Measurement of the Congestion Responsiveness of RealPlayer Streaming Video Over UDP Jae Chung, Mark Claypool, Yali Zhu Pro...
The Hungry Wolf - Bandwidth Requirements for Video <ul><li>Streaming media growing </li></ul><ul><ul><li>350,000 hours of ...
The Wolf on the Prowl – Transport Protocols for Video <ul><li>Streaming video doesn’t like TCP </li></ul><ul><ul><li>Wants...
The Wolf Pack – Commercial Video <ul><li>Commercial products have major impact </li></ul><ul><li>Have been studies charact...
Specific Motivation
Outline <ul><li>Introduction  </li></ul><ul><li>Background  </li></ul><ul><li>Experiments   </li></ul><ul><li>Results   ...
RealVideo Network Characteristics <ul><li>SureStream </li></ul><ul><ul><li>Allows bandwidth scaling </li></ul></ul><ul><li...
Outline <ul><li>Introduction  </li></ul><ul><li>Background    </li></ul><ul><li>Experiments    </li></ul><ul><li>Result...
Methodology <ul><li>Select Real Video Clips </li></ul><ul><li>Construct environment for measuring congestion response </li...
RealVideo Clip Playlist <ul><li>If put in controlled environment, may not be representative of clips “in the wild” </li></...
Responsiveness Measurement Environment <ul><li>PIII 700 Mhz, 128 MB RAM, Linux 2.4 </li></ul><ul><li>RealPlayer version 8....
Outline <ul><li>Introduction  </li></ul><ul><li>Background    </li></ul><ul><li>Experiments        </li></ul><ul><li>Re...
Distribution of Loss
Distribution of Latency
Distribution of Packet Sizes
Outline <ul><li>Introduction  </li></ul><ul><li>Background    </li></ul><ul><li>Experiments        </li></ul><ul><li>Re...
RealPlayer = FairPlayer?
RealPlayer = FoulPlayer?
Bandwidth Distribution 600 Kbps 300 Kbps 150 Kbps 75 Kbps
Head-to-Head Bandwidth
Bandwidth Difference Distribution
Mostly TCP-Friendly! <ul><li>Remove low bandwidth (36%) </li></ul><ul><li>Then remove unscalable (14%) </li></ul>
Outline <ul><li>Introduction  </li></ul><ul><li>Background    </li></ul><ul><li>Experiments        </li></ul><ul><li>Re...
Media Scaling Distribution
Media Scales
Media Scaling Dynamics (1)
Media Scaling Dynamics (2)
Media Scaling Distributions
Media Scaling Adaptation Speed
Outline <ul><li>Introduction  </li></ul><ul><li>Background    </li></ul><ul><li>Experiments        </li></ul><ul><li>Re...
Buffering Rate to Playout Rate
Buffering Rate to Playout Rate Distribution
Smoothness
Smoothness at each Bottleneck
Discussion of Results <ul><li>No concrete incentives to respond to congestion </li></ul><ul><ul><li>In fact, may be “rewar...
Future Work?
Future Work <ul><li>Other commercial Players </li></ul><ul><ul><li>Microsoft Media Player </li></ul></ul><ul><li>Live clip...
Upcoming SlideShare
Loading in …5
×

ppt

263
-1

Published on

0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total Views
263
On Slideshare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
3
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide
  • Server-Player protocol RTSP for control UDP or TCP for data Progressive Networks Architecture (PNA) for backward compatibility HTTP if going through a firewall Buffering (also called preroll) for smoothness in the face of bandwidth changes SureStream different bandwidth encodings of the same clip can switch on fly Scalable video technology Data and frame rate controlled by client speed Audio data gets priority. If bwdith left, then video. Repair Forward Error Correction Interleaving (at least for audio)
  • ppt

