Rate-control for Conversational Video Communication in Heterogeneous Networks

Rate-control for Conversational Video
Communication in Heterogeneous Networks

Transcript

1. Introduction Performance Analysis Conclusions Rate-control for Conversational Video Communication in Heterogeneous Networks VidEv 2012 Varun Singh, J¨ org Ott, Igor Curcio Comnet, Aalto University, {varun,jo}@comnet.tkk.fi Nokia Research Center, igor.curcio@nokia.com June 25, 2012 1 / 14
2. Introduction Performance Analysis Conclusions Quick introduction to Real-time Transport Protocol Is widely used for telephony, video conferencing, and telepresence applications Often used over best-effort UDP/IP networks RTP provides playout timing and packet sequencing Reception quality feedback every few seconds (RTCP) RTCP provides higher-level summary feedback instead of per-packet feedback 2 / 14
3. Introduction Performance Analysis Conclusions System Blocks for Video Communication 3 / 14
4. Introduction Performance Analysis Conclusions Key features of Video Communication Unlike video streaming, video communication has less opportunity to buffer packets/frames Delay budget is of 200-400 ms. Upper-bound 400ms recommended by 3GPP Video emerging in the Web-browser (WebRTC), FaceTime and Skype for Mobile 4 / 14
5. Introduction Performance Analysis Conclusions Real-world Mobile environment is different from fixed Internet (cable, xDSL, Ethernet) 5 / 14
6. Introduction Performance Analysis Conclusions Metrics for Rate Adaptation 6 / 14
7. Introduction Performance Analysis Conclusions Indicators Summarized Playout Delay Loss Rate Round Trip Time (RTT) Discard Rate Frame Inter- arrival time -  Long term cue -  Fairness -  Late in indicating -  Over-use -  Early sign of congestion -  Undershoot Variation indicates -  Under-utilization -  Over-use -  Tolerant to short variations -  Early sign of congestion -  Sensitive to competition 7 / 14
8. Introduction Performance Analysis Conclusions Simulator System Overview ns-2 H.264   Encoder   H.264   Decoder   REAL  Interface   REAL  Interface   APP   APP   Network   Topology   RTP  Agent   RTP  Agent   8 / 14
9. Introduction Performance Analysis Conclusions Scenario setup in ns2 9 / 14
10. Introduction Performance Analysis Conclusions Five calls in a Heterogeneous Environment 0 250 500 750 1000 0 20 40 60 80 100 120 140 160 180 0 250 500 750 1000 0 20 40 60 80 100 120 140 160 180 bit rate (kbps) time (in sec) time (in sec) Call 1 Call 2 0 250 500 750 1000 0 20 40 60 80 100 120 140 160 180 0 250 500 750 1000 0 20 40 60 80 100 120 140 160 180 time (in sec) time (in sec) Call 1 Call 3 0 250 500 750 1000 0 20 40 60 80 100 120 140 160 180 0 250 500 750 1000 0 20 40 60 80 100 120 140 160 180 time (in sec) time (in sec) Call 1 Call 4 0 250 500 750 1000 0 20 40 60 80 100 120 140 160 180 0 250 500 750 1000 0 20 40 60 80 100 120 140 160 180 bit rate (kbps) time (in sec) time (in sec) Call 1 Call 5 10 / 14
11. Introduction Performance Analysis Conclusions Performance Metrics Goodputavg Loss Rate PSNRavg ABU (kbps) (%) (dB) (%) Call 1 140.10 2.15% 31.4 (σ = 0.39) 70.1% Call 2 133.55 1.61% 31.9 (σ = 0.62) 66.8% Call 3 35.18 1.55% 22.2 (σ = 1.13) 17.59% Call 4 114.96 2.75% 31.1 (σ = 0.75) 57.5% Call 5 130.23 2.25% 31.3 (σ = 0.13) 65.1% Table: Scenario: Five calls in a heterogeneous network 11 / 14
12. Introduction Performance Analysis Conclusions Load on the Bottleneck link 0 0.2 0.4 0.6 0.8 1 1.2 0 20 40 60 80 100 120 140 160 180 bit rate (Mbps) time (in sec) Load on the bottleneck link 12 / 14
13. Introduction Performance Analysis Conclusions Conclusions We show that using Playout Delay and Frame Inter-arrival time can be used for congestion control against competing traffic. Contributions to IETF: RTP Congestion Control: Circuit Breakers for Unicast Sessionsa. RTCP Extension Report for Discard RLE Packetsb. Future Work: More complex scenarios need to be simulated. Deploy in real-world, for example in WebRTC. ahttp://tools.ietf.org/html/draft-perkins-avtcore-rtp-circuit-breakers bhttp://tools.ietf.org/html/draft-ietf-xrblock-rtcp-xr-discard-rle-metrics 13 / 14
14. Introduction Performance Analysis Conclusions Five calls in a Wired Environment 14 / 14