[vls-devel] Re: Patch to regulate network output

[Jean-Paul Saman]

Mike Albon wrote:
>>No, what you suggest is bandwidth limiting. It would not solve the large 
>>bursts Marian Durkovic is talking about.
>>It would just chop of bandwidth exceeding traffic or delay udp packets. 
>>Both are evil.
>>
>>Delaying packets of video stream will make the video/sound very choppy 
>>at the receivers end. In other words it will kill the real time 
>>behaviour of a video stream.
>>
>>A real bandwidth limiter will throw away the packets that exceeds the 
>>bandwidth limitation. This will also result in poor video and audio 
>>quality. It might even crash the video decoder.
> 
> 
> I understand what you are saying here, but isn't packet shaping what is
> actually happens within a VBR TS stream over DVB anyway. A certain
> amount is allocated to each channel with a percentage of their allocated
> bandwidth that they can burst for statistical multiplexing so the whole
> I frame wouldn't be transmitted in a single burst but smeared out a
> little bit. 
That is approximately what they are doing (AFAIK), but the PCR's are 
probably adjusted accordingly. I'm not an expert on DVB streaming anyway 
so the correct answer on what exactly happens here can not be given by me.

> 
> Certainly if this was happening in Single program transport stream you
> are not going to have problems with two elementary streams not having
> the correct timecode information.
> 
In DVB they interleave SPTS's with correct PCR's. So if you extract a 
SPTS from a MPTS (DVB) stream, then you introduce *holes*. Because the 
total bandwidth of a DVB MPTS channel is fixed, the encoders have to 
make sure that not all the SPTS's in the channel peak at the same 
moment. And if they do, then they'll adjust some streams to a lower 
bitrate for that time (hence VBR AFAIK).

> I also agree with your statement that most packet shapers will throw
> away packets, when the allocated bandwidth is used up, but this is what
> is already happening within Marian's switch?
> 
Network switch? A switch does not do bandwidth limiting it only corrects 
  checks for collisions before sending. It is not supposed to introduce 
a delay. Unless you see the max throughput rate as bandwidth limit.

Or patch? In Marian's patch there is a constant delay introduced which 
deliberately slows down any video stream. What actually should happen is 
what I explained before.

> I'm just trying to understand what the difference is with having a patch
> or using tc.   :)
No problem!

What is a tc?

Greetings,
Jean-Paul Saman.



-- 
This is the vls-devel mailing-list, see http://www.videolan.org/streaming/
To unsubscribe, please read http://developers.videolan.org/lists.html
If you are in trouble, please contact <postmaster at videolan.org>