Sei sulla pagina 1di 6

Y.-C. Chang et al.

: Cluster Based Self-Organization Management Protocols for Wireless Sensor Networks 75

Cluster Based Self-Organization Management Protocols for


Wireless Sensor Networks
Yao-Chung Chang, Zhi-Sheng Lin and Jiann-Liang Chen

Abstract — This study proposes a scheme for Self- II. PROPOSED MANAGEMENT ROTOCOLLS
Organization Management Protocols of higher-level nodes to This study implements the proposed management protocols
contest member nodes with multi-hop form hierarchical including Cover Clustering Mechanism for constructing
clusters, introduce the “20/80 Rule” for determining the ratio cluster headers to solve the problems of clustering and
of headers to member nodes, and develop a new cluster-based broadcast storm by cover contest, the Inter-cluster Routing
routing protocol that integrates the inter-cluster on-demand Algorithm and Intra-cluster Routing Algorithm to provide low
routing and the intra-cluster table-driven routing applied in
cost communications between clusters.
the Sensor Networks for further consumer applications. 1
The proposed mechanism reduces the number of
Index Terms — Wireless Sensor Networks, Cluster, Self-Organization, management nodes in a large-scale sensor network. One of the
Routing. Protocols best methods is to develop hierarchical architecture and apply
the 20/80 rule. Figure 1 illustrates the architecture of clustering
I. INTRODUCTION networks.
Wireless Sensor Network is one promising application on
wireless ad hoc networks. Sensor Network can monitor
ambient condition such as temperature, sound, light and others.
Information is collected from many sensor devices for further
consumer application in the Sensor Network [1]. The
hierarchical management architecture can be applied to handle
numerous sensor nodes [2]-[3]. The lower-level nodes are
managed and organized by the higher-level nodes by using
20/80 rules. Controlling the top-level nodes can decrease the
costs of managing nodes and the communication among them.
The 20/80 rule is a well-known “leadership shorthand term”,
Fig.1: The Architecture of Clustering Networks.
originates from Vilfredo Pareto, who discovered a common
phenomenon: about 80% of the wealth in most countries was Figure 2 introduces the architecture of self-organization
controlled by a consistent minority around 20% of the people management network by Minimum Hop Count (MHC) cover
in most countries. Various routing protocols have been algorithm. After constructing the clustering network, this paper
proposed in ad hoc networks [4]-[7], but are not exactly proposes cluster-based routing algorithm into two aspects:
suitable for Sensor Network because of the scalability. Those intra-cluster routing and inter-cluster routing.
routing protocols for ad hoc networks can be divided into two
categories: table-driven and on-demand routing protocols. The
table-driven routing mechanisms discover and maintain
routing tables even when no usage of network; in on-demand
routing mechanisms, the routes are discovered only when it is
need by the source node. These two routing categories have
their own merits and demerits [8]. This paper proposes a new
cluster-based routing protocol that combines the on-demand
routing in inter-cluster and the table-driven routing in intra-
cluster to be suited for the environment of Sensor Networks.
Fig. 2: The Architecture of Self-organization Network for Wireless
Sensor Networks

1
Yao-Chung Chang is with the National Taitung University, Taitung, This management protocol construct the intra-cluster
Taiwan, R.O.C (e-mail: ycc@nttu.edu.tw) routing with table-driven routing mechanisms similar to
Zhi-Sheng Lin is with the National Dong Hwa University, Hualien,
Taiwan, R.O.C. (e-mail: zlsin@mail.ndhu.edu.tw)
destination sequence distance vector (DSDV). The source
Jiann-Liang Chen is the Chair of Computer Science and Information node broadcasts a route query packet to nodes until the border
Engineering Department, National Dong Hwa University, Hualien, Taiwan, nodes in adjacent clusters around the local cluster receive it.
R.O.C. (e-mail: lchen@mail.ndhu.edu.tw)

Manuscript received January 15, 2006 0098 3063/06/$20.00 © 2006 IEEE

Authorized licensed use limited to: Thinagaran Perumal. Downloaded on March 19,2020 at 12:17:54 UTC from IEEE Xplore. Restrictions apply.
76 IEEE Transactions on Consumer Electronics, Vol. 52, No. 1, FEBRUARY 2006

Therefore, the intra-cluster routing limits the broadcast range


within one distance hop distance of the cluster to lower the
control overhead and reduce the interference of the shared
media. Figure 3 shows how the Intra-cluster routing builds.

