Reliable and Fault-Tolerant Storage Systems - Scalable Continuous Media Streaming Systems - page 106

50

40

30

20

10

0

0 100 200 300 400 500 600 700

Media Block Size (KB)

Block-based Rebuild (0% Utilization)
Track-based Rebuild (0% Utilization)
Block-based Rebuild (25% Utilization)
Track-based Rebuild (25% Utilization)
Block-based Rebuild (50% Utilization)
Track-based Rebuild (50% Utilization)

Figure 5.12 Rebuild time versus media block size (Quantum Atlas-10K)

5.8 Summary
In this chapter we have investigated two algorithms for rebuilding data lost in a failed disk to a
spare disk automatically and transparently. We first presented a block-based rebuild algorithm
derived from the conventional CSCAN disk scheduler and analyzed its performance. A buffer-
sharing scheme was then introduced to eliminate the additional buffer requirement during
rebuild. Next, we presented a track-based rebuild algorithm that can reduce the rebuild time
by 70-80%. The large buffer requirement incurred in track-based rebuild is then reduced to
insignificant levels by a novel pipelined rebuild algorithm. Numerical results show that it is
feasible to completely rebuild a failed disk using the presented rebuild algorithms in a practical
amount of time without causing any performance degradation to the media server.
While the algorithms presented in this chapter are designed for media servers serving CBR
media streams, they can also be extended to media servers serving variable-bit-rate (VBR)
media streams. One possibility is to replace fixed-size block retrievals with variable-size block
retrievals, with the block size corresponding to the video bit-rate. As long as the sizes of the
blocks to be retrieved in a disk round is known, we can use the same worst-case analysis as
in Section 5.5 to determine how much disk time to allocate for rebuild. The rest of the rebuild
process will be similar.