chapter 2 proposed clustering methodshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter...

39
15 CHAPTER 2 PROPOSED CLUSTERING METHOD 2.1 INTRODUCTION Vehicles are grouped based on the location of clusterhead. Road side units located at certain predefined places like junctions, traffic signals, hospitals, restaurants, congested places, shopping malls, city exit points and toll gates act as static clusterheads. Vehicles that are within the range of static clusterhead become its members and information is shared between them in a full duplex manner. All static clusterheads are attached to the central base station for regulating traffic and instructing them to take decision about path for vehicles during peak hours. During high mobility conditions, the selected optimal or shortest path for data propagation might not lead to successful communication (Rezaei et al 2009). The proposed hierarchical clustering will address this problem of link failure and dynamic mobility to a great extent. 2.2 CLUSTERHEAD SELECTION METHOD The propagation of vital data in packets from a source node to destination node without any loss is important in determining the efficiency of the overall composition of our proposed system. The location of clusterhead decides the formation of cluster, based on its transmission range respective to the other cluster members. Two types of clusterheads are incorporated for data dissemination, static and dynamic clusterheads. Predefined road side units act as static clusterheads; they can be located at road junctions, traffic

Upload: others

Post on 12-May-2020

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

15

CHAPTER 2

PROPOSED CLUSTERING METHOD

2.1 INTRODUCTION

Vehicles are grouped based on the location of clusterhead. Road

side units located at certain predefined places like junctions, traffic signals,

hospitals, restaurants, congested places, shopping malls, city exit points and

toll gates act as static clusterheads. Vehicles that are within the range of static

clusterhead become its members and information is shared between them in a

full duplex manner. All static clusterheads are attached to the central base

station for regulating traffic and instructing them to take decision about path

for vehicles during peak hours. During high mobility conditions, the selected

optimal or shortest path for data propagation might not lead to successful

communication (Rezaei et al 2009). The proposed hierarchical clustering will

address this problem of link failure and dynamic mobility to a great extent.

2.2 CLUSTERHEAD SELECTION METHOD

The propagation of vital data in packets from a source node to

destination node without any loss is important in determining the efficiency of

the overall composition of our proposed system. The location of clusterhead

decides the formation of cluster, based on its transmission range respective to

the other cluster members. Two types of clusterheads are incorporated for

data dissemination, static and dynamic clusterheads. Predefined road side

units act as static clusterheads; they can be located at road junctions, traffic

Page 2: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

16

signals, hospitals, restaurants, congested places, shopping malls, city exit

points and toll gates. Vehicles are represented as cluster members when they

are within the range of static clusterhead and information is shared between

them in a full duplex manner. The selection of dynamic clusterhead is based

on the vehicles, which travel around the city or the vehicle, which travel over

longer distance like buses, having predefined path and time chart to handle

the high mobility situations. In order to maximize the efficiency of data

dissemination, both the static and dynamic clusterheads are used as part of our

hierarchical clustering method. There can be one or many number of static or

dynamic clusterheads forming a part of global clustering phenomenon and

thereby improvising the performance of data dissemination, reducing the loss

of data and maximizing the regional coverage.

During high mobility situations the number of vehicles

participating in a VANET will be less and thereby making timely information

dissemination very difficult. Static clusterheads are able to receive

observations and aggregates from passing-by vehicles. They also send

beacons and thereby hand over their knowledge to other vehicles. The benefit

of static clusterheads, however, can also be achieved by connecting them via

dynamic clusterhead to form a backbone network irrespective of the

expanding network, allowing them to exchange information through a large

covered area. This can bring up-to-date knowledge to distant network regions

in very short time. A very limited number of static clusterheads are sufficient

for substantial benefits and when dynamic clusterheads are collaborating to a

parent static cluster head, the number of vehicle participation is improved

without much data losses.

2.3 CLUSTERING METHOD

Clustering in VANET provides a framework for management and

reduces the overhead of route attainment. Many clustering techniques had

Page 3: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

17

been proposed in the literature, but only few of them had considered the status

of network from the aspect of stability. However, the nature of dynamic

changing topology in VANETs introduces difficulties in end-to-end route

finding. Hierarchical clustering is a technique that can dynamically change

the state of clustering in VANET according to different mobility conditions.

More specifically, only stable clusters are formed around static and dynamic

clusterheads. When a new cluster is formed, the member nodes have to inform

the dynamic cluster head of their 2-hop neighbor so that the cluster head can

maintain the topological information of the cluster. Since a cluster is a

proactive domain, the cluster head is informed of the topological changes in

the cluster. Two clusters that collide have to compete according to their

number of stops and direction towards nearby static CH becomes a new CH.

The losing one must dismiss the cluster.

Vehicles are grouped based on the proposed hierarchical clustering

three tier architecture, where all vehicles are at level-0. The chosen dynamic

clusterheads (buses) are at level-1. Static clusterheads that are located at the

predefined places are at level-2. This clustering procedure is performed

recursively until the desired numbers of clusters have been constructed and

each node becomes a member of any one of the clusters. All the updated

level-0 or level-1 vehicle information is propagated into the nearby static

clusterhead whenever they cross. During the cluster formation, all the nodes

that are within the range of static clusterhead will join the nearest cluster upon

receiving a hello message.

2.4 STATIC AND DYNAMIC CLUSTERING

As per the flood-based algorithm, initial cluster construction starts

when a new node enters into the range of static cluster or if dynamic

clusterhead starts to propagate the data. Hello messages are received from

neighbour nodes within the transmission range of clusterhead. The source

Page 4: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

18

node calculates the neighbour node distance and registers the immediate

neighbour nodes along with the 2-hop neighbour information. Instead of

flooding the data to all neighbours, source would flood the data to the selected

nodes based on the direction towards the destination. This avoids cluster

congestion to a great extent. When static clusterhead, dynamic clusterhead

and cluster members are considered top most priority is given to static

clusterhead; routing and data propagation are led by the static clusterhead. In

the absence of static and dynamic clusterheads the node that has more number

of connectivity and majority direction of travel than other nodes, becomes the

clusterhead.

If the new node does not become a member of any cluster or if it is

not within the reachable area of the clusterhead, it will form the temporary

virtual cluster and this first non-cluster node will become the clusterhead for

that region. Further, all non-cluster nodes become the members of that virtual

cluster. Normally the dynamic clusterheads are designed to propagate data

and record the topology variation at different periods of time. After the cluster

construction is initiated, the topology of network will vary, as nodes are

moving with various speeds. Continuous mobility of nodes followed by its

topology changes leads to reclustering. Among the existing mobility models,

random waypoint and RPGM models (Karnadi et al 2007) are common in

urban scenario. The more the dynamicity of the node the more the topology

and its descendent links are disturbed. The nodes, which move very

frequently, are never used in data propagation. Hence frequent mobility will

lead to unbalanced overload conditions and overhead in sharing information

of topology changes. Clusterhead should update appropriate table information

frequently.