Fig. 5: The Architecture of Communications

A. CLUSTERING BY COVER CONTEST


Fig.3: Intra-cluster Routing
The coverage of nearby nodes to form a cluster by contest
Also, the propose protocol acquire the inter-cluster routing has three stages:
by exchanging the relationship between clusters in Fig.4. To
perform the inter-cluster route discovery when the route is z A header node broadcasts a “Cover REQuest” (CREQ)
demanded can reduce the overhead of building and periodically. In the mobility situation, the node is adaptive
maintaining inter-cluster routing. When the demand for inter- to optimize the setting of broadcasts.
cluster route occurs, the source node sends the inter-cluster z This request is delivered to all the sensors in the network to
route request packet (RREQ) in unicast mode to the border notify all active nodes.
nodes to acquire the adjacent cluster’s intra-cluster routing z The low-level nodes determine which of their number is
information. the cluster header using the minimum_hop_count method.
Then, the nodes will forward CREQ to cover nearby nodes
in its radio range, after they have accepted a header.

Figure6 presents how the high-level (C2) nodes compete


with the nodes of low level (C3) to form a cluster header. First,
nodes A and B each broadcast a CREQ, and the low-level
nodes record the gradient of these CREQs. The nodes in level
C3 compare the new gradient with the previous one and
determine which one is the leader of the current cluster. If the
C3 node is determined to be the new leader, this new leader
will flood the CREQ to all nodes in its radio range. Therefore,
Fig.4: Inter-cluster Routing two close headers will fairly compete with the nodes between
them. Following this competition, the hierarchical management
III. COVER CLUSTERING MECHANISM architecture on C1, C2 and C3 levels perform self-
The proposed mechanism reduces the number of organization, manifesting the intelligence, flexibility and fault
management nodes in a large-scale sensor network. One of the tolerance of the Sensor Network.
best methods of so doing is to develop hierarchical
architecture and apply the 20/80 rule. Nodes are defined as
existing in three levels depicted in Fig. 2. Figure 5 presents the
communication architecture associated the three levels. The
communication between the high-level nodes proceeds through
the low-level nodes. The functions of the high-level nodes are
assumed to include those of the low-level nodes. Hence, they
can aggregate the data of the low-level nodes. Also, all sensor
nodes are assumed to be stationary. Fig. 6: Header Broadcasts Cover Request to Form a Cluster.

Authorized licensed use limited to: Thinagaran Perumal. Downloaded on March 19,2020 at 12:17:54 UTC from IEEE Xplore. Restrictions apply.
Y.-C. Chang et al.: Cluster Based Self-Organization Management Protocols for Wireless Sensor Networks 77

