An Online Video Placement Policy based on Bandwidth to Space Ratio (BSR)

1. An Online Video Placement Policy based on Bandwidth to Space Ratio (BSR) by Asit Dan and Dinkar Sitaram IBM T.J. Watson Research Center

2. Agenda • Background • Aim • Implementation Details • Simulation Result • Difficulty

3. Background • Video Server contain several storage devices • Each storage device has different storage and bandwidth • Each video has different popularity and it change continuously • Each video has different file size and play back bit rate • Popular movie: multiple replicas created on multiple storage devices

4. Aim • Balance the load • Make the best use of bandwidth and space of the storage device

5. Implementation Details First Phase Determines whether additional replicas are necessary (Additional expected load > free bandwidth of the devices on which the replicas reside)

6. Implementation Details For Example Originally movie 1’s demand is 5 unit. Device 1 contain movie 1 After some time, Movie 1’s demand change to 11 unit New replica are needed Device 1 Bandwidth: 10unit

7. Implementation Details Second Phase: Selection of devices BSR of the video objects on a device = Total allocated bandwidth Total allocated Space

8. Implementation Details For example, device 1 contain two movies, BSR of video object on device 1 = (B+C)/(E+F) bandwidth space Free D Free A Allocated for movie 1 E Allocated for movie 1 B Allocated for movie 2 F Allocated for movie 2 C

9. Implementation Details BSR deviation of a device: Deviation of the BSR of the video object in device from the BSR of that device BSR deviation of device 1 = |(A+B+C)/(D+E+F) – (B+C)/(E+F)|

10. Implementation Details For each device without this movie, • Sort device with BSR deviation in decreasing order • Place the movie in the device if BSR deviation of the device can be reduced • Loop break if additional expected load can be satisfied If the load cannot be satisfy, constraint 2 are ignore and run the loop again

11. Implementation Details Third Phase: Reallocate the expected load Allocate the expected load over all replicas, in order to minimize the BSR deviation of the devices

12. Implementation Details Bandwidth Space Bandwidth Space 2 5 8 5 Free 8 Allocated 5 5 2 Device 1 Device 2 BSR deviation of device 1 = |8/5 – 10/10| = 3/5 BSR deviation of device 2 = |2/5 – 10/10| = 3/5

13. Implementation Details After reallocation: Bandwidth Space Bandwidth Space 5 5 5 5 5 5 5 5 BSR deviation of device 1 = |5/5 – 10/10| = 0 BSR deviation of device 2 = |5/5 – 10/10| = 0

14. Implementation Details Forth Phase: Consolidation Execute only if phase 2 failed Try to decrease the replica of other movie, in order to save space and accommodate the new replicas

15. Implementation Details For example: place a new replica of movie 2(load : 9, space : 5) Bandwidth Space Bandwidth Space 5 1 5 1 Free 9 9 Allocated for movie 1 5 5 Device 2 Device 1

16. Implementation Details Bandwidth Space Bandwidth Space Free 1 5 10 1 Allocated for movie 1 9 9 5 Allocated for movie 2 Device 2 Device 1

17. Simulation Results • The expected load of video are in Zipf distribution • Place the video into the system • The system is unable to accommodate all video(stress test) • Test for four different configurations

18. Simulation Results • Each configuration contain 4 storage device • Total Bandwidth of all configurations are the same • Total Storage of all configurations are the same

19. Simulation Results (bandwidth,storage)

20. Simulation Results

21. Simulation Results popularity Video index

22. Difficulty • Difficult to predict the expected load in the next moment accurately • Wrong Prediction has high impact on the performance of the placement policy