For fast-moving vehicles, the location updates in neighbours might

become obsolete by the time they reach the correspondent node. To get the

Page 5: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

19

exact position information of a vehicle, large routing overhead is incured.

A cooperative three-tier framework based routing in such a dynamic

environment that proves to be very efficient is proposed.

2.5 HIERARCHICAL CLUSTERING

The hierarchical clustering combines the features of static and

dynamic clustering together. Static clusters are formed around the static

clusterheads located at the road signals, street corners and congested places.

Howe ver, buses are chosen as dynamic clusterheads for having

predefined path and time chart to handle the high mobility situations.

Hierarchical clustering creates a layered environment that poses some of the

main challenges in adhoc networks. Top layer consists of static clusterhead,

middle layer consists of dynamic clusterhead and bottom layer consists of

ordinary vehicles. Because of highly dynamic vehicles, network topology also

changes. This in turn affects the performance of the network and also invokes

protocol mechanisms to react to such situations. Mobility awareness deals

with sudden changes in topology by responding against malfunctions in

routing (Basagni et al 1999). Some of mobility metrics are considered for

cluster construction in order to form a stable cluster structure thereby

decreasing its influence on cluster topology. Vehicle mobility behaviour

determines the architecture of the cluster (McDonald and Znati 1999).

Vehicles are grouped in two different ways; group vehicles which are in the

communication ranges of dynamic sources or group vehicles, which are in the

ranges of static sources mounted at traffic signals and road junctions. By

doing so, the reaffiliation and reclustering rate can be naturally decreased.

Hierarchical clustering architecture as shown in Figure 2.1 is a

three-tier architecture where all vehicles are at level-0. Those vehicles that are

within the transmission range of dynamic clusterhead become members of

that cluster. The chosen dynamic clusterheads (buses) are at level-1. Static

Page 6: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

20

clusterheads that are located at the predefined places are at level-2. Based on

the location of level-0 cluster members, static clusterhead is chosen. This

clustering procedure is performed recursively until the desired numbers of

cluster have been constructed and all the nodes become the member of any

one cluster. The formation of three-tier cluster architecture, involves the

recursive operation of clusterhead selection and topology updates. All the

updated level-0 or level-1 vehicle information is propagated into the nearby

static clusterhead whenever they cross.

Figure 2.1 Hierarchical clustering architecture

2.6 PROPOSED HIERARCHICAL CLUSTERING ALGORITHM

Dynamic clustering attempts to partition the number of nodes into

multi-hop clusters based on the following parameters (VID,LID,s,VLT)

defined in CCA Algorithm. The (VID,LID,s,VLT) criteria indicate that every

Page 7: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

21

vehicle node in a cluster has it own unique ID (VID) and Location_ID (LID)

representing the road in a particular area of the city it belongs. The symbol ‘s’

indicates the speed of the vehicle and VLT indicates the vehicle’s lifetime in a

particular cluster, regardless of the hop distance between them. The purpose is

to support robust and efficient routing, and adaptively adjust its dominant

routing scheme depending on the manner of network mobility.

As a dynamic clustering scheme, the existing parameterized

clustering scheme requires no periodic reclustering. As soon as a vehicle

enters the clustering zone its unique VID is registered into the clusterhead and

becomes a member of that cluster. Any unclustered vehicle joins a cluster by

sending out req message. Mobility also affects the size of the cluster. Low

mobility increases the size of the cluster compared to high mobility, where

increase in the number of clusters is observed. A vehicle can join a cluster if it

has a valid VID and its speed is also an important criterion. If any new vehicle

other than ambulance or rescue vehicle enters the cluster with the speed more

than an average speed it is not necessary to update it everywhere. If a vehicle

does not receive a response message after a certain period of time, it will

create a new cluster and it will become the head for this cluster.

The proposed hierarchical clustering algorithm is location based,

and it performs clustering operations on both random waypoint and RPGM.

Some of the existing routing algorithms, based on clustering uses proactive

algorithm within the clusters and reactive algorithm between the clusters

(McDonald and Znati 1999, Yu-Chee et al 2007). In the proposed solution,

CH election is a transparent one for both types of clusterheads-dynamic and

static clusterhead. City buses and selected call taxis act as the dynamic

clusterhead, and mount static clusterheads at the chosen places where

dynamic mobility is high. Each CH node acts as master for all the members

and every event is recoded and updated continuously. CH is responsible for

Page 8: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

22

controlling the data propagation inside and between the clusters. Information

storage overload problems are avoided in VANET because of its abundant

storage facility (Bao et al 2009). The CCA can reduce the burden of CH with

relatively moderate speed and more connectivity. Delay of routing is reduced

by immediate distribution and reclustering. The routing process is separated

into intra-cluster routing and inter-cluster routing.

During intra-cluster routing, the source node would check if the

destination node is in the table of the neighbour. If present, then direct

communication would proceed. Otherwise CH would list out the neighbour

nodes containing destination node entries and start to flood the data to those

nodes. In inter-cluster route, when the CH receives the route request, it would

check the destination node in its cluster table. If yes, then the route between

the destination node and neighbour cluster would be established. Otherwise,

the route request would be forwarded to other CHs. Here mobility impacts the

clustering in two different ways. On one hand, an ordinary node may move

fast, far away from its clusterhead requiring it to affiliate to another cluster.

On the contrary it may move fast within the region causing congestion.

Random mobility propagates the information throughout the cluster

unconditionally but the prediction of the location of the nodes to be updated is

a time consuming process (Kogias et al 2009). Group mobility occurs when

the vehicles wait at the signals or when there is traffic congestion. All the

events regarding the traffic information are updated during the waiting period

at the junction places. Clusters of very small size are combined together to

form a new cluster and also process for identifying clusterhead for this new

cluster is triggered.

2.6.1 Hierarchical Cluster Based Routing

The performance of VANET routing protocols tends to decrease as

the mobility of nodes increases (Wisitpongphan et al 2007, Jerbi et al 2009,

Page 9: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

23

Wenjing et al 2009) resulting in decrease in cluster stability. It is desirable to

investigate how to utilize such rich connectivity and physical heterogeneity to

improve the stability of the cluster. Non-overlapping infrastructure and the

data propagation in the form of 2-hop manner is used. Each cluster should

retain the cluster information even over longer period of time. Based on

information maintained in each clusterhead, inter-cluster routes are

dynamically discovered through flooding followed by multicasting. It has the

features of both on-demand and table driven existence of bi-directional links,

which aid in both intra-cluster and inter-cluster routing.

Static and dynamic clusterhead coordinates the data transmission

within the cluster to other clusters. During inter-cluster routing, the

clusterhead should disseminate short-lived Time-to-Live (TTL) topology

information proactively even when there is no data to be sent. This should

happen frequently at particular intervals. By doing this the nodes confirm the

availability, current status of the vehicle, major changes like clusterhead

changes and link failures. Such proactive mechanism leads to increase in

overhead.

