Information Technology Reference
In-Depth Information
L
3
Leader
S
Others
S:
Server
L
2
1
S
L
1
2
3
1
4
5
S
6
7
L
0
89210 11 12 3131415 61718419 20
1
5
21 22 23
S
24
25 26
6
27 28 29 73031
Figure 1.21
Control topology of NICE.
L
3
Leader
S
Others
S:
Server
L
2
1
S
L
1
2
3
1
4
5
S
6
7
L
0
89210 11 12 3131415 61718419 20
1
5
2122 23
S
24
25 26
6
2728 29 73031
Figure 1.22
Multicast tree of NICE.
are maintained periodically. If cluster size is smaller than
k
or larger than 3
k
- 1,
follow-up cluster merges or splits will be performed (see Figure 1.21).
A cluster leader at layer
Lj
, is also a cluster leader of a certain cluster at a
lower layer
Li
, where
Li
<
Lj
. In NICE, cluster leadership may change as mem-
bers join or leave. If there is a change in leadership of a cluster
C
, in layer
Lj
,
the current leader of
C
removes itself from all layers >
Lj
. Each affected layers
then must choose a new leader. New leaders join their super-cluster. If the
state of the super-cluster is not locally available, the node will need to contact
Rendezvous Point (RP; see Figure 1.22).
There is a possible bottleneck because in this hierarchical architecture. The
node at the top layer would need to deliver data to about
k
× (log
n
) members.
As the scale increases, this will become the bottleneck of a system.
