Re: [Covidimpacts-workshop] Forward Error Correction Qs

[Mirja Kuehlewind <[e-mail]>]

So you say the reaction to too few bandwidth was sending more data...?

Is the FEC fixed or is that somehow adopted dynamically?



> On 13. Nov 2020, at 04:55, Cullen Jennings <[e-mail]> wrote:
> 
> 
> To get a bit more specific about the webex FEC bandwidth …
> 
> I’ll talk about webex here but the the situation is similar for the other major web conferencing vendors. Note there is a lots of fine print to this… The amount of FEC bits that are  allocated is regularly changed and tuned (all the major vendors do this) and varies by things like geo, customer type, how much coffee the developers got, which version of of the product, A/B testing, and so on. What I am describing here would be the case for a significant portion of webex traffic in later summer / early fall time ranges. 
> 
> The audio codec is opus and the video codec is H.264 or slight variants of that. There are some special case use of AV1 but I'm talking just about the normal cases that are on H.264. This is for the audio / video flowing to and from a webex users desktop or mobile client. WebRTC in browsers is a different story. I’m including retransmission technique in this “FEC” bandwidth even though some people might argue it is not strictly FEC. 
> 
> The combined bandwidth for the audio + the FEC for audio is greater that the bandwidth for the video + the FEC for the video. 
> 
> Yep, I know that is not what most people would expect but it is where things ended up for some good reason, and probably some bad reasons too. The media is typically sent over SRTP which makes it pretty easy to verify theses numbers in wireshark. I don’t plan to do any tutorials on how to reverse engineer webex :-) 
> 
> The amount of audio FEC was not this high before COVID-19. This was added to deal with deteriorating network conditions relative that webex was operating in. I expect the amount of FEC to go down over time thought it is hard to guess what might happen in the short term with currently rising COVID-19 case numbers in many countries. 
> 
> I got asked about if a conference had 8 people and 5 of them had video turned off, was this a total statistic that added all the people with and without video together or something. It is not, it is about the bandwidth going from a conference to a single user that has both audio and video turned on. 
> 
> 
> 
> 
>> On Nov 12, 2020, at 9:37 AM, Livingood, Jason <Jason_Livingood=[e-mail]> wrote:
>> 
>> In Wednesday’s session I was intrigued by Cullen’s comment about forward error correction (FEC) potentially leading to audio bitrates for conferencing exceeding that of video. I was curious if that potentially pertained to add A/V codecs or just one specific codec? And can this be quantified?
>>  
>> Lastly, if this is indeed as critical an issue as it seemed, how does the feedback loop to the codec developers work so that there could be a software fix leading to a real world improvement?
>>  
>> Thanks!
>> Jason
>> -- 
>> Covidimpacts-workshop mailing list
>> [e-mail]
>> https://www.iab.org/mailman/listinfo/covidimpacts-workshop
> 
> -- 
> Covidimpacts-workshop mailing list
> [e-mail]
> https://www.iab.org/mailman/listinfo/covidimpacts-workshop