To reduce the overhead, reactive approach can be used in which the

route is searched only when there is data to be sent. Compared to other

approaches, in a reactive approach there is no initial route discovery delay,

which is undesirable in many circumstances. Instead of that initial flooding

and direction based 2-hop multicasting give better results. The propagation of

control packets is limited within each cluster by 2-hop transmission. The

transmission range of the antennas in the vehicles those to form a physical

communication hierarchy loop. Such a hierarchy enables direct information

exchange and it potentially increases routing performance. It is assumed that

there are four types of nodes, (i) static clusterhead, (ii) dynamic clusterhead,

(iii) gateway nodes and (iv) ordinary nodes in the network.

Page 10: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

24

2.6.2 Cluster Formation and Maintenance

The top layer, which contains the static clusterhead, establishes

communication over multi-hop paths. Each clusterhead periodically

broadcasts hello messages containing the cluster_id, and its location

information. Initially there is no dynamic clusterhead and it is set to 0 and the

topology is empty. During the cluster formation, all the nodes that are within

the range of static clusterhead will join the nearest cluster upon receiving a

hello message. When a node receives a new hello message from a different

cluster, based on the distance of the clusterhead, the node can continue or

leave the current cluster and join the new cluster. By checking the cluster_id

in the cluster and the distance, each hello message is relayed towards the

boundary of the cluster.

The details of the algorithm for cluster formation and maintenance

are described in CCA algorithm. The flowchart of CCA algorithm is shown in

Figure 2.2. Cluster parameters such as CJReq, CJRep, Vehicle_id, total

number of vehicles, Neighbour Vehicle_id, Vehicle counter and the number of

static and dynamic clusterheads are initialized. If there is no static CH at a

particular location, the dynamic CH(bus) is chosen as a new CH. Once the

CH has been selected, CJReq message is broadcast to initiate the cluster

formation. After the receival of CJRep cluster is formed around the static or

dynamic CH. Each Vehicle_id is registered in the CH routing table and from

this the number of vehicles in each cluster can be found out. If any one of the

vehicles in a scenario does not receive any CJReq message, it is announced as

temporary CH and it starts to propagate CJReq message to form a cluster

around it. If the temporary CH receives any CJReq from any nearby static or

dynamic CH, it will become a member of that cluster otherwise a node with

more number of stops and direction towards the neighbour becomes

a new CH.

Page 11: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

25

Figure 2.2 Flowchart of cluster construction algorithm

=Zero

No

Initialize cluster parameters

Determine the number of static and dynamic CH

Broadcast CJReq messages and

Initiate cluster formation

Calculate waiting time of CJRep messagesfrom each vehicle within its region

Waiting time

> threshold

Yes

Number of

static CHs

Choose the dynamic

CH (buses) with slow

speed and more number

of stops

>=1

Vehicle received

CJReq message

Announce it as a temp_CH

and form a cluster around it

No

Continue it as a single node

CH

Cluster formed around the static

clusterhead

Each vechile_id is registeredin the CH routing table

Yes

Calculate the number of clusters

Compute the CH and

number of vehicles in each

cluster

temp_CH broadcast CJReq

message to its neighbours

Calculate waiting time of CJRep message

from each neighbour vehicle

Waiting time >

threshold time

A virtual cluster is created

around temp_CH and itsneighbours

No

No

Neighbours are memberof another cluster?

Discard the CJReq

message

Yes

Yes

Any CJReq

from nearby

static or

dynamic CH

Virtual cluster becomes a

member of the nearby static

or

dynamic cluster

The node with the more

number of stops and

direction towards nearby

static CH becomes anew CH

Yes

No

Page 12: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

26

Pseudocode of Cluster Construction Algorithm is shown in

Figure 2.3. Cluster parameters such as CJReq, CJRep, Vehicle_id, total

number of vehicles, Neighbour Vehicle_id, Vehicle counter and the number of

static and dynamic clusterheads are initialized. Static clusterhead will

distribute CJReq to all vehicles within its range and store the number of

vehicles in the CH vehicle counter. Due to high mobility conditions many

relay nodes enter and leave between source and destination. Existing nodes

within the cluster discard CJReq to avoid congestion. If any one of the

neighbours receives CJReq, it will send the CJRep reply to the source.

Otherwise CJReq messages are distributed through the intermediate nodes

and the vehicle counter stores the number of vehicles involved during the data

propagation.

Cluster Construction Algorithm (CCA)

Parameters

CJReq (SVID,IMVID,DVID,VC,s,VLT) - Cluster Join Request

CJRep (DVID,IMVID,SVID,VC) - Cluster Join Reply

CVID - Current Vehicle_ID

DCH - Dynamic Clusterhead

DVID - Destination Vehicle_ID

IMVID - Intermediate Vehicle_ID

LID - Location ID

MV - Master Vehicle (buses)

NV - Number of Vehicles

NVID - Neighbour Vehicle_ID

SCH - Static Clusterhead

SP - Shortest Path

SVID - Source Vehicle_ID

VC - Vehicle counter

VD - Vehicle Distance

VID - Vehicle_ID

At Static Clusterhead

SCH having valid SVID distribute CJReq with LID to

all vehicles within its range

Figure 2.3 (Continued)

Page 13: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

27

Find all Immediate Neighbours(NVID) speed ‘s’ and

VLT

Find VC = # (NV) &

Register & Forward(CJReq) to NV

for each intermediate node

{

for each CJReq received

{

if new NVID == old NVID

drop(CJReq) to avoid repetition

if NVID == DVID

then DV send CJRep to source

else

{

a. CJReq to IMVID add(CVID)

b. Find VC = # (NV)

c. CJReq VC += VC

}

}

}

DCH forms cluster around its transmission range

a. If any new vehicle enters into range (DCH)

register its VLT into DCH routing table

member (DCH) <- new(vehicle)

b. If any new MV enters into the range (DCH)

{

if speed of old (DCH) < new(MV)

no change

endif

if speed of old (DCH) > new(MV)

then new(DCH) <- new (MV)

endif

Figure 2.3 (Continued)

Page 14: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

28

if speed of old(DCH) = new(MV)

vehicles with more LID entries becomes a

new (DCH)

endif

endif

}

At source

{

till(timestamp < threshold time)

{

link = SORT(CJRep IMVID)

for each link

{

find SP = Min (CJRep VC) and LID

send data to destination vehicle through SP

}

}

}

At Destination

{

for each CJReq received, send (CJRep) to source

{

if (speed > thresholdspeed)

update only VID in Clusterhead

else

sort(CJRep IMVID)

update details of vehicle to all nodes

}

}

SORT(CJRep IMVID)

{

for each IMVID in CJRep

{

calculate VD

SORT(IMVID) in ascending order of VD

calculate SP and number of vehicles

store all VID in an array of each clusterhead

}

}

Figure 2.3 Pseudocode of Cluster Construction Algorithm (CCA)