    1. 1. Measurement of the Congestion Responsiveness of RealPlayer Streaming Video Over UDP Jae Chung, Mark Claypool, Yali Zhu Proceedings of the International Workshop on Packet Video (PV) Nantes, France April 2003 http://www.cs.wpi.edu/~claypool/papers/h2h/
    2. 2. The Hungry Wolf - Bandwidth Requirements for Video <ul><li>Streaming media growing </li></ul><ul><ul><li>350,000 hours of online video broadcast ’01 </li></ul></ul><ul><li>Voice is 32-64 Kbps, but video has range of data rates </li></ul><ul><ul><li>Videoconference 0.1 Mbps (H.261, MPEG-4) </li></ul></ul><ul><ul><li>VCR 1.2 Mbps (MPEG-1) </li></ul></ul><ul><ul><li>Broadcast quality 2-4 Mbps (MPEG-2) </li></ul></ul><ul><ul><li>HDTV quality 25-34 Mbps (MPEG-2) </li></ul></ul><ul><ul><li> Thus, potential for more than network capacity </li></ul></ul>
    3. 3. The Wolf on the Prowl – Transport Protocols for Video <ul><li>Streaming video doesn’t like TCP </li></ul><ul><ul><li>Wants rate-based not window-based </li></ul></ul><ul><ul><li>Can tolerate some loss </li></ul></ul><ul><ul><li>AIMD causes rate fluctuations </li></ul></ul><ul><ul><li> So, use UDP where application controls </li></ul></ul><ul><li>But UDP has no congestion control </li></ul><ul><ul><li>Unfair, unfriendly, and even collapse! </li></ul></ul><ul><li>Approaches to have router catch </li></ul><ul><ul><li>Model video as CBR “firehose” (is it?) </li></ul></ul>
    4. 4. The Wolf Pack – Commercial Video <ul><li>Commercial products have major impact </li></ul><ul><li>Have been studies characterizing commercial traffic </li></ul><ul><ul><li>Bandwidth use, frame rate, user use… </li></ul></ul><ul><li>But no work measuring responsiveness, or lack of it, of commercial video products </li></ul>
    5. 5. Specific Motivation
    6. 6. Outline <ul><li>Introduction  </li></ul><ul><li>Background  </li></ul><ul><li>Experiments </li></ul><ul><li>Results </li></ul><ul><li>Analysis </li></ul><ul><li>Conclusions </li></ul>
    7. 7. RealVideo Network Characteristics <ul><li>SureStream </li></ul><ul><ul><li>Allows bandwidth scaling </li></ul></ul><ul><li>Buffering </li></ul><ul><ul><li>Remove jitter </li></ul></ul>Server RTSP <ul><li>Data: TCP or UDP </li></ul><ul><li>Choice, unclear </li></ul><ul><li>We’ll force </li></ul>
    8. 8. Outline <ul><li>Introduction  </li></ul><ul><li>Background  </li></ul><ul><li>Experiments  </li></ul><ul><li>Results </li></ul><ul><li>Analysis </li></ul><ul><li>Conclusions </li></ul>
    9. 9. Methodology <ul><li>Select Real Video Clips </li></ul><ul><li>Construct environment for measuring congestion response </li></ul><ul><li>Construct environment to measure ability to scale (SureStream) </li></ul><ul><li>Iteratively plan clips varying network </li></ul>
    10. 10. RealVideo Clip Playlist <ul><li>If put in controlled environment, may not be representative of clips “in the wild” </li></ul><ul><li>Select large number through search engines (Yahoo, Google …) </li></ul><ul><li>Randomly choose 100 (79 analyzed) </li></ul><ul><li>Geography results: </li></ul><ul><ul><li>76 US, 9 Canada, 8 UK, 6 Italy, 1 Germany </li></ul></ul><ul><ul><li>North American dominance likely reflected in typical user locality of reference </li></ul></ul><ul><li>Length results: </li></ul><ul><ul><li>Median 3 minutes, min 20 seconds, max 30 minutes </li></ul></ul>
