Video Staging: A Proxy-Server-Based Approach to End-to-End Video Delivery over Wide-Area Networks

1. Video Staging: A Proxy-Server-Based Approach to End-to-End Video Delivery over Wide-Area Networks Zhi-Li Zhang, Yuewei Wang, David H.C Du, Dongli Su Άννα Κυριακίδου Μ-344

2. Introduction • Novel approach to the problem of end-to-end video delivery over WANs using proxy servers situated between LANs and a backbone WAN. • Objective: reduce the backbone WAN bandwidth requirement • Development of a video delivery technique called video staging through intelligently utilizing the disk bandwidth and storage space available at the proxy servers.

3. Network Architecture

4. Video Staging • Prefetch a predetermined amount of video data and store them at proxy servers • Only part of the video is retrieved directly from the central video server across the backbone WAN. The rest of the video is delivered to the user from the proxy server • Decision on whether to stage the entire video or a portion of it hinges on many factors: • Effectiveness of video staging in reducing the WAN bandwidth requirement for the given video • Access pattern of a LAN

5. Video Staging: A single video case • Video staging methods: • Video Staging without smoothing • Video Staging with smoothing • Cut-off After Smoothing • Cut-off Before Smoothing • Integration of the optimal smoothing technique presented in previous work by Salehi, Zhang, Kurose and Towsley • Bandwidth reduction ratio: ratio of the amount of the backbone WAN bandwidth reduction to that of the disk bandwidth required at the proxy.

6. Video Staging without smoothing • Video i, F: frame period, Ni:total number of frames, Sij : size of the jth frame in bits • Peak rate Pi= (max Sij )/F (1<=j<= Ni) • Choose a cut-off rate Ci where 0<=Ci<=Pi*F and divide video i into two parts • Lower part consists of a sequence of partial frames with size Sij,l= Sij –(Sij - Ci)+ and the upper part consists of frames with size Sij,u= (Sij - Ci)+

7. Video Staging without smoothing • The upper part will be duplicated and staged at the proxy whereas the lower part will remain stored at the central server. • The smaller Ci is, the more video data is stored at the proxy. • As Ci decreases, the lower part becomes less bursty and approaches to a CBR stream.

8. Video Staging without smoothing • The backbone WAN bandwidth requirement is reduced from Pi to Ti=Ci/F • The upper part of the video consumes Di=maxSij,u /Famount of disk bandwidth in the worst case. • It also consumes ΣSij,u (j=1:Ni) amount of disk storage space. • Bandwidth reduction ratio Ri=(Pi-Ti)/Di

9. Video Staging with smoothing • Smoothing to further reduce the backbone WAN bandwidth requirement • We assume that all clients on the same LAN have a buffer of size B for smoothing • Cut-off After Smoothing (CAS) • Perform video smoothing first and then select a cut-off rate • The optimal smoothing algorithm generates the smoothest transmission schedule consisting of a sequence of transmission sizes S’ij • Cut-off rate: 0<=Ci <=Pi’*F • Ti=Ci/F, D’i=maxS’ij,u /F

10. Video Staging with smoothing • Cut-off Before Smoothing (CBS) • Select a cut-off rate first and then perform smoothing • Three ways to apply the optimal smoothing algorithm • Smoothing on the lower part (SOLP) • reduces the backbone WAN bandwidth requirement • Smoothing on the upper part (SOUP) • reduces the disk bandwidth required to transfer the data from the proxy. • Smoothing on the upper and lower parts (SOULP) • Partition the client buffer into two separate buffers • Both the reserved backbone WAN bandwidth and the disk bandwidth required at the proxy may be reduced

11. Empirical Evaluation • Based on simulation using MPEG-1 traces. • In all cases, the disk bandwidth requirement decreases as the cut-off rate increases. • CBS and no smoothing methods consume the same disk storage space. • SOUP and SOULP have smaller disk bandwidth requirement.

12. Empirical Evaluation • Ri as a function of the percentage of data staged at the proxy. • The SOULP method outperforms both SOLP and SOUP methods. • As more video is stored at the proxy, SOUP becomes more effective.

13. Empirical Evaluation • 100 streams statistically multiplexed • The effective per-stream disk bandwidth requirement is significantly smaller than the single stream case. • CAS method outperforms the three CBS method most of the time

14. Video Staging: Multiple video case • Multiple Video Staging Design Problem:Given a video access profile A and a disk system with B bandwidth and S storage capacity, determine Ci for each video such that the total reduction in the backbone WAN bandwidth is maximized subject to the disk bandwidth constraint and the disk storage constraint. • User access pattern at a LAN is characterized by a known Zipf distribution.

15. Video Staging: Multiple video case • Heuristic Algorithms • Staging Hot Video Only (SHVO) : stage hot videos entirely at the proxy • Largest Bandwidth Reduction Ratio First (LBRRF): uses Ri in determining which video and what percentage of it to be staged • Ri is a function of both the video characteristics and the user access pattern • The video with largest Ri is favored when allocating the disk bandwidth at the proxy server

16. Video Staging: Multiple video case • Empirical Evaluation • 50 videos, 500 concurrent accesses • Video peak rates range from 5.5-10.3 MBps • Backbone bandwidth requirement as a function of the number of disks available at the proxy. • LBRRF performs better than the SHVO algorithm because it utilizes the disk bandwidth more efficiently • LBRRF consumes more disk storage space