A lot of the CDNs have done a lot of work to support newer push data
>> flows. At Edgio ( or whatever it was called ) we did a lot of work to
>> support chunk-transfer encoding so that all data was streamed from origin
>> to edge as the encoder was writing its output. The nice thing about it,
>> after the chunked transfer PUT was complete, the full segment was the same
>> as a high latency segment. so everything was already cached. The problem
>> with chunked transfer is it was nearly impossible to measure line speeds in
>> the browser.
>>
>> The best alternative was to keep our data flows the same but signal in
>> the manifest when a part was fully available at the origin using byte-range
>> requests. Everyone in the industry as far as I'm aware did the same thing.
>>
>> If MOQ ends up being something similar to chunk-transfer but with
>> prioritization and better bandwidth estimate, it would be a good step
>> forward.
>>
>> What I'm trying to say is that I don't see the step from chunked transfer
>> encoding to MOQ as being that big of a leap for most CDNs. I would imagine
>> that the biggest issue for them is if the connection is persistent or not.
>> Which might be something we want to address here.
>>
>>    - How long should connections be open for?
>>    - Is there a way for the server to tell the client to go away
>>    gracefully? So the client can connect to another server or another CDN and
>>    switch the stream with minimal issues before closing the current stream.
>>
>> The other thing is that I don't think CMAF makes sense here. And perhaps
>> that is beyond the scope of MOQ. For lower latency interactive content,
>> CMAF isn't a great format. But it would make more sense for CDN
>> efficiency.
>>
>
>
> [Suhas] This was discussed during the recent interim on separating
> container format, There has  some thinking started already on these lines.
>
>
>> So as far as adoption goes, If you want to make it easy for adoption.
>> chunk-transfer v2 makes sense. But I don't think you will ever see latency
>> lower than 1 second. I would estimate 1-2 second range still.
>>
>> I don't think that's what we are really wanting to do here. Is it?
>>
>
> [Suhas] Sub-seconds latency for interactive flows is part of MoQ's scope.
>

What Suhas says is certainly true (repeating the mantra, "This solution
addresses use cases including live streaming, gaming, and media
conferencing" from our charter).

I wonder if you were saying something slightly different - that you would
be surprised to see *measured *latency under 1 second.

Best,

Spencer