Empirical Evaluation of the Congestion Responsiveness of RealPlayer Video

1. Empirical Evaluation of the Congestion Responsiveness of RealPlayer Video Jae Chung and Mark Claypool Kluwer Multimedia Tools and Applications (Accepted for Publication, 2005) http://www.cs.wpi.edu/~claypool/papers/h2h-journal/

2. The Hungry Wolf – Bandwidth Requirements for Video • Voice uses 32-64 Kbps, but video has broader range of data rates • Videoconference 0.1 Mbps (H.261, MPEG-4) • VCR 1.2 Mbps (MPEG-1) • Broadcast quality 2-4 Mbps (MPEG-2) • HDTV quality 25-34 Mbps (MPEG-2)  Thus, potential for more than network capacity

3. The Wolf on the Prowl – Transport Protocols for Video • Streaming video doesn’t always like TCP • Wants rate-based not window-based • Can tolerate some loss • AIMD causes rate fluctuations So, uses UDP where application controls • But UDP has no built-in congestion control • Unfair, unfriendly, and even collapse!

4. Caging the Wolf –Active Queue Management • Identify unresponsive flows in network • Punish those that are not TCP-Friendly • Router must monitor flows, determine characteristics • Often model video as CBR “firehose” (is it?) • A better understanding of video traffic  improve such AQM techniques

5. The Wolf Pack – Commercial Video • Commercial products have major impact Real about ½ market share [13], [LCKN05] • Have been studies characterizing commercial traffic • Bandwidth use, frame rate, user use… • But no work measuring responsiveness, or lack of it, of commercial video products • Measure responsiveness of RealPlayer over UDP

6. Outline • Introduction  • Background  • Experiments • Analysis • Conclusions

7. RTSP Server • Data: TCP or UDP • Choice automatic • (about ½ and ½ in practice) • We force RealVideo Network Characteristics • SureStream (is it effective?) • Allows bandwidth scaling • Buffer (how fast is it filled?) • Smoother playout

8. Outline • Introduction  • Background  • Experiments • Analysis • Conclusions

9. Approach • Select RealVideo URLs • Construct measurement environment • Play RealVideo clips with different bottlenecks • Analyze results • Disseminate

10. RealVideo Clip Playlist • Choose videos actually on the Internet • As opposed to a controlled server at WPI • Search through well-known search engines • Google and Yahoo • Select the first 100 valid RTSP URLs returned • Of the 100, only 79 valid until end of study

11. RealVideo Clip Duration Median about 3 minutes (min 20 sec, max 30 min) Fits duration found in [LCKN05]

12. UDP Client Router Server Hub Internet TCP DSL Client 700Kbps 10Mbps Token Bucket Filter Responsiveness Measurement Environment tcptrace • Encoding rates RealTracker • Data rates via tcpdump • Loss rates and round-trip times via ping • Token Bucket Filter to limit bandwidth • 75, 150, 300, 600 kbps • 79 (of 100) Clips, 2 measurements per clip

13. Outline • Introduction  • Background  • Experiments • Analysis  • Aggregate • Bandwidth • TCP-Friendly • Media Scaling • Buffering Rate • Conclusions

14. Aggregate Analysis

15. Sample Results 1 Pronounced buffering region Relatively fair bandwidth sharing after

16. Sample Results 2 Unfair to TCP, in general Sometimes, TCP more than UDP! Buffering less pronounced

17. Outline • Introduction  • Background  • Experiments • Analysis  • Aggregate • Bandwidth • TCP-Friendly • Media Scaling • Buffering Rate • Conclusions

18. Head-to-Head Bandwidth

19. CDF of Head-to-Head Bandwidth Less Fair under bandwidth constraints But not increasingly so 20% of the time, TCP more than UDP!

20. Fairness (Weak correlation)

21. Outline • Introduction  • Background  • Experiments • Analysis  • Aggregate • Bandwidth • TCP-Friendly • Media Scaling • Buffering Rate • Conclusions

22. A Sheep in Wolf’s Clothing? [PFTK98] • Remove low bandwidth (36%) • Then remove unscalable (14%)

23. Outline • Introduction  • Background  • Experiments • Analysis  • Aggregate • Bandwidth • TCP-Friendly • Media Scaling • Buffering Rate • Conclusions

24. Media Scaling Distribution Some cannot even scale Maybe these are low bwidth?

25. Media Scaling per Clip Some high bwidth clips have no levels Fewer levels at higher bwidth

26. Media Scaling Dynamics 1 Takes longer for TCP

27. Media Scaling Adaptation Speed Takes longer for TCP  more difficult to figure out network conditions

28. Media Scaling Dynamics 2 TCP has fewer level changes

29. Media Scaling Level Changes TCP cannot adapt as finely

30. Outline • Introduction  • Background  • Experiments • Analysis  • Aggregate • Bandwidth • TCP-Friendly • Media Scaling • Buffering Rate • Conclusions

31. Buffering Rate to Playout Rate Both are bursty!

32. Is TCP Smooth Enough? When no congestion Slope comparable for some regions

33. Summary • No concrete incentives to respond to congestion • In fact, may be “rewarded” for not • However, RealVideo clearly adjusts • Often TCP-Friendly • Content providers need to provide chance for scalability • Scaling levels • Buffering at higher rate is bad for network since bursty • Why not TCP?  API is limiting

34. Future Work?

35. Future Work • Other commercial Players • Microsoft Media Player • Perceptual quality of video over TCP versus UDP • Characterization of clips on Internet • So can examine “typical” clips