Page 15: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

29

Dynamic clusters are formed around the transmission range of

dynamic clusterhead. If any new dynamic clusterhead (bus) enters into the

dynamic cluster, clusterhead election process starts. The vehicle having more

number of stops and minimal speed becomes a new CH. The source can find

the shortest path to the destination by the distance and the number of

intermediate nodes from the CJRep message. All the events are updated

immediately in each clusterhead

2.6.3 Topology Discovery

In high mobility conditions, only the gateway nodes are involved in

the discovery of topology. All changes in topology should be immediately

propagated without delay. In regular interval period each node can sense the

presence of its neighbours by propagating hello messages. By flooding the

hello messages and using the reply from neighbours, clusterhead can choose

the neighbours based on the nodes having the updated destination information

and that are close to the destination. If a vehicle reaches the boundary of that

cluster and tries to enter the neighbour cluster it should initiate the topology

discovery process for that cluster to find out its initial scenario and update its

topology information. The node, which has the longest distance from the

clusterhead and the direction of travel towards the destination, becomes the

gateway node of that cluster. Those nodes which are at the border should

exchange the cluster update messages with the neighbour clusters’ gateway

nodes too. All the updates received by the border nodes from different

clusters are combined and sent to other nodes for further processing by the

clusterhead. Each node should enable their availability by registering their

node_id to its clusterhead. Flooding the topology updates in a reactive manner

reduces the traffic overhead to a greater extent.

Page 16: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

30

2.6.4 Packet Buffering

The packets are stored temporarily in dynamic clusterhead and

permanently in static clusterhead. Each route is computed from the routing

table entries associated with a lifetime in each node. Each clusterhead will

forward the packet to the neighbour clusterhead through relay nodes or

gateway nodes and reach the clusterhead of the destination cluster. The route

will be removed from the table entries when it expires. For destination node,

which moves frequently and is far away from source node, the route for the

destination cannot be found.

2.6.5 Route Recovery

In VANET, clustering and mobility are the basic parameters, which

determine the efficiency of a routing algorithm. This is because when a

vehicle is moving, it can be placed beyond the coverage range of its

neighbours. This will result in link failure. A novel routing algorithm has to

be quick: vehicles that belong to the failed route must be informed and a valid

replacement route must be discovered for further processing (Noureddine et al

2006). The proposed routing algorithm adopts the common approach for the

detection of route break used in most existing protocols. A link_error packet

is sent back to the source to notify the path break. The source initiates the

reroute process if it still has data to send. However, this approach will lead to

increase end-to-end delay and overhead. In order to recover routes efficiently

a more efficient method is used based on the lifetime of the dynamic

clusterhead at the destination node cluster. In this method the destination node

broadcasts backward beacons to the source. Once, the source of data receives

backward beacons, it chooses the adequate route with respect to the path with

minimal interference. However, due to frequent mobility and topology

changes, the backward beacons may not arrive at the source within a

Page 17: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

31

predefined time. The source of data may trigger a new route discovery by

broadcasting a new route request.

Route recovery process can be initiated without any delay if any

data loss occurs due to route failure. If the route failure is detected within the

cluster, route discovery process is initiated by the clusterhead through the

relay nodes. The relay nodes repair the broken path with minimum overhead.

If this method fails, the pair of nodes where the link has been disconnected,

should flood their status to their neighbouring nodes and find their alternate

way of connection and reinitiate a route discovery.

2.6.6 Routing in Clusters using 2-hop Information

Most research work on VANET clustering is focused on single-hop

clustering. Due to the limited area covered by each cluster and mobility

variations, the single-hop clustering needs frequent updates on routing

information (Vidhyapriya and Vanathi 2007). Recent researchers have shown

enormous interest in 2-hop routing (Uichin et al 2009). 2-hop routing will

play a central role in future enhanced VANET environments because it

improves network routing, avoiding overhead impact as node membership

increases. As a special case, a dense VANET that includes a large number of

vehicles located in a small area at a random period of time is considered.

Dense VANET deployment environments are not only of growing

commercial interest but also they can significantly simplify the design and

implementation of specific protocols and support solutions. In particular, the

2-hop routing, together with the comparative evaluation of novel heuristics

improves its effectiveness that addresses the formation of non-overlapping

clusters. The CHs assign all nodes within the 2-hop neighbourhood to the CH

and assign the role of relay to the nodes placed within the 2-hop

neighbourhood. The proposed algorithm respects the following constraints:

Page 18: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

32

1. Each node is identified by its 2-hop neighbour,

2. Every node must belong to at least one cluster under static

clusterhead or dynamic clusterhead,

3. Each cluster should contain minimum one clusterhead and

4. All the neighbour nodes should receive the initial flooding

information that later on they process with multicast operation

based on the destination location.

2.6.7 Link Path Maintenance

Data packets are able to continuously relay as long as connectivity

between source and destination remains in place. The packet delivery may fail

due to network partition, which is caused by the mobility of vehicles (Xiaoxia

et al 2008). Connectivity information between source, intermediate nodes,

gateways and destination mainly relies on periodical control messages. For

the path maintenance, link_flag of control messages is used to indicate the

connectivity status. Once it is fixed, the source of the link should flood the

status of the link to its nearby neighbours and execute the rerouting process.

The link_flag might be zero in two conditions: (i) when a vehicle moves to its

destination, which is far away; (ii) when the destination becomes unreachable

due to intermediate node movement. If a vehicle does not receive any control

message, either the static or dynamic source would initiate path setup phase to

designate other links and delete old links.

2.7 DATA FORWARDING

Depending on the density of the cluster, source forwards a data

packet towards the immediate neighbours. Source will wait for the

acknowledgement from the neighbours with valid cluster_id, to avoid the

Page 19: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

33

duplicate data forwarding. If the number of responding group members

exceeds predetermined threshold, a temporary queue is constructed for

collecting the reply messages from the cluster until the flooding is over. If a

requesting member does not get enough replies, data packet is propagated

along a separate path using the shortest path algorithm. A newly built cluster

is confirmed by a control message. If a temporary cluster is constructed, a

node requesting cluster creation broadcasts data packets during cluster

construction period. All nodes within a cluster, rebroadcast the packet if TTL

value is valid. All vehicles that were already forwarded with data packet of

same sequence number ignore packet during flooding. Accordingly, data

forwarding path rooted at sender does not have reversed link toward sender.

The parameter max_cluster_count can be used to avoid broadcast storm

problem by restricting the number of group members building their cluster.

Sometimes cluster members are used to forward the packet between two

clusters. In such cases, cluster members just rebroadcast receiving data

packets. When an intermediate cluster member receives the information from

a node at boundary, it attempts to identify the surrounding cluster members by

sending special control_queue packets.

When a group member does not belong to any cluster for some

period of time, it will announce itself as a clusterhead and start to create a

cluster around it by the data propagation (Gau et al 2002). Such being the

case, instant reliable delivery and efficient resource should be provided to the

new clusterhead. The redundancy can be allowed to some extent, if some link

failure occurs or the destination ignores the data packet. Data delivery failure

will be noticed when a destination cluster does not exist. In such a case, the

member relays data packet again if a node does not request to stop packet

relaying. Data packets can be continuously delivered to group members

through flooding without computation of routing table. It is intended for

networks with unpredictable topological changes and highly dynamic nodes.

