KNOWLEDGE MANAGEMENT USING CLUSTERS IN VANETS

Description, Simulation and Analysis

C. Caballero-Gil, P. Caballero-Gil and J. Molina-Gil

Department of Statistics, Operations Research and Computing, University of La Laguna, La Laguna, Tenerife, Spain

Keywords:

VANET, Cluster, Knowledge management, Wireless ad-hoc networks.

Abstract:

Vehicular Ad Hoc Network (VANET) is a type of mobile network that allows Knowledge Management of

the road providing communication among nearby vehicles and between vehicles and nearby ﬁxed roadside

equipment. The lack of centralized infrastructure and high node mobility and number of vehicles gener-

ate problems such as interrupting connections, difﬁcult routing, security of communications and scalability.

Clusters are here proposed as a solution to avoid data collisions by decreasing the number of connections

exchanged among vehicles and to reach this goal, nodes must cooperate, and form or join clusters depending

on their state. This paper provides a global vision of the life cycle of cooperative nodes who form clusters and

a description of how to deal with the information within a cluster. Simulation results show that the proposed

scheme reduces the number of communications, avoiding data loss due to collisions. This paper provides a

full description of each cluster management process and of how to deal with the information within a cluster.

1 INTRODUCTION

A VANET is a spontaneous Peer-To-Peer (P2P) net-

work formed by moving vehicles. As any other

MANET (Mobile Ad-hoc NETwork), a VANET has

no central infrastructure, which implies the need of

self-management in a distributed environment where

nodes have to adapt to unpredictable changes. Such

autonomic networks present unique challenges such

as high mobility, real-time constraints, scalability,

gradual deployment and privacy.

Intelligent VANETs hybridly integrate multiple

P2P networking technologies such as WiFi IEEE

802.11 b/g, WiMAX IEEE 802.16, Bluetooth, etc. to

achieve effective wireless communication. Such net-

works constitute a fundamental part of the Intelligent

Transportation System (ITS). Different possible situ-

ations exist where Knowledge Management between

vehicles would help to prevent accidents and to avoid

trafﬁc jams, which would save time and money, re-

duce contamination of the environment and consump-

tion of fuel reserves.

Several characteristics can be considered in wire-

less networks: authenticity, privacy, anonymity, coop-

eration, low delay, stability of communications, scal-

ability, etc. (Caballero-Gil et al., 2009), (Raya et al.,

2006) However, when dealing with VANETs, the pro-

tection of those properties is an even greater problem

due to speciﬁc characteristics of these networks, such

as very changing scenarios, from local roads with

very few vehicles to cities or highways full of vehi-

cles. In this work we propose the use of clusters in

VANETs, which will allow to optimize communica-

tion in dense trafﬁc situations.

2 CLUSTERS

A cluster in a VANET is deﬁned as a set of vehicles

that are located in a close geographic area whose for-

mation is determined by the mobility pattern of vehi-

cles. The cluster needs a minimum of vehicles and is

managed by a given node called ”leader of the clus-

ter”. All vehicles forming part of a cluster have a

direct wireless connection with the leader of such a

cluster and share a secret key.

There are several bibliographic references that

propose the use of clusters in VANETs. (Fan et al.,

2008) presents a theoretical analysis of a directional

stability-based clustering algorithm. (Gunter et al.,

2007) describes clusters where the leader is the node

in the middle with the lowest identiﬁer. (Rawashdeh

and Mahmud, 2008) proposes a different scheme of

clusters to maximize the advance of the relayed in-

formation and to avoid interferences. None of these

170

Caballero-Gil C., Caballero-Gil P. and Molina-Gil J..

KNOWLEDGE MANAGEMENT USING CLUSTERS IN VANETS - Description, Simulation and Analysis.

DOI: 10.5220/0003070501700175

In Proceedings of the International Conference on Knowledge Management and Information Sharing (KMIS-2010), pages 170-175

ISBN: 978-989-8425-30-0

 2010 SCITEPRESS (Science and Technology Publications, Lda.)

works deﬁne in detail the processes that nodes have

to complete for cluster management and do not show

any implemented scheme to demonstrate the relia-

bility of obtained data, (Caballero-Gil et al., 2007),

which is a priority in this work.

