HotStreaming: A Cross-Layer Optimized P2P IPTV System

1. HotStreaming: A Cross-Layer Optimized P2P IPTV System Polly Huang 黃寶儀 Homer Chen 陳宏銘 Jason Yao 姚介新 GICE, NTU

2. Outline • CoolStreaming, PPLive, Peercast • What they are • Secret of success • HotStreaming • The Differences • The system • The evaluation • Current Status

3. CoolStreaming  PPLive • A peer-to-peer (P2P) overlay network for IPTV • Hongkong-China based • To an average user, it’s free live sports event broadcast service • 10,000s of subscribers

4. Secret of the CoolStreaming’s Success • Free content • Smooth and quality video • The more users, the better the quality

5. 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 Peer Why? Media Source 1 2 3 4 5 6 7 8 9 10 Peer Peer Peer

6. Major Technical Components

7. Outline • CoolStreaming, PPLive, Peercast • What they are • Secret of success • HotStreaming • The Differences • The system • The evaluation • Current Status

8. Differences

9. System Advantages

10. Three Components • Partnership formation • Multiple description coding • Video segment scheduling

11. Partner Peers HotStreaming Peer Video Buffer MDC Decoder Main Buffer Map Player Partnership Manager Scheduler System Architecture peer peer peer P2P Overlay Network peer peer

12. Consuming Peer Providing Peer 4 2 3 5 7 6 8 9 Partnership Formation: Idea • Every peer: • Periodically send • An advertisement message (ADV) 1 • Upon receiving the ADV: • Roll a dice • If win, select the ADV sender to provide video segments Partnership Relationship

13. Qualitative Comparison • SCAMP • Forwarding of ADV • To all partners • Selection of partner • Depending on current #partners • TYPHOON • Forwarding of ADV • To partners who have few partners • To avoid peers being isolated  better stability • Selection of partner • Depending on current #partners • Bounded by the max #partners  better load balance

14. Quantitative Comparison Stability Improvement ~ 20 fold

15. ... ... 0 2 4 6 ... ... 1 3 5 7 Adding Redundant Quarter - Sized Streams E f ... ... 0 2 4 6 1 3 5 7 O q O f ... ... 1 3 5 7 2 4 6 0 E q MDC: Idea MDC-STHI: MDC with Spatial-Temopral Hybrid Interpolation

16. Very Key to Mobile IPTV N 1 N 2 E + O O + E Ef: Full size even frame Eq:1/4 size even frame Of: Full size odd frame Oq: 1/4 size odd frame f q f q N 3 E + O + E + O E + O f f q q q q E + O f q N 6 N 4 N 5

17. PSNR: Stefan (CIF)

18. Segment Request Scheduling • From a peer’s partners • Available segment type (e.g. Ef, Oq) and size • The peer calculates • Maximize the video quality • Score of full-sized segments > quarter-sized ones • Under 2 constraints • Estimated time for segment to arrive < playback time • Total segments to request < estimated available bandwidth • A linear programming problem • Efficient approximation algorithm • Computation < 10s milliseconds

19. 2: {Ef(1), Of (1), …}, 3: {Of(1), Ef (1), …} 6: {Oq(2), Eq(2), …} Segment Request: Illustrated 2 1 3 6

20. Testbed Experiments • All three components integrated

21. Impact of TYPHOON

22. Impact of MDC-STHI+Opt Schedule

23. Summary: Our Strength

24. Outline • CoolStreaming, PPLive, Peercast • What they are • Secret of success • HotStreaming • The Differences • The system • The evaluation • Current Status

25. On-Going • Small-scale testing on Campus Network • 15-node scale testing • 1 server, 14 peer users • 1 Mbps video source for an hour • Network centric measurement • Scalability • Packet loss rate (arrival rate within deadline) • Control message overhead

26. Video Receiving Rate: per Peer

27. Video Sending Rate: Total Scalability (server load) MOD ~ stream_size * #user P2P ~ stream_size * in_degree  For non-live-programs, no user limit on HotStreaming!

28. Playback Continuity

29. Overhead

30. Questions? Polly Huang http://nslab.ee.ntu.edu.tw/