Reliable and Fault-Tolerant Storage Systems - Scalable Continuous Media Streaming Systems - page 108

Table 5.3 Buffer requirements ( Q = 64KB, N D = 5)

Quantum Quantum Seagate Seagate

Rebuild Algorithm Atlas-III Atlas-10K Barracuda Cheetah IBM 9ES

Block-based rebuild w/o
buffer sharing

107 MB 161 MB 67 MB 175 MB 126 MB

Block-based rebuild with
buffer sharing

69 MB 104 MB 43 MB 112 MB 81 MB

Track-based rebuild w/o
buffer sharing

182 MB 326 MB 132 MB 244 MB 272 MB

Track-based rebuild with
buffer sharing

114 MB 222 MB 89 MB 132 MB 191 MB

Pipelined rebuild 70 MB 105 MB 44 MB 113 MB 83 MB

In addition to the rebuild algorithms studied in this chapter, there are also other techniques
that may further improve rebuild performance. In particular, when serving active media streams
in rebuild mode, the system has to recover lost data blocks for playback purpose. By storing
these already reconstructed data blocks to the spare disk, one may be able to further shorten
the rebuild time.
However, storing individual data blocks to the spare disk might also adversely affect disk
efficiency in track-based rebuild. First, compared to track-based rebuild, more time is spent
seeking rather than data transfer in block-based rebuild. Hence, the reduction in reading from
the data disks is offset by the loss in disk efficiency in the spare disk. Second, depending on
the placement policy, rebuilding individual blocks may also require changes to the track-based
rebuild algorithm as some tracks will have some of the blocks already reconstructed. Therefore,
the performance impact is not obvious and more work is required to determine the applicability
of such technique.

References
[1] F.A. Tobagi, J. Pang, R. Baird, and M. Gang, Streaming RAID TM : A Disk Array Management System for Video
Files, Proceedings of ACM Conference on Multimedia '93 , Anaheium, CA, USA, Aug. 1993, pp. 393-400.
[2] S. Berson, L. Golubchik, and R.R. Muntz, Fault Tolerant Design of Multimedia Servers, Proceedings of ACM
SIGMOD International Conference on Management of Data , San Jose, CA, USA, May 1995, pp. 364-375.
[3] A. Cohen, W.A. Burkhard, and P.V. Rangan, Pipelined Disk Arrays for Digital Movie Retrieval, Proc. of IEEE
International Conference on Multimedia Computing and Systems , May 1995, pp. 312-317.
[4] M.S. Chen, H.I. Hsiao, C.S. Li, and P.S. Yu, Using Rotational Mirrored Declustering for Replica Placement in a
Disk-Array-Based Video Server, Proc. of ACM Multimedia '95 , San Francisco, CA, USA, Nov. 1995.
[5] B. Ozden, R. Rastogi, P. Shenoy, and A. Silberschatz, Fault-tolerant Architectures for Continuous-Media Servers,
Proc. of ACM SIGMOD International Conference on Management of Data , Montreal, Canada, June 1996, pp.
79-90.
[6] A.N. Mourad, Issues in the Design of a Storage Server for Video-on-Demand, ACM Multimedia Systems , vol. 4,
April 1996, pp. 70-86.
[7] A. Cohen and W.A. Burkhard, Segmented Information Dispersal (SID) for Efficient Reconstruction in Fault-
tolerant Video Servers, Proc. of 4th ACM International Multimedia Conference , Boston, USA, Nov. 1996, pp.
227-286.