Clusters will be used only when the conditions

of the routes require it. Examples are dense trafﬁc,

trafﬁc jams or congested highways, where the den-

sity of vehicles in a geographic zone causes that the

number of communications is huge. But clusters are

formed before the number of nodes begins to degrade

the network. Without any mechanism to minimize

the number of communications, a simple broadcast

will be launched from every vehicle generating a lot

of unnecessary redundancy. The number of packets

generated depends on the number of nodes in the net-

work and interconnection among them. Therefore, it

will be generated n packets for each data communica-

tions where n is the number of vehicles with On Board

Unit (OBU) in the network (in the scope of interest).

Some studies like, (Dousse et al., 2003) show that this

number of connections causes collisions in the infor-

mation that is sent, which degrades communication

quality.

3 MESSAGE MANAGEMENT

INSIDE CLUSTERS

By using clusters the number of communications can

remarkably decrease without missing any useful in-

formation.

Algorithm 1 shows the steps that a vehicle bel-

oging to a cluster must follow in order to process an

input signal.

If the node is the ﬁnal destination, it simply pro-

cesses the information. Otherwise, it checks whether

data were sent by the cluster leader. In particular,

the leader can send two types of packets towards any

member of the cluster that is not the ﬁnal destination

of the data:

• A connection of a vehicle to Internet services, or

any other supplied service where it is necessary a

relay of an information sequence,

• A packet of other type of information that must be

forwarded towards other parts of the network.

With respect to this second type of packets there

are two types of communications that must be differ-

entiated:

• safety-related information

• commercial advertising

Algorithm 1: Message Management inside Clusters.

01:function MessageManagement (...)

02: if(AmIﬁnalDestination(packet)) then

03: TreatData(packet);

04: else

05: if (AmILeader()) then

06: if (IsPublicInformation(packet)) then

07: TreatData(packet);

08: Multicast(packet, ClusterKey);

09: else

10: relayer =

10: estimatePosition(DestinationNode);

11: Unicast (packet, relayer);

12: end

13: else

14: if(IsForwardingSequence(packet))then

15: relayer =

15: estimatePosition(DestinationNode);

16: Unicast (packet, relayer);

17: end

18: else

19: if(IsSentbyLeader(packet))then

20: relayer =

20: estimatePosition(DestinationNode);

21: Unicast (packet, relayer);

22: end

23: else

24: Unicast (packet, ClusterLeader);

25: end

26: end

27: end function

In both cases the vehicle belonging to the cluster

that receives or produces the communication, sends it

to the cluster leader who will forward it to all con-

nected members of the cluster and towards the zones

where the message has not been yet spread.

An Internet connection with a RSU (Road Side

Unit) can be relaying through another clusters us-

ing intermediate nodes who forward the information.

For these types of communications, mechanisms for

enforcement cooperation, like in (Molina-Gil et al.,

2009) are necessary because without them, intermedi-

ate vehicles would not have the necessary incentives

to relay others connections, what would disable any

type of service that incorporates an indirect connec-

tion with the RSU.

4 CLUSTERS STAGES

We distinguish among several stages in cluster man-

agement, corresponding to different situations of ve-

hicles, depending on the route and on their status in

KNOWLEDGE MANAGEMENT USING CLUSTERS IN VANETS - Description, Simulation and Analysis

171

each moment. The stages are: Detection, Election,

Creation, Membership and Life of a cluster.

VANETs are wireless networks where there are a

large number of highly volatile connections between

vehicles. For this reason it is necessary to deﬁne in

detail the way in which vehicles must act according

to their situation.

The global network life scheme proposed in this

paper is as follows. Initially all nodes start in the

Cluster Detection stage. After this, they can enter the

Creation or the Election stage, depending on the cir-

cumstances. After Cluster Creation, the node would

be the cluster leader, while after Cluster Election, the

node would proceed to Cluster Membership.

4.1 Cluster Detection

This is the ﬁrst stage, where vehicles are in normal

conditions without dense trafﬁc. This stage is de-

scribed in Algorithm 2, where neighbor(i) denotes

the i-th neighbor of the node that initiates the stage.

From time to time the vehicle checks the number of

neighbors and the number of leaders among them. If

there is at least one neighbor who is leader of a clus-

ter, the node proceeds to the Election stage, and oth-

erwise to the Creation stage. This stage does not gen-

erate any trafﬁc of control due to the fact that all the

necessary information is contained in the beacons that

nodes generate.

