Information Technology Reference
In-Depth Information
First, RSS requires only the redundant servers, instead of all servers, to have the additional
memory to buffer redundant units. Therefore, the total server buffer requirement is reduced.
Second, redundant units can be stored continuously on the disks in the redundant servers so
that retrievals are much more efficient. By contrast, redundant units in the original distributed
scheme are scattered on the disk and hence a separate disk I/O is required to retrieve each
redundant unit. Third, under RSS, retransmission of the redundant units can start as soon as the
failure is detected, without the need to wait for the current stripe unit to complete transmission.
This is possible because the redundant servers are idle before a failure is detected.
Assume failure occurs at time t f during the transmission of stripe j , then it will be detected
latest by time ( t f
T D ). Since retransmission of redundant stripe units can start immediately
upon failure detection as shown in Figure 11.7, the required redundant unit will be transmitted
by time ( t f
+
+
T D +
T a v g ). Now let t j be the time for which transmission of stripe j ends. Then,
The retrieved redundant units have
been discarded here because the
failure is not yet detected.
Failure occurs here
Failure detected here
. . .
Transmission
Server 0
. . .
Retrieval
Hence the needed redundant units
have to be retrieved again.
Figure 11.6 The AGSS scheduler discards retrieved video blocks once transmission is completed
S 0
36
32
28
24
20
16
12
8
4
0
S 1
37
33
29
25
21
17
13
9
5
1
S 2
38
34
30
26
22
18
14
10
6
2
S 3
11
7
3
S R
P9
P8
P7
P6
P5
P4
P3
P2
P3
P2
P1
P0
T D
S 3 failed
Failure detected,
start transmitting
redundant units
Redundant units
retrieved but
not transmitted
Figure 11.7 Transmission scenario for the redundant server scheme
Search WWH ::




Custom Search