1. An adaptive video multicast scheme for varying workloads Kien A.Hua, JungHwan Oh, Khanh Vu Multimedia Systems, Springer-Verlag 2002

2. Outline Introduction Related Work Proposed Approach Performance Model Performance Conclusion

3. Introduction Maximize the efficiency of server resource with –Periodic broadcastPeriodic broadcast –Scheduled Multicast (batching)Scheduled Multicast –Hybrid Design Will show that existing scheduled multicast techniques are not suited for hybrid designs

4. Related Work Skyscraper Broadcasting Scheme Fragmentation recursive function –Series [1,2,2,5,5,12,12,25,25,52,52,…]

6. Related Work Skyscraper Broadcasting Scheme Apply aforementioned series function => segment size = 10.4 !! 22-mins video 1.5 Mbits/sec 5 min buffer 0.5 min latency

7. Related Work Scheduled Multicast Differ primarily in the criterion used to select which batch will receive service –First come, first served (FCFS) –Maximum queue length first (MQL) –Maximum factored queue length first (MFQ)

8. Related Work Scheduled Multicast FCFS MQL timeVideo 1 timeVideo 2 time Video 2 Video 1 Fair, maybe bad throughput Better throughput, maybe not fair

9. Related Work Maximum factored queue length first (MFQ) Applying a discriminatory weighting factor to the length of the queue : video i length, : the request frequency of video i Schedule video with the largest value of

10. Related Work Maximum factored queue length first (MFQ) d Still not fair, because not average waiting time !!

11. Proposed Approach Adaptive Hybrid Approach (AHA) –With a novel scheduled multicast -- “Largest aggregated waiting time first scheme” (LAW) –And SB (skyscraper broadcast)

12. LAW Compare with MQL, it take account of the distribution of the request –With considering “aggregated waiting time” : the total number of pending requests for video i : the arrival time of the jth request for video i c : the current time

13. LAW S1=128*5-(107+111+115+121+126)=60 S2=128*4-(112+119+122+127)=32 S1=128*5-(107+111+115+121+126)=60 Compute the sum of video i service latency

14. Adaptive hybrid approach With following procedures to decide which videos to broadcast

15. Adaptive hybrid approach

16. Performance model Compare AHA with MFQ-SB-n Performance metrics –Defection rate –Unfairness –Average service latency –Throughput 100 videos, each 120 mins, avg. playback rate 1.5 Mbits/sec.

17. Performance model

18. Performance - LAW vs. MFQ Arrival rate : 8 req/min Skew factor : 0.3 LAW perform slightly better than MFQ in service latency, throughput, defection rate

19. Performance

20. Performance Compare MFQ-SB-n with altering one of –Server Capacity (channels) –Request Arrival Rate –Skew Factor

21. Alter Server Capacity (channels)

22. Alter Request Arrival Rate

23. Alter Skew Factor

24. Alter request rate & skew factor

25. Conclusion Prove that existing scheduled multicast schemes are not suited for hybrid design Proposed a new technique called Largest Aggregated Waiting time first (LAW) AHA is capable of coping with a changing workload

26. Periodic Broadcast (1996) video … … … time PB v.s. batch: –Short initial delay –Large client-side buffer Client requests

27. Batching (1993) 0t1t1 t2t2 t3t3 Client requests time Batch window: –The time interval to initiate a batch stream.

28. Adaptive hybrid approach