Algorithm 2: ClusterDetection.

01:function ClusterDetection (...)

02: numberOfNeighbors = 0;

03: numberOfLeaders = 0;

04: while (neighbor(i) exists) do

05: if (isLeader( neighbor(i) )) then

06: numberOfLeaders = 0;

07: end

08: numberOfNeighbors++;

09: i++;

10: end

11: if (numberOfLeaders == 0) then

12: ClusterCreation();

13: else

14: ClusterElection();

15: end

16: end function

4.2 Cluster Election

This stage starts when the vehicle has found among

its neighbors at least one node that is leader of some

cluster. If there is only one neighbor who is a clus-

ter leader, the choice is automatic. Otherwise, if there

are several leaders, the vehicle has to choose one of

them to join it. Algorithm 3 shows this stage, where

clusterValue denotes a quantity used for the choice

and clusterLeader( j) represents the j-th neighbor of

the node that is leader of a cluster. If there are sev-

eral leaders among its neighbors, the vehicle chooses

one according to the clusterValue that depends on the

following values for each cluster j:

• Density A( j) of vehicles.

• Average quality of signal B( j) in dBm within the

vehicles.

• Time C( j) during which it has been connected to

the leader.

Algorithm 3: ClusterElection.

01:function ClusterElection (...)

02: if (numberOfLeaders ≥ 1) then

03: j = 1;

04: e = 0;

05: clusterVal[e] = 0;

06: while (clusterLeader[j] exists) do

07: clusterVal[j] = A(j)+B(j)+C(j);

08: if(clusterVal[j] ≥ clusterVal[e]) then

09: clusterVal[e] = clusterVal[j];

10: end

11: j++;

12: end

13: else

14: e = 1;

15: endif

16: sendRequest (clusterLeader[e]);

17: receiveClusterKey(clusterLeader[e]);

18: ClusterMembership();

19:end function

Once the cluster has been chosen, the vehicle

sends a login request encrypted with its public key to

the cluster leader. After authenticating it, the leader

sends the cluster secret key encrypted with such a

public key and from then, the vehicle becomes part

of the cluster.

4.3 Cluster Creation

In the Cluster Creation stage (Algorithm 4), the ve-

hicle is not close to any leader of a cluster. It should

check whether within their neighbors there are at least

X nodes that do not belong to any cluster, plus a vari-

able Y that indicates the number of vehicles that can

either turn off, separate or not join the new cluster that

is being created. If the number of neighbors without

cluster is lower than the minimum threshold required

for cluster creation, the vehicle waits a period time1

KMIS 2010 - International Conference on Knowledge Management and Information Sharing

172

and starts again the Cluster Detection stage. Other-

wise, if the number of neighbors is greater than the

threshold X +Y , the vehicle begins a new Cluster Cre-

ation process. In order to do it, it multicasts a cluster

creation request towards all neighbors with distance

equal to 1. Nodes that receive this request respond

accepting or rejecting the invitation. If the number of

neighbors that accept the invitation is greater than the

minimum threshold X, the new cluster leader sends to

each node the secret key of the cluster encrypted with

the public keys of each node. In this moment the new

cluster is formed. Otherwise, the number Y of esti-

mated vehicles is increased by adding the number of

vehicles that did not accept the invitation.

Algorithm 4: ClusterCreation.

01:function ClusterCreation (...)

02: if (numberOfNeighbors ≥ X + Y) then

03: AcceptedNeighbors = 0;

04: n = 1;

05: l = 0;

06: MulticastNeighbors (NeighborsList[]);

07: for(n=1;n≤ numberOfNeighbors;n++) do

08: ReceiveClusterElection(n);

09: if (neighbor(n) accept) then

10: acceptedNeighbors(l)=neighbor(n);

11: l++;

12: acceptedNeighbors++;

13: end

14: end

15: if (acceptedNeighbors ≥ X) then

16: for(n=1;n≤ acceptedNeighbors;n++) do

17: SendClusterKey(acceptedNeighbors(n),

18: PuKacceptedNeighbors(n));

19: end

20: ClusterLife();

21: else

22: Y = Y + X - acceptedNeighbors;

23: Wait(time1);

24: ClusterDetection();

25: end

26: else

27: Wait(time1);

28: ClusterDetection();

29: end

30:end function

In conclusion, this stage requires: a multicast of

invitation to join the new cluster, unicast responses

from n users and a multicast to relay a message that

enables the members to build the cluster secret key.

