Information Technology Reference
In-Depth Information
and the maximum buffer requirement can be computed from
B share =
max
{
B share ( u )
|
u
=
0
,
1
,...,
K
}
(5.36)
Intuitively, the larger the track size (i.e., SY max ) compared to the block size (i.e., Q ), the
more likely that the buffer requirement will be dominated by the rebuild process, and vice
versa. In the next section, we present a novel pipelined rebuild algorithm to reduce this buffer
requirement.
5.6 Pipelined Rebuild
The possibility of track-based rebuild in multi-zone disks stems from the fact that rebuild
requests are non-real-time and hence can be served at variable rates. Another observation is
that tracks are always retrieved sequentially to avoid seek overhead. This sequential property
differs from normal data requests where the order of retrieval can change from round to round
due to the CSCAN algorithm. We present in this section a pipelined rebuild algorithm to take
advantage of this sequential property to reduce the buffer requirement to insignificant levels.
5.6.1 Buffer Requirement
Figure 5.6 depicts the pipelined rebuild algorithm. The scheduling algorithm for retrieving data
from the data disks are the same as track-based rebuild. The difference is in scheduling the
write operations to the spare disk. Specifically, tracks reconstructed from track-based rebuild
are buffered until all track retrievals are completed before writing to the spare disk. By contrast,
Figure 5.6 Pipelined rebuild under ideal scenario of synchronized disks
Search WWH ::




Custom Search