Page 20: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

34

However, it uses adaptive mechanisms for restricting flooding in the static

network.

Packets are transmitted between the vehicles within the cluster

during intra-cluster transmission. The routing updates are done by the

clusterhead to its cluster member periodically and the members will update

their routing table after they receive the control packet. Routing tables are

updated by the corresponding clusterhead and gateway nodes (Gunter et al

2007). Packets are processed based on the priority of packets and the longest

path from different clusters. The timeout route items are removed from the

routing table. The overhead during updation process can be allowed. In order

to avoid unnecessary bandwidth utilization and to avoid overhead, the

clusterhead and gateways act as the backbone of routing and all the routing

activities are reflected only on this set of nodes. In dynamic mobility

conditions it is very difficult to maintain the stable link structure and the

topology also varies continuously that the clusterhead might change as well.

Hence the backbone architecture is eliminated altogether and a fully

distributed approach has to be convened.

During inter-cluster communication, two separate channels are used

between clusters: i) control channel; and ii) data channel. The routing

algorithm proposes proactive method for intra-cluster transmission and

reactive method for inter-cluster transmission. Each member of the cluster can

transmit/receive emergency messages and data packets to and from the

cluster-head during flooding. After receiving and processing each CH assigns

channel to transmit and receive different kinds of signals to the corresponding

cluster members. During inter-cluster routing the gateway to be used to

forward packets should be mentioned explicitly. Localization of certain nodes

and addressing packets will forward them to their destination. Each CH

collects the messages from its own cluster members and its nearby clusters

Page 21: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

35

and then consolidates information based on its priority. The CH uses the

control channel for beacon and hello messages. The data channel is used to

forward the data packets to the neighbouring cluster-heads. The vehicles from

different clusters will try to share the common channels to transmit/receive

the packets. Each relay station can randomly pick one of its neighbours,

which is closer to the destination and forward the message towards that node.

Therefore each packet should contain the information about destination,

location, direction and a set of neighbours to be traversed along. The closest

node to the destination location broadcasts the message to all stations in its

radio coverage area (Wiegel et al 2007). The clusterhead construction

algorithm provides the location information of nodes.

Each intermediate node should be aware of the traffic conditions.

The destination will receive the data packets and traffic conditions from

intermediate nodes instead of source. Once the path of transmission has been

established the shortest path algorithm is executed to find the path with

minimal interference. It is removed whenever the next shortest path has been

found. This process is continued until data transmission is over. In each

cluster all the clusterhead assigns its transmission power equal to the power

required to reach the furthest nodes of the cluster. However, the stable

structure has been maintained because all the calculated shortest paths are

available in clusterheads.

2.8 HANDLING FLOODING OVERHEAD

Large clusters are split into smaller ones for better management

purpose and routing. Due to large number of vehicles and different mobility

scenarios it is essential to implement inter and intra-cluster routing to localize

and manage packet routing. Source node will check the routes inside the

cluster by referring its 2-hop neighbour routing table after which direct

communication will happen. Otherwise the next level of 2-hop nodes will be

Page 22: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

36

searched for further process until the destination has been reached. If the

search process continues until the border of the cluster and could not find the

destination node in its cluster region, then it would request route to the nearby

CH. After receiving the route request, it would check if the destination node is

in the nearby cluster table or not. If it is in the table, data will be delivered to

the destination. Otherwise, the route request would be delivered to other CHs.

Source node must know the topology of each cluster through the

gateway nodes before performing flooding. Each clusterhead should know the

vehicle_id of all the vehicles in that cluster and the connectivity among the

cluster nodes and relay nodes for the corresponding cluster. The intra-cluster

connectivity contributes the most to the level-0 routing table size and inter-

cluster connectivity contributes to level-1 routing table size. To distribute the

routing table size relevant to each node, a recursive flooding procedure is

employed for each level in the hierarchal clustering. Dissemination of the

cluster topology is initiated by the level-1 clusterhead.

The topology information in level-0 cluster is flooded only to

members with level-1 hierarchy. The flooding is initiated by the level-1

clusterhead. With the hierarchical structure, each node can unambiguously

forward packets towards the destination direction, which will reduce the

overhead incurred by flooded routing even at high mobility conditions.

2.9 RECLUSTERING

Reclustering provides clustering stability, network robustness,

longer connectivity and message reliability. Distributed mechanism for

clusterhead election, first demands certain level of consistency to be reached

among the nodes. In Vehicular Adhoc Networks (VANET) all the vehicles

communicate in a peer-to-peer fashion, without the need for a network

infrastructure. Reclustering will insert a kind of modified infrastructure

functionality and produce stable hierarchical structures to be built over

Page 23: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

37

dynamically changing clusterheads which perform multi-hop forwarding,

efficient bandwidth allocation and cluster management. Mobility of vehicles

and topological parameters could reduce the advantage of using clusters by

requiring frequent updates of the routing tables. A chosen clusterhead should

keep track of routing table changes and ensure link maintenance.

The existing reclustering algorithms are either dynamic or static

(Avresky and Natchev 2005, Elmusrati et al 2007). The Lowest-ID algorithm

is one of the most popular static algorithms where the clusterhead role is

assigned to the node with the lowest ID (Lian et al 2007). Highest-Degree

(HD) algorithm is one of the dynamic algorithms in which the node with the

largest number of 1-hop neighbours is selected as clusterhead. Each node

should be aware of node mobility pattern, link failures and link reconnections

are caused by the movement of the nodes (Stojmenovic et al 2002). An

efficient reclustering in the proposed hierarchical clustering scheme is

suitable for the infrastructure-less VANET environment. During the

reclustering procedure, the decision-making is based on local information,

i.e., the node priorities based on node degree for clusterhead selection (Yu

and Chong 2005). The stability of the algorithm relates to the communication

overhead that is imposed by the underlying reclustering scheme, which affects

the network throughput. Compared to other reclustering schemes CCA

achieves better connectivity and hence better end-to-end message reliability

but overhead is increased because of the number of clusterhead modifications

due to mobility. The parameters that clusterhead election takes into account in

its procedure are, (i) node degree i.e., the number of 2-hop neighbours that

each node has inside its radio range and (ii) distance traveled by the vehicle.

For example, if the traveling time is calculated CCA will select as

clusterheads the nodes with more number of stops, a choice more suitable for

sparse networks. If it weighs more on the connectivity degree, the nodes

having a large number of 2-hop neighbours will be selected, a choice that is

Page 24: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

38

more suitable for dense networks. Reclustering consists of a setup phase; then

nodes are deployed in the geographical areas and their mobility, topology,

speed, direction, counters and routing tables are initialized. Then the

algorithm of clusterhead selection and the update of the routing tables are

performed.

Vehicles are dispersed across the geographical area according to a

specific deployment pattern. CCA allocates to all nodes a unique vehicle_id.

Each vehicle holds a counter containing the information about the number of

times that it has been elected as clusterhead and a counter for the node degree.