B. COVER ALGORITHM N41 is belong to cluster C4 and finds the route to C4 is


available in the third entry of Table 1. Accordingly, node N32
z All nodes determine their levels by measuring their sends the packets to next hop N35. While this step is applied
capability when the power is on. iteratively, the packets are transmitted to node N41. It will use
z High-level nodes periodically broadcast CREQ packets. the Intra-cluster routing of Cluster C4 to convey the packets to
z Nodes in an uncovered set register the high-level node as N44 iteratively in the same way.
the set’s header and record the number of covered wave
hops when they receive a CREQ packet. Then, the node
modifies the source and the hop count of the packet before
re-broadcasting.
z When the covered nodes receive a CREQ packet with the
same number of sequences in the header again, they drop
this packet and count the number of drops.
z If the nodes of the covered set receive a CREQ packet with
various numbers of sequences of headers, to register the
header, then they compare the hop count with the record
hop count and select one of the headers with the fewest
hops. If the chosen header is the new one, then it modifies
the source and the hop count of the packet, and
rebroadcasts the packet.
Fig. 7 Intra-cluster routing algorithm
z A node waits for a random delay whenever it decides to
rebroadcast CREQ.
Furthermore, these gateways, the border nodes to other
z If the drop count exceeds threshold C, then the rebroadcast
is canceled. clusters, had been selected while the intra-cluster routing built.
It doesn’t need to make extra effort to do selection of
gateways. It diminishes the overhead on selection of gateways.
IV. INTER-CLUSTER ROUTING ALGORITHM AND INTRA- The example is depicted in Fig. 3. The cluster C3 can
CLUSTER ROUTING ALGORITHM communicate with C4 through the gateway N41 and with C1
through N13. Moreover, the border nodes between two adjacent
clusters may be more than one node, these candidate nodes can
A. INTRA-CLUSTER ROUTING
be the multi-path gateways to adjacent clusters. It offers not
The intra-cluster routing is constructed with table-driven only the fault tolerance, the failure of sensor nodes caused by
routing mechanisms like destination sequence distance vector being blocked due to the depletion of power, physical damage,
(DSDV) with a bit modification. In this routing mechanism, or the power-saving schedule would not affect the task of
the source node broadcasts a route query packet to every node sensor network, on inter-cluster communication, but also the
until the border nodes in adjacent clusters around the local load balance to avoid the wear out problem. In Fig. 3, the
cluster receive it. By the way, it limits the broadcast range cluster C3 can communicate with C6 by passing either gateway
within the cluster and the next hop of it to lower control N66 or N64. The multi-path routing enhances the robustness of
overhead and to reduce interference of the shared media. wireless sensor network. When one of these gateways is failed,
Every node maintains other nodes’ routing information (such the packets can be transmitted through another gateway.
as destination, next hop, sequence number and the cluster id of Therefore, the relation between local cluster and adjacent
the destination node) within its cluster. Besides, the border clusters can be established with local routing table (as shown
node in adjacent clusters is added into the local routing table, in Fig. 8).
it’ll be the route to adjacent clusters. The routing information
mentioned above is used to make routing selection in intra- Table 1 N32’s Routing information
cluster routing. There is no cluster head being selected to be in C_ID Dest. Next Metric Seq. No ಹ
charge of transmission and routing maintain so that it avoids C6 N66 N34 2 N66-200
the bottleneck and reduces the control packets of choosing C6 N64 N34 2 N64-103
cluster head. The intra-cluster routing algorithm is shown in C4 N41 N35 2 N41-140
Fig. 7. C2 N21 N30 3 N21-201
C3 N31 N30 2 N31-340
The node N32’s routing information of Fig. 3 is described in C3 N30 N30 1 N30-110
Table 1 for example. The node N32 can deliver the packets to … … … … …
node N31 using the routing information in Table 1 easily.
When the packets are going to be transmitted to N41 located on
the adjacent cluster C4, the node N32 has no route to N41 in its
routing table. However, node N32 knows that destination node

Authorized licensed use limited to: Thinagaran Perumal. Downloaded on March 19,2020 at 12:17:54 UTC from IEEE Xplore. Restrictions apply.
78 IEEE Transactions on Consumer Electronics, Vol. 52, No. 1, FEBRUARY 2006

Fig. 8 Intra-cluster routing

B. INTER-CLUSTER ROUTING
The relation between clusters can be extracted from local
routing table while the next hop nodes in adjacent clusters
from the cluster edge are including in it. It is the foundation of
constructing inter-cluster routing. In Fig. 4, the relation
between C3 (cluster 3) and its adjacent clusters is the lines in
red and the C6 is a dotted line in blue. Fig. 9 Inter-cluster routing algorithm

This method acquires the inter-cluster routing by For example (as shown in Fig. 10), the node N32 in C3 want
exchanging the relationship between clusters. To perform the to communicate with the node N52 in C5, but the route to N52
inter-cluster route discovery when the route is demanded can or C5 can’t be found in intra-cluster routing table of N32. The
reduce the overhead of building and maintaining Inter-cluster inter-cluster RREQ packet will be sent to the border nodes,
routing. The on-demand routing protocols in ad hoc network like N66, N64, N41, N13 and N20, for obtaining the adjacent
like AODV and DSR used to utilize flooding as route cluster’s intra-cluster routing. The inter-cluster RREQ will be
discovery method, since there are no previous routes to guide cached in the node which received it and not be removed until
the packets to destination. Nevertheless, it will increase not the expired time is due. When node receives the inter-cluster
only route discovery latency but also the overhead depending RREQ with the same source and sequence number, the RREQ
on the range of flooding. This work proposes a new routing packet will be dropped. If the route can’t be found in the
mechanism to solve this problem. When the demand for inter- border node’s intra-cluster routing table, the border node will
cluster route occurs, the source node sends the inter-cluster send the inter-cluster RREQ again to other border nodes. This
route request packet (RREQ) in unicast mode to the border procedure is repeated as required till it reaches the edge of
nodes to acquire the adjacent cluster’s intra-cluster routing sensor network. In this case, the path to C5 can be found from
information. With these, the inter-cluster route can be created. the intra-cluster routing table of border nodes in C6 and the
The algorithm is depicted in Fig. 9. border node will transmit inter-cluster route reply (RREP)
packet back to the source node; the route to C5 will be added
into the intra-cluster routing table of the nodes located on the
reverse path of inter-cluster RREP traveled. That is, C3 got the
path to C5 by passing C6; and further, if the link between C3
and C6 is broken caused by nodes failure or other irresistible
reasons, the route from C3 to C5 will be reconstructed along
the path through C4, C6.

