Re: [codec] #28: Layered bit-stream

["Dmitry Yudin" <[e-mail]>]

Hi,

> I understand. Because you as an application programmer want to have an
easy life, 
> the codec designer shall develop a more complicated codec? In
addition, everybody 
> should suffer from a higher bit rate? No, that is not fair.
Sorry Christian, but I'm not share your idea to personalize this
discussion.
Definitely, there are number of applications that can (more or less
efficiently) utilize 
layered stream structure. At the same time, scalability makes codec more
complex and
cause bitrate penalty due to a simple fact that
entropy(a+b)<=entropy(a)+entropy(b). 
The only question is: what penalty is acceptable and what is not? Based
on IP-MR codec 
experience I consider the price for scalability as not too high.

>> Obviously, conferencing is the most important use-case.
> No, end-to-end connections are more frequent than conference calls.
Actually, the statement was "Obviously, conferencing is the most 
important use-case for layered coder".


> I mean that the layered coding is only used within one computer. 
> It is not important in-between computers. And, it is only a
> performance optimization that make the conference gateway faster.
This is correct. No other benefits except performance saving.

Regards,
Dmitry

-----Original Message-----
From: Christian Hoene [mailto:[e-mail]] 
Sent: Thursday, May 06, 2010 5:13 PM
To: Dmitry Yudin
Cc: [e-mail]
Subject: RE: [codec] #28: Layered bit-stream

Hi Dimitry,

>Hi Christian,
>
>From application point of view, the layered stream structure allows
>server manipulate channel bandwidth individually for each user with
zero
>performance overhead. 

I understand. Because you as an application programmer want to have an
easy life, the codec designer shall develop a more
complicated codec? In addition, everybody should suffer from a higher
bit rate? No, that is not fair.

> Obviously, conferencing is the most important use-case.

No, end-to-end connections are more frequent than conference calls.

>
>> a) First, this use case is a local optimization only. Thus, the must
>not be standardized.
>What do you mean exactly? "local optimization" of what?

I mean that the layered coding is only used within one computer. It is
not important in-between computers. And, it is only a
performance optimization that make the conference gateway faster.

Sincerely,

 Christian


>
>> b) Second, instead of layered coding one can use other ways of
>tweaking the implementation
>> performance. For example, if you calculate a 512 FFT do get two 256
>FFTs for free.
>> I bet there are thousand other shortcuts which I am not aware of.
>How do this interrelates with scalability? Please, explain.
>
>Let's return back to the subject:
>    Shall layered coding be supported? - we think "yes", because ...
>(see my first sentence)
>    Who needs it?                      - answered
>    Can we drop this requirement?      - only if we have real good
>reasons for it. Do we have them?
>
>Best regards,
>Dmitry
>
>-----Original Message-----
>From: Christian Hoene [mailto:[e-mail]]
>Sent: Wednesday, May 05, 2010 11:02 PM
>To: Vladimir Sviridenko; [e-mail]
>Cc: Slava Borilin; Dmitry Yudin; [e-mail]
>Subject: RE: [codec] #28: Layered bit-stream
>
>Hi Vladimir,
>
>>2/ we think that VoIP and Videoconferencing systems are users of such
>>codecs.
>
>Could you please explain your position a bit?
>
>As far as I understand, layered coding helps if multiple streams having
>the sample content but different rates must be generated.
>For example, if a conferencing system stream the same audio stream to N
>users but each users has a different bandwidth. Just encode
>all layers and drop the higher layers for the low bandwidth users. This
>approach is easy and efficient and reduce the encoding
>complexity.
>
>The arguments against are simple.
>a) First, this use case is a local optimization only. Thus, the must
not
>be standardized.
>b) Second, instead of layered coding one can use other ways of tweaking
>the implementation performance. For example, if you
>calculate a 512 FFT do get two 256 FFTs for free. I bet there are
>thousand other shortcuts which I am not aware of.
>
>Thus, I have the opinion that layered coding is not worth the extra
>bandwidth of 20 or more percentage. It is just good locally but
>not needed for interoperability.
>
>Yours,
>
> Christian
>
>
>
>
>
>
>>Yours,
>>Vladimir Sviridenko
>>SPIRIT
>>
>>-----Original Message-----
>>From: [e-mail] [mailto:[e-mail]] On Behalf
>>Of codec issue tracker
>>Sent: Sunday, May 02, 2010 5:10 PM
>>To: [e-mail]
>>Cc: [e-mail]
>>Subject: [codec] #28: Layered bit-stream
>>
>>#28: Layered bit-stream
>>------------------------------------+---------------------------------
-
>-
>>----
>> Reporter:  hoene@...                 |       Owner:
>>     Type:  defect                  |      Status:  new
>> Priority:  minor                   |   Milestone:
>>Component:  requirements            |     Version:
>> Severity:  Active WG Document      |    Keywords:
>>------------------------------------+---------------------------------
-
>-
>>----
>> Shall layered coding be supported?
>> Who needs it?
>> Can we drop this requirement?
>>
>>--
>>Ticket URL: <http://trac.tools.ietf.org/wg/codec/trac/ticket/28>
>>codec <http://tools.ietf.org/codec/>
>>
>>_______________________________________________
>>codec mailing list
>>[e-mail]
>>https://www.ietf.org/mailman/listinfo/codec