The routing table should contain the information about 2-hop neighbour list

and the control message with all the packets that the node receives. The

neighbour and 2-hop neighbour list for each node is built by reading the

vehicle_id, which is included in the reply hello messages. After some

threshold time if no neighbour is heard, then that node becomes clusterhead.

Each vehicle in the network contains the information about node degree and

its 2-hop information, which are included in the reply packets that each

neighbour broadcasts. Different types of packets are generated during

reclustering cluster_JOIN, cluster_LEAVE and the ACK message to

acknowledge both the JOIN and LEAVE requests of the nodes during mobility

as shown in Table 2.1.

Table 2.1 Message parameters

Message Parameters

cluster_JOIN (vehicle_id,CH_ID)

cluster_LEAVE (vehicle_id,CH_ID)

ACK (vehicle_id,CH_ID)

HELLO (vehicle_id,CH_ID,Source

vehicle_id,Destination

vehicle_id,counter)

Page 25: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

39

CCA sets a maximum to the number of neighbours that a CH is

allowed to advertise by flooding. Gateway nodes allow the acceptable number

of entries into the cluster that is equal to the cluster’s size. Parameters like

size of cluster, transmission rate and connectivity degree should not be

distributed to all nodes. If the cluster member connectivity degree is more

than the threshold, the node should run the rerouting algorithm and maintain

the load balancing with the neighbour nodes. CCA maintains stability

between the clusterhead and the neighbours who send control messages for

the communication establishment. If certain level of stability exists, the

cluster member becomes a CH, and then the selection of new clusterhead

triggers the reclustering procedures and updates the new events in all routing

tables.

2.10 EFFECT OF RWP MOBILITY MODEL ON CLUSTER

COUNT

The behavior of RWP mobility model by applying proposed and

existing algorithms for various vehicle counts is as follows. Even though

RWP model is easy to simulate, it has some unrealistic assumptions about

node movements like sharp turns and sudden stop. Sharp turns occur

whenever a node changes its direction after traveling for a random amount of

time and sudden stops occur when the node decides to stop at a particular time

instant. During a direction change, the speed chosen by a node is totally

independent of the previous speed. In Lowest-ID algorithm, nodes with the

lowest ID number become the clusterhead. In RWP, initially there is more

competition about the election of clusterhead and only small amount of

clusters are created. Later on whenever the vehicle count increases more

number of nodes are involved in cluster creation process leading to increase in

number of clusters. In MOBIC, cluster creation considers the relative mobility

of nodes. A group may consist of clusters that have similar mobility

Page 26: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

40

characteristics. Several groups can be merged into one group depending on

the mobility of each group. A node with higher relative mobility is more

prone to be unstable and therefore this node should not be elected as a

clusterhead. Among the group of vehicles the vehicle with lowest relative

speed becomes the clusterhead. At RWP mobility conditions its difficult to

group the vehicles with relative speed when less vehicles are available.

Whenever the vehicle count increases, we can group the vehicles with the

relative speed and hence more number of clusters is created.

In MCC, the node having maximum connection is chosen as a

clusterhead. The random distribution of less number of vehicles in a

simulation environment creates only less number of clusterheads initially. The

random behavior decreases the clusterhead’s lifetime in MCC is more than

Lowest-ID and MOBIC as the degree of connectivity changes very rapidly. In

DDCA, the idea is to dynamically partition the network into nonoverlapping

clusters of nodes consisting of one parent with zero or more children. To join

a cluster, a node is expected to survive for a period of time ‘t’ with a

probability of at least ‘ ’. Initially more number of clusters are created

dynamically even at random situation. However when vehicle count increases

the maximum number of hops between any pair of nodes in the same cluster

varies dynamically depending on the mobility characteristics of the nodes

leading to decrease in the number of clusters. In CCA, at random mobility

conditions initially more clusters are created based on the predefined static

and dynamic clusterhead. After dynamic clusterhead election there is a

decrease in cluster count. Once the election process is over, CCA forms

constant number of clusters throughout the simulation. Even when vehicle

count increases it will produce constant number of clusters, avoiding

overhead. The number of reaffiliation in CCA increases with increased

number of vehicles and it decreases with further increase of vehicles as the

Page 27: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

41

nodes tends to stay inside the transmission range of clusterhead despite their

random motion.

2.11 EFFECT OF RPGM MOBILITY MODEL ON CLUSTER

COUNT

The behavior of RPGM mobility model by applying proposed and

existing algorithms for various numbers of vehicles is as follows. RPGM

cannot be used as a basis for cluster formation, but can be useful for

predictive group mobility management. In RPGM, each group has a logical

“center” and the center’s motion defines the entire group’s motion behavior

including location, velocity, acceleration etc. If a node with a Lowest_ ID

happens to be highly mobile, it will cause severe re-clustering when it moves

into the transmission range of other clusterheads. MOBIC forms 1-hop

clusters with the relative speed of vehicles resulting in more reclustering in

initial phase. Whenever vehicle count increases there is a competition among

them to become the clusterhead. Based on the nodes with relative mobility,

with increase in vehicle count the number of clusters decreases. MOBIC does

not perform as well as Lowest_ID for vehicle count increase, because clusters

break up frequently, and the neighbour set changes frequently at the time of

group formation. It outperforms as vehicle count increases and the groups are

fixed.

In MCC, the nodes that have more number of neighbours are

elected as clusterheads. Initially less number of clusters is created during the

cluster formation phase; once the clusterhead is selected the further movement

of nodes along with the clusterhead will increase the number of clusters even

when the vehicle count increases. MCC builds 1-hop clusters with the

participation of clusterheads. During group behavior at peak time, pause times

exist between the vehicles resulting in increase in cluster stability for MCC.

DDCA forms less number of clusters but there is always a need to keep more

Page 28: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

42

cluster member information leading to more reaffiliations. It does not require

pause time for initial cluster formation, which results in decrease in the

number of clusters if vehicle count increases. In CCA, clusterhead update is

very limited and hence brings no ripple effect. When there is no cluster to

join, it forms a new cluster. Hence, stable cluster structure is achieved thus

reducing the clustering overhead. The average number of clusters in CCA

decreases with increase in number of vehicles as it simply results in a

different configuration irrespective of its mobility. Thus, the role of the

clusterhead in CCA is retained as long as until simulation ends and its lifetime

are higher than others.

2.12 SIMULATION AND RESULTS

The simulation experiments are conducted by using Ns-2 simulator

with the objective to evaluate the efficiency of reclustering in CCA against

existing well known algorithms; Lowest-ID (LID), Mobility Metric Based

Clustering (MOBIC), Maximum Connectivity Clustering (MCC) and

Distributed Dynamic Clustering Algorithm (DDCA). The accounts are taken

at different possible mobility and topology conditions in order to analyze the

behaviour of the algorithms under these conditions. To this end, the input

parameters that were studied in the simulation were network density, speed

and direction variations, initial node deployment pattern, the user mobility

pattern, the radio transmission range and the packet transmission ratio.

