1. 1 Layer-Encoded Video in Scalable Adaptive Streaming Michael Zink, Jens Schmitt, and Ralf Steinmetz, Fellow, IEEE IEEE TRANSACTIONS ON MULTIMEDIA, VOL. 7, FEBRUARY 2005

2. 2 Outline What is scalable adaptive streaming? Maximize the quality of the cached video Average PSNR is not a good assessment Subjective impression of variations in layer- encoded video Retransmission scheduling Simulation results Conclusion

3. 3 What is scalable adaptive streaming Main goal: Provide a true video-on-demand (TVoD) system Issues: Quality Adaptation (congestion control) Layer-encoded technique Scalability (performance) Proxy caching

4. 4 Layer-encoded adaptation If the network is congested, Use end-to-end congestion control mechanisms Allow dropping segments of the video Advantages: Prevent unfairness against TCP-based traffic Increase the network utilization

5. 5 File Tradition method v.s The method in this paper Tradition method Origin Server First client Second client Proxy Cache File The method in this paper

6. 6 Proxy caching (1/4) use write-through caching Case 1:forwarded through the proxy cache First clientSecond client Proxy Cache Origin Server Cache replacement strategy decided unicast : request path : transmit path

7. 7 Proxy caching (2/4) Case 2:streamed via a multicast group which clients and proxy caches join Origin Server Proxy Cache First clientSecond client Cache replacement strategy decided multicast unicast : request path : transmit path

8. 8 Proxy caching (3/4) A proxy cache may cache : a layer-encoded video after its initial transmission, or the missing segments (from different layers) being retransmitted.

9. 9 Proxy caching (4/4) challenges : What is a “ high quality ” video? Need an appropriate assessment When to schedule the retransmissions? The time to perform retransmission How to schedule the retransmissions? Choose the segments to be retransmitted. The retransmission of missing segments is a complex problem. Devise a number of heuristic algorithms.

10. 10 How to assess the quality of video Use a subjective assessment The experiment was performed with 94 test candidates (62 males and 32 females), between the age of 14 and 64. Video: Consists of 4 layers The size of the n th layer: s n = 2s n-1

11. 11 Subjective assessment v.s PSNR (1/2) Six variations in layer-encoded video Farm1 – Amplitude: Decrease slightly has better quality Farm2 – Frequency: Lesser layer changes has higher quality

12. 12 Subjective assessment v.s PSNR (1/2) M&C1 – Closing the Gap: Close a gap in a lower layer has higher quality M&C2 – Constancy: No layer changes has better quality

13. 13 Subjective assessment v.s PSNR (2/2) M&C3 – Constancy at a Higher Level: Tennis1 – All is Well That Ends Well: Increasing the amount of layers in the end leads to a higher quality

14. 14 Compare Subject and PSNR quality Subject assessment is more precise than PSNR. s(v) represents the same result as subject assessment. Minimize the frequency and amplitude will have the better quality (reducing the layer variations).

15. 15 When to schedule the retransmissions Retransmission time: Directly after the initial streaming process During subsequent requests During requests for different content from the server

16. 16 How to schedule the retransmissions Scheduling goals: Decreasing quality variations for a cached video is important. Minimize both frequency of variations and amplitude of variations. s(v): spectrum frames layers

17. 17 Algorithms for retransmission scheduling (1/3) Optimal retransmission scheduling is a discrete nonlinear stochastic optimization problem (NP-complete) Heuristics for retransmission scheduling Unrestricted priority-based heuristics Send periodic retransmission requests to the server to ensure that the server obtains an up-to- date schedule of retransmissions.

18. 18 Optimal retransmission scheduling model

19. 19 Algorithms for retransmission scheduling (2/3) Three heuristics of unrestricted priority- based retransmission scheduling algorithm: Case1 : Unrestricted lowest layer first (U-LLF)

20. 20 Algorithms for retransmission scheduling (3/3) Case2 : Unrestricted shortest gap lowest layer first (U-SG-LLF) Case3 : Unrestricted lowest layer shortest gap first (U-LL-SGF)

21. 21 Simulation results (1/2)

22. 22 Simulation results (2/2)

23. 23 Conclusion Promise a scalable TCP-friendly TVoD system Scalable adaptive system Use both caching and layer-encoded Develop the retransmission algorithms to retransmit missing segments Minimize the frequency and amplitude of the variations of the cached video