Minutes IETF100: rmcat
minutes-100-rmcat-00

RTP Media Congestion Control Working Group Minutes - IETF 100
=============================================================

Reported by Colin Perkins and Gorry Fairhurst

The RMCAT working group met at IETF 100 in Singapore, on 15 November 2017.
The chairs started the meeting with a review of the working group status:

- The cc-requirements, coupled-cc, and SCReAM drafts are with the RFC Editor
- The SBD draft has passed WG last call and the write-up is ready, minor 
  issues to resolve then will go to the IESG
- The NADA draft has been updated based on WG last call comments, and is
  on the agenda for this meeting
- The GCC draft is waiting an update from the authors
- The eval-test draft is blocked on eval-criteria, and eval-criteria is 
  waiting on an update. Varun Singh volunteered to do this update at the 
  last meeting, and it's hoped this will be received soon. If not, the 
  chairs will try to find an additional editor to complete the work.
- The wireless tests draft has no open issues. Reviews and implementation
  experience are needed.
- The video-traffic-model draft is ready for review, and close to WG last
  call.
- The feedback-message draft will be discussed later in the meeting.
- The cc-feedback draft has expired, but will be updated to track changes
  to the feedback-message.
- The cc-codec-interactions and framework draft are blocked waiting for a
  proposal for updates that Varun and Zahed volunteered at IETF 99. Zahed
  noted that the framework needs only small updates to align with the RTP
  topologies RFC, but that it's unclear how to align cc-codec-interactions 
  and the framework, and he's had no time to review. Thinks the framework 
  is in good shape. Xiaoqing Zhu noted that Cisco is open sourcing code 
  that implements the framework, but again she has issues finding time.
  Zahed suggested to ship the framework, and leave cc-codec-interactions
  until later: he's not convinced they need to align. Zahed to discuss with
  Varun soon, and update the WG on the mailing list with how they plan to
  proceed

The status update concluded with a review of the milestones.


## NADA

Xiaoqing Zhu summarised the changes made to the NADA draft based on
implementation feedback from Julius Flohr, and presented performance
results to show their benefits and the behaviour of the algorithm.

She also noted that they'd added Fixed-FPS and Trace-driven codecs to their
synthetic codecs test suite. These are believed to be more realistic than
the previously implemented models, and have been used for their latest
experiments. Xiaoqing believes the video-traffic-model draft has now been
completed, but it needs further review from the working group.

Xiaoqing noted that their ns3-rmcat RMCAT simulation code, that has been
previously discussed in the working group, has now been made open source.
Jonathan Lennox asked if their simulator uses RTCP feedback as in the
cc-feedback draft? Not yet, since the cc-feedback draft is still changing.

An update will be submitted to address review comments from Roland Bless
and Michael Welzl, then they believe the draft will be completed. It's
expected this will be ready for a second WG last call.

Ingemar Johansson asked if the evaluations were with CBR-like video, or
something more realistic. They're a mixture. 

Xiaoqing noted their next steps are to experiment with embedding NADA in
Mozilla browsers. Jonathan Lennox asked what feedback mechanism they're
using, and if they could use the cc-feedback draft.


## SCReAM

Ingemar Johansson presented experiments using SCReAM for congestion control
of video over mobile 4G/5G links for remote control of vehicles. This needs
low latency and good quality video, so similar requirements to interactive 
video applications that RMCAT has targetted.

Their implementation uses a non-standard RTCP feedback format, but they
intend that this will be replaced with the cc-feedback message once it's
stabilised.

Ingemar noted that video codecs can be challenging to work with, and have
limited tuning capabilities in many cases. Randell Jesup asked a number of
questions for clarification on these results.

The perform strea prioritisation using weighted credit based scheduling.
Ingemar noted some issues with this priorisation, and it works quite ok,
but there is room for improvement. He believes use of ECN would improve
performance (the ECN results presented used a simulated bottleneck, but
the others use a real network).

Xiaoqing Zhu asked for clarification on the packet loss rates observed,
and whether there were self-induced losses as the flows compete. Ingemar
noted that it was around 0.5% packet loss in the real network tests. It
would perhaps be interesting to see packet loss traces, and to explore
the effects of non-congestive packet loss on the algorithms. There was
some discussion about packet loss patterns, and it seems that it was a
little bursty.  There was also discussion on the causes of loss in the 
LTE scenario.


## RTCP Feedback for Congestion Control

Zahed Sarker presented the results of the design team on RTCP feedback.
The design team was formed after IETF 94 to propose a generic congestion
feedback RTCP message. This has now been done, and the draft has been
passed to the AVTCORE working group to finalise. Expected outcome is a
standards track RFC from AVTCORE. The chairs thanked the design team for
their work.

                                   - + -