The Ns-2 simulation model simulates nodes moving in an open

plane (Marc Greis 1995). NSG-2 tool is used to generate different scenarios

for Ns-2 simulator (Wu 2007). Motion follows the RWP and PRGM model,

source and relay nodes chooses a destination uniformly at random in the

simulated region, chooses a velocity uniformly at random from a configurable

range, and then moves to that destination at the chosen velocity. The nodes

are initially placed randomly in a rectangular region. Routing protocols

Page 29: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

43

AODV and DSDV are considered. Simulations are done for networks with

50, 100,200 and 500 vehicles with 802.11 radios, with a nominal 250-meter

range. Table 2.2 summarizes parameters that are considered for CCA

simulation.

Table 2.2 Simulation parameters for CCA

Parameter Value

Number of nodes 50-500

Routing protocol AODV and DSDV

Mobility Model Random Waypoint and RPGM

Data flow CBR

Data packet size 512 bytes

Node placement Random

Terrain Area 2000m 2000m

Simulation time 2000s

In the experiments the mobility was simulated with the random

waypoint model and Reference Point Group Mobility Model. The nodes were

simulated to travel with an average speed in the range between the low

pedestrian speed of 5 km/h and the high vehicular speed of 120 km/h.

Initialization of VANET is done by deploying the nodes both randomly and in

groups. In the random deployment model the node mobility follows the

exponential distribution. Transmission range can be varied upto 250m.

However, if the transmission range is less, there are many clusters with

minimum overlapping instead of covering the entire area by a single cluster

resulting in overlapping of clusters with large membership.

Page 30: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

44

N um ber of Vehicles

50 100 150 200 250

Nu

mb

er

of

Clu

ste

rs

0

5

10

15

20

25

30

CCA

Lowest-ID

M OBIC

M CC

DDCA

Figure 2.4 Number of clusters created under RWP mobility model

Figure 2.4 compares the number of clusters that were created by the

proposed and existing algorithms for various numbers of vehicles. CCA is the

most efficient with respect to this metric creating considerably same number

of clusters when the radio transmission covered 40 meters and more than

Lowest-ID, MOBIC, MCC (Maximum Connectivity Clustering), DDCA

(Distributed Dynamic Clustering Algorithm) and CCA. Cluster construction

can be performed parallel in the whole network. In Lowest-ID algorithm,

nodes with the lowest ID become the clusterhead. Election of clusterhead

competition increases whenever the vehicle count increases. Lowest-ID need

pause time of nodes for initial cluster formation to guarantee the accurate

information exchange between the neighbourhood node but in random

waypoint model all the vehicles are dynamic ones. In MOBIC, among the

Page 31: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

45

group of vehicles the lowest relative speed of the vehicle becomes the

clusterhead. In dynamic mobility situations vehicles are distributed with

different speed at different times, so the number of clusters also increases.

MOBIC outperforms Lowest-ID by forming larger size clusters for medium to

high values of transmission range. In MCC, the node having maximum

connection is chosen as a clusterhead. However, cluster overhead increases at

the time of election resulting in increase in the number of clusters moderately

if vehicle count increases. Random behaviour decreases the clusterhead life

for MCC than Lowest-ID since the degree of connectivity changes very

rapidly, thereby reducing its lifetime but in the case of Lowest-ID clusterhead

change is not as frequent as in MCC. MCC performs worse than the Lowest-

ID algorithm when the number of vehicles is increased.

DDCA forms large size clusters at low mobility conditions and

cause table updates whenever topology changes occurs resulting in overhead

increase. During initial cluster formation DDCA does not require pause time.

The overhead and pause time leads to generate lesser number of clusters

throughout the process. In CCA, predefined static and dynamic clusterhead

forms constant number of clusters throughout the simulation. Even when

vehicle count increases it will produce constant number of clusters leading to

lesser overhead. In CCA, average number of clusterhead remains the same

even when the number of vehicle increases. Thus, the clusterhead change rate

is moderate than the existing algorithms. The number of reaffiliation in CCA

increases with increased number of vehicles and it decreases with further

increase of vehicles as the nodes tends to stay inside the transmission range of

clusterhead despite of their random motion. In CCA, a non-overlapping

cluster structure can be achieved with the introduction of predefined static and

dynamic clusterhead. This can reduce the number of small unnecessary

clusters. In CCA clusterhead update very limited and hence brings no ripple

effect.

Page 32: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

46

N um ber o f Veh icles

50 100 150 200 250

Nu

mb

er

of

Clu

ste

rs

5

10

15

20

25

30

CCA

Lowest-ID

M O BIC

M CC

DDCA

Figure 2.5 Number of clusters created under RPGM mobility model

Figure 2.5 shows the number of clusters that were created by the

existing and proposed clustering algorithms under RPGM mobility model for

various numbers of vehicles. In group mobility behaviour, if the number of

vehicles increases at the peak time then the selection of clusterhead becomes

a competition in Lowest-ID, leading to decrease in number of clusters.

MOBIC forms 1-hop clusters with the relative speed of vehicles results in

more reclustering during simulation starts. Whenever vehicles count increases

there is a competition occurring between them to become clusterhead, this

decreases the number of clusters. Although MOBIC does not perform as well

as Lowest-ID when vehicle count increases, because clusters are broken up

frequently, and the neighbour set changes frequently at the time of group

formation. It does outperform the later when vehicle count increases. In MCC,

the clusterhead is elected based on the nodes having more number of

Page 33: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

47

neighbours. MCC builds 1-hop clusters with the participation of clusterheads.

MCC needs pause time of motion of nodes for initial cluster formation to

guarantee the accurate information exchange between the neighbourhood

nodes. During group behaviour at peak time, pause time exist between the

vehicles resulting in cluster stability increase for MCC. DDCA forms clusters

with large size and need to keep more cluster member information leads to

more reaffiliations. It does not require pause time for initial cluster formation

resulting in decrease in number of clusters if vehicle count increases

(Sharmila John and Blessing Rajsingh 2008). In CCA, a non-overlapping

cluster structure can be achieved with the introduction of predefined static and

dynamic clusterhead. This can reduce the number of a small unnecessary

cluster.

In CCA, clusterhead update is very limited and hence brings

minimal ripple effect. If there is no cluster to join, it forms a new cluster.

Hence, stable cluster structure is achieved and minimized clusterhead count

can reduce the clustering overhead. The number of reaffiliations is less when

compared to other algorithms as its cluster members are at 2-hops from their

static and dynamic clusterhead. The clusterheads are nodes tend to stay inside

the clusters for longer time. The average number of clusters in CCA decreases

with increase in number of vehicles as it simply results in a different

configuration irrespective of its mobility. Thus, the role of the clusterhead in

CCA is retained as long as the simulation ends and it has a longer lifetime

than others.