Authorized licensed use limited to: Thinagaran Perumal. Downloaded on March 19,2020 at 12:17:54 UTC from IEEE Xplore. Restrictions apply.
Y.-C. Chang et al.: Cluster Based Self-Organization Management Protocols for Wireless Sensor Networks 79

Fig. 13 illustrates the average latency of the end-to-end


communication. The pair of source and destination is elected
randomly. This measure the delay time of a packet between
sent by source and received by destination. Whereas the DSR
and AODV are reactive routing protocols, route is obtained
while it is demanded. The route discovering may delay the
transmission of data packets by queuing them in the buffer.
Therefore, the latency will be more than CB-BGP which is
using proactive scheme in clusters. In CB-BGP, The packets
could be delivered from one cluster to next cluster quickly by
using intra-cluster routing information. The track of clusters
Fig. 10 Inter-cluster routing also can be obtained on demand by using unicast instead of
broadcast to reduce the energy consumption.

V. EXPERIMENTAL RESULTS

3DFNHW'HOLYHU\5DWLR

To evaluate the performance of the Cluster-Based Border
Gateway routing Protocol (CB-BGP), the comparison with two 
existing well-known protocols, AODV and DSDV, is archived '65
by using these metrics of throughput, average latency, control 
$2'9
overhead and the packet delivery ratio for performance 
&%%*3
comparison. The packet delivery ratio is the rate between
numbers of data packets received by the destination and the 
number sent by the source. The average latency is the average 
delay time of a packet between transmitting from the source      
1XPEHURI1RGH
and receiving at the destination. Routing overhead is the total
number of packets generated during discovering routing paths. Fig. 11 Packet delivery ratio with different network size
Throughput, expressed in data units per period of time, is the
number of bits passing through a data communication system. 

&RQWUR2YHUKHDG SNW

In the experiments, the network consists of 50 to 150 sensor '65



nodes in a 250m by 250m square region. These nodes having
 $2'9
radio coverage of 50 meters are randomly placed in the sensor
field, and have already been clustered into several clusters  &%%*3
with a unique composite-id including cluster-id and node-id. 
Constant bit rate (CBR) is used as traffic source and no sensor 
node with mobility occurred. The sources send 64 byte date 
per packet at the rate of 1 packet per second to destination. 
The capability of channel is set to 19.2kbps.      
1XPEHURI1RGH
The Packet Delivery Ratio will decrease with the increase of Fig. 12 Control overhead with different network size
sensor nodes in wireless sensor network. The main cause is
that the packets were dropped due to the high congestion of
routing overhead around sensor nodes. The Fig. 11 plots the These measured values of throughput with three protocols
Packet Delivery Ratio of DSR and AODV have a big descent are depicted in Fig. 14. It shows that the amount of throughput
when the size of network is more than 100 nodes. This is decreases as the size of nodes in network increases. Heavy
because that DSR and AODV will broadcast RREQ routing overhead will result in the congestion and the data
throughout the entire network to discover a new route while no packets will be crowded out. However, the CB-BGP takes the
route was found to destination. The flooding of routing advantages of cluster structure and multi path to target to avoid
overhead will increase with network scale (as shown in Fig. bottleneck. Therefore, the slope of CB-BGP decreases slowly.
12) and have great influence on the performance of whole This is why it still perform well throughput in large network.
network. Although the increasing overhead also effect on the
CB-BGP, but the Packet Delivery Ratio declined slowly VI. CONCLUSION
because that CB-BGP limit the flooding of RREG within the This study implements the proposed management protocols
cluster. including Cover Clustering Mechanism for constructing
cluster headers to solve the problems of clustering and

