Adaptive FEC updates

RMCAT WG, IETF91, Honolulu

Transcript

1. Adaptive FEC for Congestion Control Varun Singh, Marcin Nagy, Jörg Ott, Lars Eggert IETF 91, Honolulu, 09. November 2014 https://tools.ietf.org/html/draft-singh-rmcat-adaptive-fec-01
2. IPR •  2 IPR disclosures – Nokia – Polycom
3. Error Resilience NACK (RTX) RPS SSA UEP 3G Link Packet Loss End-to-End Delay FEC SSA: Adaptive packet sizes RPS: reference picture selection
4. RTCP Interval time Throughput media rate FEC STAY state PROBE state STAY state STAY state UP state UP state DOWN state PROBE state (A) Adding FEC (B) Swapping FEC with media (C) Partially swapping FEC with media Available Capacity CONCEPT
5. CC Framework Media Encoder Rate Controller RTP Packets Transport Layer (UDP) IP Source RTP RTCP FB Pacing Buffer
6. RFC 6363: FEC Framework Media Encoder Rate Controller RTP Packets FEC Code FEC Module RTP (De)mux Transport Layer (UDP) IP Repair RTP RTCP FB Source RTP
7. State Machine STAY UP DOWN PROBE Good conditions, probe network No more bandwidth available Stable conditions Stable conditions Unstable conditions Unstable conditions Unstable conditions STEADY PROBE DECREASE INCREASE
8. FEC Scheme •  Currently, using – parity, 1- or 2-d interleaved XOR (for burst loss) – There is discussion in RTCWEB on FEC schemes. •  But, – It could be applied to other schemes S, Holmer, et al., "Handling  Packet  Loss  in  WebRTC", Proc. of IEEE International Conference on Image Processing (ICIP 2013) , 9 2013.
9. RTCP Feedback •  RLE of Post-repair (RFC5725) •  RLE of loss packets (RFC3611) •  RLE of discarded packets (RFC7097) •  Packet count of lost and repaired packets – (draft-ietf-xrblock-post-loss-repair)
10. Calculating goodput •  Receiver reports goodput: – Goodbytes in reporting interval = bytes received – bytes discarded OR •  Sender calculates goodput: – Goodbytes in reporting interval = bytes sent – bytes lost – bytes discarded RLE reports
11. Undershoot •  Congestion reported –  Sender calculates duration of congestion = HSN when RTCP scheduled – HSN when congestion detected –  undershoot bytes = sending rate * duration/8 –  New rate = Goodbytes – undershoot bytes [1] OR –  Delta = Sending rate – goodput –  New rate = sending rate – 2 x delta [2] [1] http://www.netlab.tkk.fi/~jo/papers/2012-videv-rate-control.pdf [2] http://www.netlab.tkk.fi/~varun/nagy2014mmsys.pdf
12. Typical results •  ~90% recovery in time for decoding when FEC interval was short – Fewer packets protected •  ~80% recovery in time for decoding when FEC interval was long – More packets protected
13. Applicability •  Implemented over a delay-based congestion control – See paper for details •  However would like to generalize it. – Apply to SCReAM, GCC, … Nagy M., Singh V., Ott J., Eggert L., Congestion  Control  using  FEC  for  Conversational  Multimedia   Communication, Proc. of ACM Multimedia Systems, Singapore, SG, Mar, 2014,
14. Evaluation (1/3) •  Ns-2 simulation, Variable link capacity, Single flow on the link 0 50 100 150 200 250 300 350 400 0 100 200 300 400 500 600 700 800 900 Throughput [kbps] Time [s] OWD: 50ms
15. Evaluation (2/3) 0 100 200 300 400 500 600 700 800 900 Time [s] OWD: 100ms
16. Evaluation (3/3) 0 100 200 300 400 500 600 700 800 900 Time [s] Link capacity FBRA FEC probe rate OWD: 240ms
17. 50ms 0 500 1000 1500 2000 2500 3000 0 20 40 60 80 100 Throughput [kbps] Time [s] Link capacity FBRA FEC probe rate
18. 50ms 0 0.2 0.4 0.6 0.8 1 0 20 40 60 80 100 RTT [ms] Time [s] RTT
19. 100ms 0 500 1000 1500 2000 2500 3000 0 20 40 60 80 100 Throughput [kbps] Time [s] Link capacity FBRA FEC probe rate
20. 100ms 0 0.2 0.4 0.6 0.8 1 0 20 40 60 80 100 RTT [ms] Time [s] RTT
21. RTT fairness (50ms) 0 500 1000 1500 2000 2500 3000 3500 4000 0 20 40 60 80 100 Throughput [kbps] Time [s] Link capacity FBRA flow 1 FBRA flow 2 FBRA flow 3
22. 5.4: RTT fairness 0 0.2 0.4 0.6 0.8 1 0 20 40 60 80 100 RTT [ms] Time [s] RTT flow 1 RTT flow 2 RTT flow 3
23. RTT fairness (100ms) 0 500 1000 1500 2000 2500 3000 3500 4000 0 20 40 60 80 100 Throughput [kbps] Time [s] Link capacity FBRA flow 1 FBRA flow 2 FBRA flow 3
24. RTT fairness (100ms) 0 0.2 0.4 0.6 0.8 1 0 20 40 60 80 100 RTT [ms] Time [s] RTT flow 1 RTT flow 2 RTT flow 3
25. Compete with short TCPs 0 100 200 300 400 500 600 700 800 900 Time [s]
26. TESTBED Evaluation (1/2) •  1Mbit/s link capacity, 50ms one-way-delay, Two RTP flows on the link 0 200 400 400 200 0 0 50 100 150 200 250 300 350 400 0 40 40 0 Throughput [kbps] PSNR [dB] Time [s] FBRA Call1 FBRA Call2 PSNR Call1 PSNR Call2
27. TESTBED Evaluation (2/2) 0 200 400 400 200 0 0 50 100 150 200 250 300 350 400 0 40 40 0 Throughput [kbps] PSNR [dB] Time [s] FBRA Call1 FBRA Call2 PSNR Call1 PSNR Call2 •  1Mbit/s link capacity, 100ms one-way-delay,
28. Next steps •  Code: (coming soon) –  https://github.com/protocols-comnet/rmcat-adaptive-fec-code •  Evaluation Paper: –  Nagy M., Singh V., Ott J., Eggert L., Congestion Control using FEC for Conversational Multimedia Communication, Proc. of ACM Multimedia Systems, Singapore, SG, Mar, 2014, •  More feedback is appreciated J