1. August 23, 2001ITCom2001 Proxy Caching Mechanisms with Video Quality Adjustment Masahiro Sasabe Graduate School of Engineering Science Osaka University E-mail : m-sasabe@ics.es.osaka-u.ac.jp

2. Background Growth of computing power Proliferation of the Internet Distributed multimedia applications Increasing of video streaming services over the Internet High-quality and low-delay video streaming service is required

3. Technique for low-delay data delivery Proxy mechanisms –Widely used in WWW systems –Reduce network load Applying to the video transfer Problems A current proxy system only handles files whereas a single video stream is very large Client requests on the video quality considerably differ due to heterogeneity in the available bandwidth Segmentation of video data Video quality adjustment in the proxy

4. Research Targets Proposal of proxy caching mechanisms with video quality adjustment –Low-delay video streaming service while meeting user’s demand Data retrieval with consideration on client’s request Prefetching data that clients are going to require in the future –Reduce required cache buffer size Segmentation of video data for retrieval, caching and forwarding Replacement of cached data with consideration of size, quality and re-usability of data

5. Video streaming system using proxy cache with video quality adjustment Server Client 1 Client 2 Network Proxy Request (high quality) Request (low quality) Cache Buffer Forward (high quality) high quality Cache high quality MPEG-2 Video Stream Request (high quality) Forward (high quality) Read low quality Forward (low quality) Quality Adjustmen t Read

6. Assumptions on our proposed mechanisms MPEG-2 video Unit of data for retrieval, caching and forwarding –GoP (Group of Pictures) A client periodically requests the proxy to send a GoP with user’s demand Rate control for sending video data –TFRC (TCP Friendly Rate Control) Available Bandwidth Video quality Quality adjustment is performed by re-quantization filter 0 5 10 15 20 0102030405060 Average rate [Mbps] Quantizer scale Reciprocal of a quantizer scale  TFRC rate 

7. Data retrieval Cache hit –The proxy adjusts cached GoP to the request and transmits it to the client Cache miss –Server-Proxy bandwidth is sufficient Data retrieval with consideration on current and future demands –Server-Proxy bandwidth is insufficient Data retrieval with consideration on trade-off between quality and delay

8. Data retrieval - Sufficient bandwidth - Considering clients that are going to require the GoP in the future, the proxy retrieves the GoP of maximum quality among their demands GoP BeginningEnd Video Client 4 2 Client 3 5 Client 2 3 Client 1 4 User’s request on video quality 5 Cache miss

9. Data retrieval - Insufficient bandwidth - Trade-off between quality and delay Which does the user give priority to, quality or timeliness? Introducing a parameter   is defined as the ratio of the acceptable quality to the demand  10 timelinessqualityuser’s preference

10. Data retrieval - Insufficient bandwidth - Control based on  Quality of the client’s request Quality of cached GoP Quality of the client’s request Quality of the GoP which the proxy can retrieve from the server Quality of the client’s request Quality of the GoP which the proxy can provide The proxy sends the cached GoP The proxy retrieves the GoP from the server The proxy retrieves the GoP of the acceptable quality from the server  < Yes No    

11. Prefetching mechanism Range of examination for prefetching (prefetching window) Position in the video stream Candidate GoP for prefetching Requested GoP Requested quality Quality of the cached data

12. Replacement Algorithm BeginningEnd PriorityHigh Low 7 2 64 6427538 1. Choose a candidate GoP for replacement 2. Try the quality adjustment to decrease the size of the candidate 3. Eject the candidate from the cache Video No priority GoP in the prefetching window Requested GoP by a client 1 Candidate GoP for replacement 65

13. Simulation model Video stream is two hours long 10 clients watch the same video stream from the beginning to the end without interactions The inter-arrival time between two successive client participations follows the exponential distribution whose average is 1,800 seconds Server Router Proxy Client 150 Mbps50 Mbps TCP sessions UDP sessions

14. Evaluation criteria Required cache buffer size –Amount of cached data Playout delay –Maximum jitter Degree of satisfaction on delivered video –The average ratio of the delivered video quality to the demand

15. Simulation result - Amount of cached data - Time [sec] 0 20 40 60 80 100 120 140 160 05000100001500020000 Traditional method (infinite cache, no quality adjustment) Proposed mechanism (infinite cache, no prefetching,  ) Proposed mechanism (20 Gbits cache, prefetching,  ) Amount of cached data [Gbit]

16. Simulation result - Playout delay - Client 10 15 20 25 0 0123456789 5 Playout delay [sec] Traditional method (infinite cache, no quality adjustment) Proposed mechanism (infinite cache, no prefetching,  ) Proposed mechanism (20 Gbits cache, prefetching,  )

17. Simulation result - Degree of satisfaction on delivered video - 0 0123456789 1 0. 8 0. 6 0.4 0. 2 Client Degree of satisfaction on delivered video Traditional method (infinite cache, no quality adjustment) Proposed mechanism (infinite cache, no prefetching,  ) Proposed mechanism (20 Gbits cache, prefetching,  )

18. Conclusion Conclusions –We proposed proxy caching mechanisms with video quality adjustment –Simulation results show that our system can accomplish a low-delay video streaming service while meeting user’s demand and available bandwidth Future works –Reducing playout delay –Considering interactions such as rewinding, pausing and fast-forwarding