Authorized licensed use limited to: Thinagaran Perumal. Downloaded on March 19,2020 at 12:17:54 UTC from IEEE Xplore. Restrictions apply.
80 IEEE Transactions on Consumer Electronics, Vol. 52, No. 1, FEBRUARY 2006

broadcast storm by cover contest, the Inter-cluster Routing REFERENCES


Algorithm and Inra-cluster Routing Algorithm to provide the
low cost communications between clusters. In a fair [1] C.-C. Chiang and M. Gerla, “Routing in Clustered Multihop Mobile
competition among numerous nodes, the number of headers is Wireless Networks,” Proceedings of 11th Internal Conference of
Information Networks, vol. 9B-1, pp. 1–10, 1997.
decided by the 20/80 rule, which is a good guide for wireless
[2] D. Ganesan, R. Govindan, S. Shenker, and D. Estrin, “Highly-resilient,
sensor networks. Each cluster has the minimum hop count and Energy-efficient multipath Routing in Wireless Sensor Networks”
maintains the balance of cluster headers in the area. Both of Mobile Computing and Communications Review, vol. 4, no. 5, October
node density and the number of headers can be varied to 200
[3] T. Clausen et al. “Optimized link state routing protocol (OLSR)”
improve cover loss. The cluster-based routing protocol this [Online] Available: http://www.ietf.org/internet-drafts/draft-ietf-
paper proposed uses table-driven routing mechanism to manetolsr-07.txt
construct the intra-cluster routing and builds the inter-cluster [4] C. E. Perkins et al. “Ad hoc on-demand distance vector (AODV)
routing” [Online] Available: http://www.ietf.org/internet-drafts/draft-
routing by exchanging the relation of clusters on demand. ietf-manet-aodv-13.txt
Applying this cluster based self-organization management [5] D. B. Johnson. “The dynamic source routing for mobile ad hoc networks
protocols in Sensor Network can reduce the amount of packets (DSR)” [Online] Available: http://www.ietf.org/internet-drafts/draft-
ietf-manet-dsr-08.txt
flooding throughout entire network and improve the routing [6] Charles Perkins and Pravin Bhagwat, “Highly dynamic destination
efficiency in the further consumer applications. sequenced distance vector routing (DSDV) for mobile computers”
Proceedings of ACM/SIGCOMM’94 Conference on Communications
Architectures, Protocols and Applications, pp. 234–244, 1994.
 [7] J. Broch, D. A. Maltz, D. B. Johnson, Y. C. Hu, and J. Jetcheva. “A
'65 Performance Comparison of Multi-Hop Wireless Ad Hoc Network

$YHUDJH/DWHQF\ VHF

$2'9 Routing Protocols.” Proceedings of ACM/IEEE MobiCom, October


 1998.
&%%*3


Yao-Chugn Chang (M’03) was born in Taiwan on

August 29, 1974. He received the BS degree (1996) in
 Computer Science and Information Engineering from
Tamkang University (TKU) at Taipei, the MS degree

(1998), and the PhD degree (2006) in Computer Science
 and Information Engineering from National Dong Hwa
           University (NDHU), Hualien, Taiwan. His main research
1XPEHURI1RGH focuses on the network related topics including Transition of IPv4/IPv6,
Fig. 13 Average latency with different network size Mobile IPv6, Network Mobility and Sensor Network. He is also the member
of R&D Division, Taiwan National Information and Communications
Initiative (NICI) IPv6 Steering Committee. Currently, he is working on the
 Mobile IPv6 API for Win CE Project of R&D Division.

 Zhi-Sheng Lin was born in Taiwan 1974. He is the
 master student in the Department of Computer Science
7KURXJKSXW

 and Information Engineering of National Dong Hwa


 University, Hulaien, Taiwan. His current research
'65 interests is in the wireless sensor network.


$2'9
 &%%*3

 Jiann-Liang Chen (M’96) was born in Taiwan on
           December 15, 1963. He received the Ph.D. degree in
1XPEHURI1RGH Electrical Engineering from National Taiwan University,
Taipei, Taiwan in 1989. Since August 1997, he has been
Fig. 14 Throughput with different network size
with the Department of Computer Science and
Information Engineering of National Dong Hwa
University, where he is a professor and department chair
now. His current research interests are directed at cellular mobility
management and personal communication systems

Authorized licensed use limited to: Thinagaran Perumal. Downloaded on March 19,2020 at 12:17:54 UTC from IEEE Xplore. Restrictions apply.

Potrebbero piacerti anche