Page 34: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

48

V ehic le speed (km /h)

20 40 60 80 100 120

Clu

ste

rhe

ad

ch

an

ge r

ate

0

1

2

3

4

5

6

7

CC A

Lowest-ID

M O B IC

M C C

DD C A

Figure 2.6 Clusterhead change rates at various vehicle speeds under

RWP mobility model

Figure 2.6 shows the clusterhead change rate at various vehicle

speeds under RWP mobility model. An experimental result shows the CH

change rates for Lowest-ID, MOBIC, MCC, DDCA and CCA when nodes

were placed randomly in the field. The average clusterhead lifetime is lower

in Lowest-ID, since the degree of connectivity changes rapidly for the

vehicles and its clusterhead change also occurs very frequently, thereby

reducing its average lifetime whenever the vehicle speed increases. The nodes

tend to stay inside its cluster for a longer time at low mobility situations,

thereby increasing clusterhead change rate. In MCC the connectivity of the

clusterhead changes more rapidly than Lowest-ID. Thus the number of

reclustering is higher than Lowest-ID. The stability of the clusters in MCC

increases only when the pause time of the nodes increase which is impossible

Page 35: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

49

in real life scenarios. MOBIC considers the vehicles with low relative speed;

cluster count increases throughout the area whenever the speed of the vehicles

is less. Competition occurs between them during low mobility condition and

hence the clusterhead change rate also increases initially. MOBIC forms large

size clusters vehicle speed is high. DDCA does not require pause time on

motion. It creates more number of clusters at high mobility condition leading

to increase in clusterhead change rate from the beginning of the simulation

itself. CCA provides higher stability than other existing algorithms as it

outperforms in terms of clusterhead lifetime, reclustering and clusterhead

change metrics by the introduction of predefined static and dynamic

clusterhead. CCA may be more feasible for somewhat congested places,

where vehicles are highly connected.

Vehicle Speed (km /h)

20 40 60 80 100 120

Clu

ste

rhe

ad

ch

an

ge r

ate

1

2

3

4

5

6

7

CCA

Lowest-ID

MO BIC

MCC

DDCA

Figure 2.7 Clusterhead change rates at various vehicle speeds under

RPGM mobility model

Page 36: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

50

Figure 2.7 shows the clusterhead change rate at various vehicle

speeds under RPGM mobility model. In Lowest-ID, initially at low mobility

conditions, because of continuous election of clusterhead more numbers of

clusters are formed resulting in increased clusterhead change rate. Compared

to Lowest-ID, MOBIC will generate more number of clusters at low mobility

conditions and hence the rate of change is higher. In group mobility all the

vehicles are migrate from one place another to another along with their

neighbours. Initially, at low mobility conditions there is more connectivity

between the vehicles and hence clusterhead change rate is also high in MCC.

In MCC, the change of rate decreases whenever the speed of vehicles

increases because the link between them is less compared to low mobility

conditions. DDCA does not require the static assumption at the time of initial

cluster formation, because it utilizes the mobility behaviour of nodes to decide

the relative speed or path available resulting in increase of clusterhead change

rate. In CCA, the constant number of static and dynamic clusterhead leads to

stable cluster formation resulting in clusterhead change rate.

Overhead occurring at various vehicle densities under RWP

mobility model is shown in Figure 2.8. In Lowest-ID, each node broadcasts

hello messages periodically to participate in the network. Every node of

cluster radius ‘R’ retransmits each hello message. Thus, the drawback of

Lowest-ID algorithm is that certain nodes processing power is decreased due

to serving as clusterhead for longer period of time. There is a maximum

number of temporary clusterhead elections at high-density conditions leading

to increase in overhead. In MOBIC, each node in the cluster formation phase

needs two more hello messages to calculate its mobility metric, which results

in longer cluster formation duration and hence increased overhead.

Page 37: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

51

Vehicle density

100 200 300 400 500

Ove

rhead

(%

)

2

4

6

8

10

12

14

16

CCA

Lowest-ID

M OBIC

M CC

D DC A

Figure 2.8 Overheads at various vehicle densities under RWP mobility

model

In MCC, the neighbours of a clusterhead become members of that

cluster and can no longer participate in the election process. Since no

clusterheads are directly linked, only one clusterhead is allowed per cluster.

As the number of nodes in a cluster is increased, the throughput drops. As the

density of the cluster increases, reclustering is done and hence overhead

increases. In DDCA, any node that possesses a low relative speed than its

neighbours has the ability to become a CH. This operation requires a lot of

pair wise communication before the decision is made. This leads to increase

in overhead if the number of vehicles increases. In CCA, the election of slow

speed vehicle shares the information with other vehicles and static clusterhead

in random scenario. Reclustering process is reduced by the election of

predefined time chart vehicles which also decreases the hello messages count

even if vehicle count increases.

Page 38: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

52

Vehicle density

100 200 300 400 500

Overh

ead

(%

)

2

4

6

8

10

12

14

16

18

CCA

Lowest-ID

MO BIC

MCC

DDCA

Figure 2.9 Overheads at various vehicle densities under RPGM

mobility model

Figure 2.9 shows overhead at various vehicle densities under

RPGM mobility model. In Lowest-ID, the node with lowest ID in its k-hop

neighbourhood becomes the clusterhead. In Lowest-ID algorithm certain

nodes are prone to power drainage due to serving as clusterheads for longer

periods of time. In RPGM model, reclustering in Lowest-ID indicates that a

clusterhead movement may still invoke the complete cluster structure re-

computation leading to overhead increase. In MOBIC, the frequent

reclustering results in lower throughput and longer delay. In MOBIC, all the

decisions are based on the group reference member, the frequent reclustering

and the corresponding clusterhead election results in increase in the cluster

maintenance overhead. In MCC, the node with maximum number of

neighbours is chosen as a clusterhead. Each node either becomes a

Page 39: CHAPTER 2 PROPOSED CLUSTERING METHODshodhganga.inflibnet.ac.in/bitstream/10603/9844/7/07_chapter 2.pdf · hierarchical clustering method. There can be one or many number of static

53

clusterhead or an ordinary node. This system has a greater rate of clusterhead

change but the throughput is low. Whenever the vehicle count increases the

reaffiliation count of nodes is high due to node movement and as a result, the

current clusterhead may not be re-elected.

At high-density conditions the reclustering in MCC may cause

large communication overhead in the network, when there is frequent CH

disconnects in the cluster architecture. In DDCA, group movement of nodes

in urban area creating and maintaining cluster structures within an adhoc

network comes with additional communication and computation costs.

Cluster related information is exchanged rapidly; causing higher bandwidth

consumption and reduced network performance leading to overhead increase.

However in CCA, all the nodes become a member of either static or dynamic

clusterhead. During high mobility, if any node does not become member of

cluster, it declares itself as a clusterhead and forms a cluster around its

transmission range until a predefined clusterhead comes within its range. This

will reduce the number of reclustering and minimize overhead.

In this chapter the performance of hierarchical clustering-Cluster

Construction Algorithm (CCA) is discussed in detail. Overheads and

clusterhead change rates are reduced by the combination of static and

dynamic clusterheads. Various simulations are carried out and results are

compared with the existing algorithms. It is observed that the CCA algorithm

performs well under different mobility conditions. In the next chapter the

importance of the selection of proper mobility model is discussed.