Information Technology Reference
In-Depth Information
S
0
S
1
S
2
S
3
S
4
S
5
s
0
u
0
u
1
u
2
u
3
p
0
u
4
u
5
u
6
p
1
s
1
u
7
u
11
u
8
u
9
p
2
s
2
u
10
u
12
p
3
s
3
u
13
u
14
u
15
p
4
s
4
u
16
u
17
u
18
u
19
Figure 14.3
Storage configuration for distributed sparing
S
0
S
1
S
2
S
3
S
4
S
5
u
0
u
1
u
2
u
3
p
0
u
2
u
7
u
4
u
5
u
6
p
1
u
6
u
9
p
2
p
2
u
10
u
11
u
8
u
12
p
3
s
3
u
13
u
14
u
15
p
4
u
16
u
16
u
17
u
18
u
19
Figure 14.4
Storage configuration after lost-data rebuild in distributed sparing
Distributed sparing has the advantage that all servers are utilized under normal mode. How-
ever, the extra restoration phase increases system repair time considerably because the network
and server throughout are limited. Therefore, we consider only hot-sparing in this chapter.
14.1.2 Data Rebuild Computation
To rebuild the data in the failed server, we can perform erasure correction computation using
the remaining data and redundant data units. To do this, we will need to send the remaining
data and redundant units to a host to perform the erasure correction computation. These data
transmissions obviously will add to the streaming workload of the remaining servers. To avoid
adversely affecting the on-going streaming sessions, it is therefore necessary to use only the
residual capacities in the retrieval and transmission schedulers for such a purpose.
In the following we investigate five rebuild algorithms - disk migration, reloading data from
back-up, baseline rebuild, distributed rebuild, and mixed distributed-baseline rebuild. The first
two are simple solutions which also require extra equipment and/or human intervention in the
process. The last three algorithms are automatic and transparent, automatic in the sense that
Search WWH ::
Custom Search