1. Receiver Capability Heterogeneity in the Internet

2. Agenda Introduction Some proposed approaches Performance comparison Summary Discussion

3. Introduction Evolution of VoD systems  Video rental  Video over Internet  Unicast  Multicast 3 data streams1 data stream 4 data streams1 data stream VS

4. Introduction Problem created by receiver capability heterogeneity 3 Mbps 1 Mbps

5. Introduction Trivial solutions, we either  1. leave video stream rate at 3 Mbps  Unable to provide real-time streaming  2. reduce video stream rate to 1 Mbps  Video quality degradation

6. Other Approaches Replicated stream approach Layering approaches  Cumulative layering approach  Non-cumulative layering approach

7. Replicated Stream Approach Aggregate server bandwidth: 4 Mbps Sender Group of clients (C A ) – downlink: 1 Mbps Group of clients (C B ) – downlink: 3 Mbps Full quality stream (3 Mbps) Low quality stream (1 Mbps)

8. Layering Approaches Cumulative layering  Base layer + enhancement layers  Cumulative decoding  E.g. MPEG-2 and H.263 standards Spatial scalability, temporal scalability, data partitioning and SNR scalability Non-cumulative layering  Independently decodable video layers  E.g. Multiple Description Coding (MDC)

9. Layering Approaches Aggregate server bandwidth: 3 Mbps Enhancement layer (2 Mbps) Base layer (1 Mbps) Sender Group of clients (C A ) – downlink: 1 Mbps Group of clients (C B ) – downlink: 3 Mbps

10. Comparison between the Two Approaches Common argument:  Stream replication wastes server bandwidth by stream duplication However, no quantitative and systematic comparison has been given

11. Some Counterarguments Kim and Ammar [1] take into account of  Layering overhead  Protocol complexity for fair comparison [1] T. Kim, M. H. Ammar, "A comparison of layering and stream replication video multicast schemes", Proc. NOSSDAV‘ 01, Port Jefferson, NY, June 25-26, 2001.

12. Layering Overhead Information theory states: For the same source and same distortion, (1)layered encoding requires at least as much data rate as a non-layered encoding (2)equality requires a strict Markov condition to apply to the source

13. Layering Overhead Protocol and packetization overhead  Source of overhead: start codes, GOP information, picture header, macroblock header etc.  More severe at low data rates  According to literature, overhead can be as much as 20% ~ 30%

14. Layering Overhead - Example Aggregate server bandwidth: 3 Mbps Take into account the overhead (e.g. 20%), data rate contributing to video data:  C A  0.83 Mbps  C B  2.5 Mbps Enhancement layer (2 Mbps) Base layer (1 Mbps) Sender Group of clients (C A ) – downlink: 1 Mbps Group of clients (C B ) – downlink: 3 Mbps

15. Video Quality Degradation Layered (2 layers with different quantizer scales) vs non-layered

16. Protocol Complexity In layering protocols, number of channel subscriptions >= 1, which incurs  More join / leave group messages  Better synchronization capability

17. Summary Three basic approaches to Internet heterogeneity problem Superiority not always goes to layered multicast protocol

18. Discussion Possible applications in multicast VoD systems  Fast-forward (FF) VCR operations  Normal playback resumption after VCR operations

19. Q & A Thank you