RTP Media Congestion Avoidance Techniques (RMCAT) Working Group

IETF 106, Singapore, 19 December 2019
Reported by Colin Perkins.


1. Working Group Status

   The RMCAT working group met at IETF 106 in Singapore, in November 2019.
   The meeting was chaired by Colin Perkins (in person) and Anna Brunstrom
   (remote).

   The chairs presented the document status, noting that the requirements
   for RTP congestion control, the candidate congestion control algorithms
   and related drafts, and the evaluation drafts have been published or are
   with the RFC Editor. The feedback message format is nearing completion
   in AVTCORE, and the guidelines for providing congestion control feedback
   will be updated once the feedback message format is complete. None of
   the other drafts are active.

   The chairs suggested updated milestones for the group. The proposed
   goals are to gather evaluation results and experience for 12 months,
   with meetings only if there are results to report, and then to evaluate
   status and decide whether to pursue further work. There were questions
   around the type of evaluation results expected, with Zahed Sarker and
   Xiaoqing Zhu raising this issue during the meeting and Ingemar Johansson
   making similar comments on the mailing list. Mirja Kühlewind noted that
   it's important to consider carefully to ensure the candidate algorithms
   are safe to use. We want experimental deployment results to show this
   before the group can take candidate algorithm to Proposed Standard.
   Mirja also noted that we don't have to pick a winner. The goal is to
   show that a congestion control algorithm can be deployed without causing
   problems, allowing it go to standards track. The goal is to show safety,
   not that the candidate algorithm outperforms others.  Nitin Garg and
   Jonathan Lennox noted that they have some experiments and deployment
   experience that will be ready to present in a future meeting, to help
   support this proposal.

   Jonathan Lennox reported from the Hackathon, noting that he'd been
   working on adding visibility into tracing and debugging, working on a
   common log format with Sergio Murillo to feed into visualisation tools.

   Jonathan also noted that is would be useful to have a practical way to
   turn the test case draft into concrete tests (e.g., that could be used
   via mininet or Linux TC to run evaluations). The various algorithms and
   test cases are written from the perspective of end points, and can be
   difficult to interpret as an SFU implementer.


2. Update on NADA implementation and new evaluation results

   Xiaoqing Zhu presented some NADA evaluation results. The latest
   version of the draft-ietf-rmcat-nada-13 that's now been send to the RFC
   Editor has some clarifications, but there have been no algorithmic
   changes. The algorithm and feedback has been implemented in Firefox.
   Xiaoqing showed evaluation results compared the behaviour with NADA
   to that with the default congestion control for scenarios within the
   US and for trans-Atlantic flows to demonstrate the behaviour of the
   algorithm.

   Nitin Garg asked about use with production traffic. Xiaoqing noted that
   this has not yet been done, but that they are interested in doing so.
   Another question asked about evaluating different combinations of the
   default and NADA algorithms, but this has not yet been done.

   The draft is with the RFC Editor. This initial implementation experience
   is extremely valuable in starting to understand its behaviour.


3. RTCP feedback for congestion control

   Colin Perkins briefly presented the RTP Congestion Control Feedback work
   that's being done in AVTCORE(draft-ietf-avtcore-cc-feedback-message-05).
   The main change has been to add an additional paragraph explaining the
   relation to draft-holmer-rmcat-transport-wide-cc-extensions. There have
   been no format changes. The draft is expected to go to WG last call in
   AVTCORE shortly.

                                   - + -