1 / 27

Experience with Control Mechanisms for Packet Video in the Internet

Experience with Control Mechanisms for Packet Video in the Internet. J-C. Bolot and T. Turletti. Presented by Michael Wong Course 18-845, 4/25/00. Outline. Introduction Rate control mechanisms Loss control mechanisms Limitations and outlook Conclusion. Conventional Video.

Télécharger la présentation

Experience with Control Mechanisms for Packet Video in the Internet

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Experience with Control Mechanisms for Packet Video in the Internet J-C. Bolot and T. Turletti Presented by Michael Wong Course 18-845, 4/25/00

  2. Outline • Introduction • Rate control mechanisms • Loss control mechanisms • Limitations and outlook • Conclusion

  3. Conventional Video • Rate of video sequence vary rapidly with time • Constant capacity channel • Solution: Vary video quality according to scene complexity so as to keep output rate approx. constant • Rate control based on buffer feedback

  4. Best effort service • Offer channel with characteristics: • Available bandwidth, end-to-end delay and loss • Dependent on other connections in the network • Difficult to determine a priori

  5. Approach 1 • Provide performance guarantees at network level • Requires specific resource allocation and/or reservation mechanisms • Not widely deployed

  6. Approach 2 • Adapt application behavior to network characteristics • Also applicable to network with explicit resource allocation and reservation

  7. Focus • End-to-end delay requires elaborate queuing disciplines • Delay control with adaptive playout algorithms is a well understood problem • Adapt to available bandwidth and loss • Rate control mechanisms • Error control mechanisms

  8. Rate control mechanisms • Examples of source-based control: • TCP – Adjust window size based on network congestion • CATV – Local buffer capacity • WAN – Feedback information on state of network

  9. Difficulties • Characteristics of channel is time varying • Variations dependent on other traffic sources • The “channel” is a heterogeneous multicast tree

  10. Rate adjustment • Parameters for block-based transform video encoding • Frame sampling rate (PQ mode) • Quantization levels (PR mode) • Movement detection threshold (PR mode) • User specifies mode and max_rate

  11. Packets sent using Real-Time Transport Protocol (RTP) Feedback carried by RTCP packets as Receiver Reports (RR) 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |V=2|P|X| CC |M| PT | sequence number | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | timestamp | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | synchronization source (SSRC) identifier | +=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+ | contributing source (CSRC) identifiers | | .... | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Feedback collection

  12. Rate control algorithm • Maximum output rate is controlled by linear increase/multiplicative decrease algorithm: If Congestion max_rate := max(max_rate/GAIN, MIN_RATE) Else if NoCongestion max_rate := min(max_rate + INC, MAX_RATE)

  13. Observations • Algorithm can be used in PQ or PR mode • Rate control vs. window control in TCP • Rate adjustments do not match window adjustments in TCP • No good values of Congestion, NoCongestion, GAIN, and INC.

  14. Rate control (cont’d) • Analytic models of TCP, bandwidth equ of a TCP connection [23], • MTU = maximum packet size • rtt = mean round trip time • p = mean packet loss rate

  15. Limitations • Feedback explosion • Probabilistic sampling scheme in IVS • Scaled feedback scheme of RTCP • Heterogeneity problem • Convert loss rates into a global loss rate • Average or highest loss rate

  16. Experimental evaluation Unicast over INRIA Slow CPU

  17. Experimental Evaluation (cont’d) One Source Three Sources

  18. Experimental Evaluation (cont’d)

  19. Loss control mechanisms • Lost packets degrade video quality • Intracoded frames sent at regular intervals • Approaches to loss control: • Reduce the time between intra-coded blocks of image • Intra-code and transmit only blocks that change the most • Use intra- and inter-frame coding (IVS)

  20. Loss control (cont’d) • Destination recovery • Recover at destination using loss concealment techniques (i.e. spatial and temporal interpolation) • Forward Error Correction (FEC)-based error control mechanisms

  21. Loss control using FEC • Redundant data about packet n – k in packet n is made up of macroblocks (MB) • MBs are stored at lower definition • Feedback mechanism is used to control the amount of redundant information sent by source

  22. FEC loss rate

  23. Experimental Evaluation

  24. Limitations and Outlook • Source-based control mechanisms • Heterogeneity of network exhibit widely different characteristics • Either low capacity participants are overwhelmed or • High capacity participants receive low quality video

  25. Solutions • Video gateways • Layered coding • Encode video with a layered or hierarchical coding scheme • Receivers join one or more multicast groups suited to network conditions • “Shared learning” to prevent load explosion

  26. Solutions (cont’d) • Replace join experiments by an explicit estimation at each receiver of equ • Find the largest integer L s.t. • Advantages • TCP-friendly • Does not rely on active probing • Does not require shared learning

  27. Conclusion • Prevents real-time UDP apps from swamping Internet and TCP • Rate adjustment algorithm and rate parameters unclear • Scalability issues with heterogeneous multicast groups • No minimum quality guarantees

More Related