This means a total of 2n + 1 packets in case of posi-

tive cluster creation, and n+2 if the process fails. The

Cluster Creation starts when the appropriate number

of neighbors reaches a certain threshold of trafﬁc, but

without to be dense trafﬁc. Consequently, manage-

ment packets generated at this stage not increase com-

munications in dense trafﬁc conditions.

4.4 Cluster Membership

Once the cluster is formed, the leader must periodi-

cally validate that the cluster continues being useful.

Otherwise, it would be necessary to change the leader

or to end the cluster.

Algorithm 5 shows the process where a node

leaves its cluster. When the node loses any contact

with the leader of the cluster for certain time, the

node stops to belong to its cluster and begins the Clus-

ter Detection stage if node density exceeds the corre-

sponding threshold.

Algorithm 5: ClusterMembership.

01:function ClusterMembership (...)

02: if (See( clusterLeader )) then

03: Wait(timeMemberCheck);

04: ClusterMembership();

05: else

06: Wait(timeReCheck);

07: if (See( clusterLeader )) then

08: Wait(timeMemberCheck);

09: ClusterMembership();

10: else

11: ﬁnalClusterMembership();

12: ClusterDetection();

13: end

14: end

15: end function

4.5 Cluster Life

Algorithm 6 shows how the leader of a cluster period-

ically checks that the cluster is still useful. If cluster

size falls below a certain threshold, the leader checks

whether it has a number of neighbors greater or equal

to D (dense trafﬁc threshold) and waits for time2 in-

stead of ending the cluster in order to avoid introduc-

ing cluster management trafﬁc when the vehicle is in

a dense trafﬁc situation. If the leader is not in a dense

trafﬁc situation, it begins a leader change or a clus-

ter ending process. First, the leader asks about the

neighborhood density in order to know if neighbor-

hood density (number of neighbors of the same clus-

ter or without any cluster near) is bigger than X. It

also ﬁnds out which of its neighbors has the largest

number of neighbors. After this, it sends a multicast

signal of leader change to all its neighbors. The new

leader will begin a Cluster Creation stage with those

nodes without any cluster that are in its transmission

KNOWLEDGE MANAGEMENT USING CLUSTERS IN VANETS - Description, Simulation and Analysis

173

range. In the absence of any neighbor exceeding the

threshold, the leader sends the cluster ending signal

through multicast to all its neighbors.

Algorithm 6: ClusterLife.

01:function ClusterLife (...)

02: for(n=1;n≤ numberOfNeighbors;n++)do

03: if(Belongs(neighbor(n),cluster(a)))then

04: clusterSize++;

05: end

06: end

07: if (clusterSize ≥ X) then

08: Wait(time2);

09: ClusterLife();

10: else

11: if (numberOfNeighbors ≤ D) then

12: newLeader=0;

13: for(n=1;n≤ numberOfNeighbors;n++)do

14: //clusterSize+withoutCluster

14: pot = potential(neighbor(n));

15: if((pot≥ X)and

15: (pot≥ clusterSize))then

16: clusterSize=clusterSize(n);

17: newLeader=n;

18: end

19: end

20: if (newLeader == 0) then

21: Multicast (End-of-Cluster-Signal);

22: ClusterDetection();

23: else

24: Multicast (Leader-Change-Signal);

25: //New leader init ClusterCreation proccess

26: ClusterDetection();

27: end

28: end

29: end

30:end function

5 SIMULATION AND ANALYSIS

Both the feasibility and effectiveness of our approach

are shown through the ﬁgures where a simulation ex-

empliﬁes its performance. In the ﬁrst part of our

demonstration (Figure 1), a NS-2 and SUMO display

shows the VANET state in one moment when clusters

are operating. The most relevant options selected for

the demonstration have been: Total number of vehi-

cles: 80, number of vehicles with OBUs: 80, number

of lanes for each direction: 3 and 3, simulation time:

100 seconds, moment when retransmissions begins:

40 seconds, retransmission period: 10 seconds, dis-

tance relay nodes: 75 meters, traveled distance before

the trafﬁc jam happens: 800 meters.

