Information Technology Reference
In-Depth Information
Figure 5.3 Retrieval and transmission scheduling algorithm
enables the server to sustain non-stop service even when one of the disks fails by computing
the unavailable media block using erasure-correction computation over the remaining blocks
in the parity group.
Let R v be the media bit-rate. Then the retrieved ( N D
1) media blocks will be transmitted
in the next service round and the service round length is thus given by
( N D
1) Q
T r =
(5.1)
R v
Under this scheduling algorithm, the total number of buffers required is given by
B p =
KN D Q
+
K ( N D
1) Q
(5.2)
where the first term is the buffer requirement for retrieval, the second term is the buffer
requirement for transmission, and K is the maximum number of requests that can be served in
a service round (see Section 5.3.4). Transmission requires fewer buffers because the retrieved
parity block is not transmitted and hence the buffer can be reused.
For a server with a large number of disks, the single parity disk may not provide sufficient
redundancy to maintain acceptable reliability. This problem can be solved by dividing the disks
into clusters where each cluster has its own parity disk (e.g., Streaming RAID [1]). Multiple
disk failures can be sustained as long as no more than one disk fails in a cluster. Results
Search WWH ::




Custom Search