1. A scalable technique for VCR-like interactions in video-on-demand applications Tantaoui, M.A.; Hua, K.A.; Sheu, S.; IEEE Proceeding of the 22nd International Conference on Distributed Computing Systems Workshops, 2002. Hsin-Hua, Lee

2. Outline  Introduction  Broadcast Based Interaction Technique (BIT)  Simulation Results  Conclusion

3. Introduction(1/2)  Existing techniques solved only partially the problem of VCR interactivity in a broadcast framework. 1. Deliver the emergency stream to some dedicated client.  Limited to small-scale deployment. 456 3 2 891011 … 47 12 … 1 10 Prefetch buffer Emergency stream Existing streams 10 jump

4. Introduction(2/2) 2. ABM (Active Buffer Management): Maintain the play point at the middle of the video segment currently cached in the prefetch buffer.  Prefetching rate cannot keep up with typical fast-forward speeds.

5. BIT Channel Design  Two versions of the same video: a normal version and a version compressed by a compression factor, f.  The server bandwidth is divided into K logical channels. K=K r +K i, and K i =K r /f. K r : number of the regular channels, K i : number of interactive channels.

6. BIT Data fragmentation  Regular channels --CCA (Client Centric Approach)  Interactive channels 1234 5678 1 2 f=4 ViVi Si’Si’

7. BIT Broadcasting Scheme

8. Implementing VCR functions with BIT  Two buffers : normal buffer holds the normal video interactive buffer caches the compressed video. To ensure good interactive service, the size of interactive buffer is set twice of the normal one.  C+2 loaders : c normal loaders and 2 interactive loaders.

9. Implementing VCR functions with BIT  To playback and provide VCR functions, we identify two components that work together: Player: rendering the frames and accepting interactions. Loader: tuning to the appropriate channels and downloading the segments.

10. BIT Player  Play action player renders the current play point of the normal buffer, and the play point of the interactive buffer moves accordingly. 21222324 9105678 Normal buffer Interactive buffer 34

11. BIT Player  Fast-Forward/Fast-Reverse/Pause actions Interactive buffer not exhausted  player switches to the normal mode,  loaders downloads the appropriate segments such that the normal playback is resumed at a frame closest possible to the destination point. 3738 9105678 Normal buffer Interactive buffer 34

12. BIT Player Interactive buffer exhausted  player forces the user to resume the normal play by setting the destination point to the newest frame of the interactive buffer in case of a Fast- Forward action, or the oldest frame in case of a Fast-reverse or pause actions.

13. BIT Player  Jump Forward/Jump Backward actions Destination in normal buffer  Player simply moves the play point to the destination point, and continues the normal play. Destination not in normal buffer  Player allocates the loaders to start downloading the appropriate segments, and continues the normal play from the closest point.

14. BIT Loader  If the current segment is in the first half of its interactive group group j, the two interactive loaders are allocated to the previous and to the current interactive groups, i.e. group j-1 and group j.  Otherwise, in the second half, group j-1 and group j.  By prefetching the compressed segments, we make sure that the play point of the interactive buffer is always in the middle.

15. Performance Study  User Interaction Model P i ’ s : the frequency of interactions; m i ’ s: the average VCR interaction duration, interactive mean.

16. Simulation Assumption  Durations for a play and for interactive actions are exponentially distributed with their respective mean, m i.  m ff = m fr = m pause = m jf = m jb  P i =P p =0.5, and P pause = P ff = P fb =P jf = P jb = P i /5=0.1.  We define duration ratio d r = m i /m p. This ratio measures the degree of interaction.

17. Simulation results  Effect of the duration ratio c=3, 2 hr video, f=4, regular buffer size= 5mins, interactive buffer size=10mins, K=40(K r =32, K i =8)

18. Simulation results  Effect of the buffer size

19. Simulation results  Effect of Compression factor

20. Conclusion  BIT offers better interaction quality by repeatedly broadcasting the interactive (compressed) version of the video.  By prefetching data from this version and caching them in the local buffer, the client is able to support longer-duration interactive actions.  In terms of scalability, since the clients can share the interactive broadcasts, the bandwidth requirement of BIT is independent of the number of users. This strategy is consistent with the broadcast paradigm.