Re: [rtcweb] RTP Usage: Is RTP Retransmission REQUIRED or RECOMMENDED

[Cameron Byrne <[e-mail]>]

On Wed, Jun 27, 2012 at 1:50 PM, Fabio Pietrosanti (naif)
<[e-mail]> wrote:
> On 6/27/12 10:23 PM, Cameron Byrne wrote:
>> On Wed, Jun 27, 2012 at 1:06 PM, Colin Perkins <[e-mail]> wrote:
>>> On 27 Jun 2012, at 17:19, Cameron Byrne wrote:
>>>> I think Bernard noted this as well, but it is not so much buffer
>>>> bloat... but the issue is indeed latency, not loss.  Consequently,
>>>> re-transmission not productive.
>>>
>>> This surely depends on the network being used, and the acceptable latency for the application. Making blanket statements that retransmission is, or is not, productive are not helpful, since it's usefulness is very context dependent.
>>>
>>> Colin
>>
>> Allow me to set context.
>>
>> A comment was made that re-transmission of packets is a good idea
>> because mobile networks drop a lot of packets.
>>
>> My comment is that re-transmission of RTP is not productive in 3GPP
>> wireless networks such as GSM, UMTS, or LTE.
>>
>> This is because L2 mechanisms do forward error correction and PDU
>> acknowledgement and re-transmission.  The RAN system does not have
>> packet loss, generally speaking.  Packet loss may occur on the
>> internet prior to entering the RAN, this is a separate issue than the
>> specific case of mobile.
>
> You are saying that in EDGE, UMTS, HSDPA when an IP stack send an UDP
> packet, it will handled from L2 like in the old age of GSM CSD and be
> automatically retransmitted?
>

Yes. LTE too.

Cameron

> The L2 RLC can goes in:
>
> * "acknolwedged mode" (handling retransmission)
>
> * "unacknowledged mode" (no retransmission).
>
> UDP streaming packets over Mobile Data Networks goes normally in
> "unacknowledge mode".
>
> As far as my professional experience VoIP over mobile can be normally be
> subject to packet loss.
>
> Fabio
>
> p.s. however i agree that's a particular use cases of WebRTC