On 03/01/2012 09:18 AM, Michael Welzl wrote:
Hi!
Fine by us, and certainly a welcome topic in ICCRG. We can give you a 1 hour slot within our meeting. Thank you! We'll plan accordingly, and put together a sub-agenda that we can send you ASAP.
Cheers, Michael
On 2/29/12 1:19 PM, Harald Alvestrand wrote:
Hello,
our transport AD, Wesley Eddy, suggested I should contact you about the issue of getting discussion time at the Paris IETF for the problem of congestion control for RTP-carried real-time media.
We (a small group of people working on the rtp-congestion@alvestrand.no mailing list) have come up with a few drafts, mainly based on work done at Global IP Sound and Google:
- draft-alvestrand-rtcweb-congetsion-01 - draft-alvestrand-rmcat-remb-00
These drafts suggest that one implement a delay-based, recipient-measured estimation algorithm for bandwidth estimation over RTP flows, and define a feedback message (analogous to TMMBR) for the recipient telling the sender how much data to send at any time.
There has also been discussion on the rtcweb mailing list about the desirable properties of such congestion control, including how it should interact with congestion control in SCTP, which is suggested for a data channel transport in the RTCWEB effort.
In an informal lunch meeting in Taipei, we discussed having a three-pronged approach:
- Defining a set of best practices for what one should really not do with RTP flows (such as saying "when packet loss exceeds some percentage, reduce send rate significantly"); this seems like appropriate work for AVTCORE
- Define a set of feedback messages that can be used to implement algorithms like the one mentioned above, or others, without standardizing the algorithm itself
- Creating a working group for publishing the ideas and pushing a long term effort for standardizing a congestion control algorithm. A draft charter is here: http://www.ietf.org/iesg/evaluation/rmcat-charter.txt
Due to bad planning, we did not get a BOF request in for the Paris meeting, but would really like to have a place to have this discussion in that timeframe; also, the ICCRG should have quite a few people whose input we would definitely want into this discussion.
Can you consider whether or not you can offer us some time on the ICCRG agenda for this?
Harald Alvestrand