Information Technology Reference
In-Depth Information
To tackle this problem, we can reshuffle the order of the elements in the data vector for
computing
c
1
,
1
to obtain
c
1
,
1
=
f
2
,
1
v
8
+
(
f
2
,
2
v
5
+
f
2
,
3
v
6
+
f
2
,
4
v
7
)
+
f
2
,
5
v
9
(15.10)
which then allows us to reuse
c
1
,
1
in the computation:
7
c
1
,
1
=
f
2
,
1
v
8
+
f
2
,
j
−
4
+
1
v
j
−
f
2
,
1
v
4
+
f
2
,
5
v
9
j
=
4
=
f
2
,
1
v
8
+
(
c
1
,
1
−
f
2
,
1
v
4
)
+
f
2
,
5
v
9
(15.11)
This reduces the number of data block transmissions from 5 to 3. Note that the client will
also need to use the reshuffled order when decoding the parity group for playback. This parity
group order information can either be generated dynamically, or sent along the video data
blocks as meta-data.
Interestingly, there may be more than one way to reuse redundant block in updating the
redundant data, and possibly with different redundant update overhead. For example, consider
the computation of
c
2
,
1
:
11
c
2
,
1
=
f
2
,
j
−
8
+
1
v
j
j
=
8
=
(
f
2
,
1
v
8
+
f
2
,
2
v
9
)
+
f
2
,
3
v
10
+
f
2
,
4
v
11
(15.12)
If we reshuffle the order of computations for
c
1
,
1
to
c
1
,
1
=
(
f
2
,
1
v
8
+
f
2
,
2
v
9
)
+
f
2
,
3
v
5
+
f
2
,
4
v
6
+
f
2
,
5
v
7
(15.13)
then we can reuse
c
2
,
1
in the computation:
11
c
1
,
1
=
f
2
,
j
−
8
+
1
v
j
−
f
2
,
3
v
10
−
f
2
,
4
v
11
j
=
8
+
f
2
,
3
v
5
+
f
2
,
4
v
6
+
f
2
,
5
v
7
(15.14)
=
(
c
2
,
1
−
f
2
,
3
v
10
−
f
2
,
4
v
11
)
+
f
2
,
3
v
5
+
f
2
,
4
v
6
+
f
2
,
5
v
7
However, in this case the number of data block transmissions is five, two blocks more than that
of reusing
c
1
,
1
. Thus, in the SRDU algorithm, the system will first compute the redundant data
update overhead for all reusable redundant blocks and select the one with the lowest overhead
for reuse.
Search WWH ::
Custom Search