    11. 11. Responsiveness Measurement Environment <ul><li>PIII 700 Mhz, 128 MB RAM, Linux 2.4 </li></ul><ul><li>RealPlayer version 8.0.3 </li></ul><ul><li>Sniffing via tcpdump </li></ul><ul><li>Loss and round-trip time via ping </li></ul><ul><li>TBF to limit bandwidth </li></ul><ul><li>2 Measurements for each clip </li></ul><ul><li>(Note, RealTracer for MediaScaling) </li></ul>Internet Client Hub Router 10Mbps DSL 700Kbps Client Token Bucket Filter UDP TCP Server
    12. 12. Outline <ul><li>Introduction  </li></ul><ul><li>Background  </li></ul><ul><li>Experiments  </li></ul><ul><li>Results  </li></ul><ul><li>Analysis </li></ul><ul><li>Conclusions </li></ul>
    13. 13. Distribution of Loss
    14. 14. Distribution of Latency
    15. 15. Distribution of Packet Sizes
    16. 16. Outline <ul><li>Introduction  </li></ul><ul><li>Background  </li></ul><ul><li>Experiments  </li></ul><ul><li>Results  </li></ul><ul><li>Analysis </li></ul><ul><ul><li>Head to Head  </li></ul></ul><ul><ul><li>Bandwidth </li></ul></ul><ul><ul><li>Scaling </li></ul></ul><ul><ul><li>Buffering </li></ul></ul><ul><ul><li>Smooth </li></ul></ul><ul><li>Conclusions </li></ul>
    17. 17. RealPlayer = FairPlayer?
    18. 18. RealPlayer = FoulPlayer?
    19. 19. Bandwidth Distribution 600 Kbps 300 Kbps 150 Kbps 75 Kbps
    20. 20. Head-to-Head Bandwidth
    21. 21. Bandwidth Difference Distribution
    22. 22. Mostly TCP-Friendly! <ul><li>Remove low bandwidth (36%) </li></ul><ul><li>Then remove unscalable (14%) </li></ul>
    23. 23. Outline <ul><li>Introduction  </li></ul><ul><li>Background  </li></ul><ul><li>Experiments  </li></ul><ul><li>Results  </li></ul><ul><li>Analysis </li></ul><ul><ul><li>Head to Head  </li></ul></ul><ul><ul><li>Bandwidth  </li></ul></ul><ul><ul><li>Scaling  </li></ul></ul><ul><ul><li>Buffering </li></ul></ul><ul><ul><li>Smooth </li></ul></ul><ul><li>Conclusions </li></ul>
    24. 24. Media Scaling Distribution
    25. 25. Media Scales
    26. 26. Media Scaling Dynamics (1)
    27. 27. Media Scaling Dynamics (2)
    28. 28. Media Scaling Distributions
    29. 29. Media Scaling Adaptation Speed
    30. 30. Outline <ul><li>Introduction  </li></ul><ul><li>Background  </li></ul><ul><li>Experiments  </li></ul><ul><li>Results  </li></ul><ul><li>Analysis </li></ul><ul><ul><li>Head to Head  </li></ul></ul><ul><ul><li>Bandwidth  </li></ul></ul><ul><ul><li>Scaling  </li></ul></ul><ul><ul><li>Buffering  </li></ul></ul><ul><ul><li>Smooth </li></ul></ul><ul><li>Conclusions </li></ul>
    31. 31. Buffering Rate to Playout Rate
    32. 32. Buffering Rate to Playout Rate Distribution
    33. 33. Smoothness
    34. 34. Smoothness at each Bottleneck
    35. 35. Discussion of Results <ul><li>No concrete incentives to respond to congestion </li></ul><ul><ul><li>In fact, may be “rewarded” for not </li></ul></ul><ul><li>However, RealVideo clearly responds </li></ul><ul><ul><li>Often TCP-Friendly </li></ul></ul><ul><li>Content providers need to provide chance for scalability </li></ul><ul><ul><li>Scaling levels </li></ul></ul><ul><li>But buffering at higher rate is bad for network </li></ul><ul><li>TCP can be as smooth as UDP </li></ul><ul><li>So why not TCP?  API is limiting </li></ul>
    36. 36. Future Work?
    37. 37. Future Work <ul><li>Other commercial Players </li></ul><ul><ul><li>Microsoft Media Player </li></ul></ul><ul><li>Live clips (versus pre-recorded clips) </li></ul><ul><li>Perceptual quality of video over TCP versus UDP </li></ul><ul><li>Characterization of clips on Internet </li></ul><ul><ul><li>So can examine “typical” clips </li></ul></ul>
    1. A particular slide catching your eye?

      Clipping is a handy way to collect important slides you want to go back to later.

    ×