A Novel Video Layout Strategy for Near-Video-on-Demand Servers

1. A Novel Video Layout Strategy for Near-Video-on-Demand Servers Shenze Chen & Manu Thapar Hewlett-Packard Labs 1501 Page Mill Rd. Palo Alto, CA 94304

2. Table of Contents • Introduction • Data Layout Strategy • Matching Disk Bandwidth with Application Requirement • Disk Optimization Strategies • Conclusion

3. Introduction • TVoD allocates each client a dedicated channels for video streaming. • For NVoD system, the number of channels needed is significantly smaller. • The cost of TVoD system is much higher than that of NVoD • NVoD requires a much smaller number of video channels • Usually, no VCR control is provided for NVoD system. • Scalability is no longer the problem (with multicast /broadcast), instead disk throughput may now be the bottleneck.

4. Data Layout Strategy • Movie is broken down in to segment or logical block. • The segments are placed on the disk as in the following diagram.

5. Data Layout Strategy • Movie is not placed across a disk array. • By using this layout, • The disk head can read continuously track by track without any seek within a period. • Disk seeks are eliminated, except for the seeks from the innermost to the outermost track between sweeps. • When a disk fails, only one movie is out of air, other movie storing on other disks are not affected.

6. Matching Disk Bandwidth with Application Requirement • Placing Multiple Movies on a Single Disk • In most cases, the disk bandwidth is much more than that of the video playback rate • To utilize these bandwidth, we can put multiple movies on a single disk. • The data rates of the movies must be identical. • However, the size of the movies need not be the same, approximately the same is already ok.  movies need not to be of same length.

7. Matching Disk Bandwidth with Application Requirement • The following diagram illustrates the placement of 3 movies on a single disk.

8. Disk Optimization strategies • With zoning in disk, the disk is divided into multiple zones with different sizes and transfer rates.

9. Disk Optimization strategies • When doing deterministic performance analysis, we are limited to use the lower data rate at the inner zone for calculation. • Thus the deterministic performance greatly reduced. • Two strategies are developed to deal with this problem.

10. The Segment-Group-Pairing (SGP) Strategy • This strategy is based on track-pairing scheme. • In track pairing scheme, an outer track is paired with an inner track so as to average out the discrepancy in transfer rate. • However, since the each retrieval of data block involve a seek, the performance of the disk is greatly reduced. • SGP is devised to tackle this problem.

11. The Segment-Group-Pairing (SGP) Strategy • In SGP, • Instead of pairing an outer track with an inner track, we pair an outer zone with an inner zone. • Data to be retrieved in a service round is broken down into two segments, one is placed in the outer zone, and the other is placed in the inner zone. • Advantages: • Larger block size and longer service round can be used. • Only one seek is involved in a single service round. • Data rate of the disk is averaged out. • The deterministic performance of the disk is improved.

12. The Disk Pairing (DP) Strategy • A similar approaches to average out the data rates of zones. • Suppose we have movies of different data rates to be placed in n disk. • In DP, • Every movie is divided into n parts • If movie A is of highest data rate, then each part of movie A is placed in the outer zones of the disks. • If movie B is of second highest data rate, then after the data of movie A is allocated on disk, B will be allocated on the unused highest data rate zones remaining on the disks.

13. The Disk Pairing (DP) Strategy • Playback time of the movies are staggered by Tr/n. where Tr is the period of the transmission. • This architecture also enable load balancing between the disk because at each time instant, every disk is scheduling the channels for a movie only.

14. The Disk Pairing (DP) Strategy • Example: • suppose we have 2 disk and 2 movies. • Movie A and B are divided into 2 parts, according to the Data Layout Strategy. • First part of movie A is stored on outer zone of disk 1; First part of movie B is stored on the inner zone of disk 1; • Second part of movie A is stored on outer zone of disk 2; Second part of movie B is stored on the inner zone of disk 2;

15. The Disk Pairing (DP) Strategy • Movies are divided “horizontally”.

16. The Disk Pairing (DP) Strategy • By the time when the first part of movie A in disk 1 is serving the channels, the second part of movie A in disk 2 should be idle. • By the time when the first part of movie B in disk 2 is serving the channels, the second part of movie B in disk 1 should be idle. • Perfect Load Balancing can be achieved. • Movies with higher data rate can utilize the higher data rate zones in the diskperformance improve.

17. Conclusion • In this paper, a disk layout strategy is proposed. • According to the proposed disk layout strategy, two disk optimization techniques can be applied. • Segment Group Pairing (SGP) • Disk Pairing (DP) • In SGP, the deterministic performance is improved by average out the discrepancy of data rate in different zones. • In DP, movies with higher data rate are allocated to zone with higher throughput, thus improving the deterministic performance.