Figure 1: Simulation.

The implemented simulations for clusters con-

sider four levels of development: vehicle mobility,

node energy, cluster formation and P2P communica-

tions in the network.

• The vehicle mobility layer manages the node

movement in the movement pattern, which deﬁnes

roads, lines, different speed limits for each line,

trafﬁc jams, etc.

• The node energy layer is used to distinguish be-

tween vehicles with and without OBUs. Vehicles

without OBUs are present in the road but do not

contribute in the communications.

• The cluster formation layer deﬁnes which vehi-

cles belong to each cluster, who is the leader

of each cluster, who generates trafﬁc information

and who relays information to other clusters.

• The P2P communications layer is responsible for

the deﬁnition of which nodes are in the transmis-

sion range of the retransmitting node at any time.

Statistics Extraction. Simulations give essential

statistics such as number of generated, dropped or lost

packets or bytes. These basic statistics data are useful

to make efﬁcient simulations for large scale scenarios.

Two Implementation Mechanisms. Simulations

provide two mechanisms to implement VANETs:

One with clusters and the other without them, and

they can be compared with the same topology (see

Figure 2).

Figure 2: Generated packets without and with Clusters.

Among the obtained information from the simula-

tions we have the number of packets and bytes gen-

KMIS 2010 - International Conference on Knowledge Management and Information Sharing

174

erated, sent, broadcast, received, lost, etc. for each

node. Also, other general information shown is the

number of packets generated or lost in the whole net-

work, the number of formed clusters, which nodes are

the leaders of the clusters, which nodes generate pack-

ets and which nodes forward them, etc. In addition to

all this information, another interesting aspect is that

it provides a detailed simulation of what happens in

each moment in the VANET thanks to the use of the

NS-2 display. It also shows the trafﬁc model through

the SUMO tool while the information is represented

using TraceGraph. Table 3 shown some result of sim-

ulations. We have chosen the same set of common

parameters previously mentioned.

Figure 3: Generated and lost packets.

We can observe in Figure 3 the comparison be-

tween the average generated and lost packets: it is

clear that, without the use of clusters in VANETs,

the number of generated packets grow up much faster

than with the use of clusters. But also the lost pack-

ets grow up much faster. The main reason is likely to

be the heaviest trafﬁc load that VANETs generates in

trafﬁc jams conditions: indeed, the original protocol

makes a massive use of broadcast operations. Clus-

ters will help to decrease the percentage of lost pack-

ets and to perform the VANETs operation.

6 CONCLUSIONS

In this paper the use of clusters has been proposed

as a solution to decrease the number of communica-

tions in VANETs under dense trafﬁc conditions when

the overhead of transmitted data causes a consider-

able drop in communication quality. In particular, a

complete description of the proposed scheme for au-

tonomic cluster management in VANETs is provided,

which includes differentiation among possible vehicle

states: from the initial state when it does not belong

to any cluster, to the choice of an existent cluster to

join it, the creation of a new cluster, and the end of

a cluster. This paper also shows how to proceed with

cluster communications.

A complete analysis has been done through simu-

lations using the open source trafﬁc simulator SUMO

and network simulator NS-2. Such simulations al-

low the analysis of the operations at each stage, and a

comparison between communication overhead when

using clusters and without using them in VANETs.

ACKNOWLEDGEMENTS

Research supported by the Ministerio de Ciencia e

Innovaci

on and the European FEDER Fund under

Project TIN2008-02236/TSI, and by the Agencia Ca-

naria de Investigaci

on, Innovaci

on y Sociedad de la

Informaci

on under PI2007/005 Project.

REFERENCES

Caballero-Gil, P., Caballero-Gil, C., Molina-Gil, J.

& Hern

andez-Goya, C., 2007 ’A simulation

study of new security schemes in mobile Ad-hoc

NETworks’, Lecture notes in computer science,

EUROCAST:Vol:4739:73-81

Caballero-Gil, P., Caballero-Gil, C., Molina-Gil, J. &

Hern

andez-Goya, C., 2009 ’Flexible Authentication

in Vehicular Ad hoc Networks’, APCC IEEE.

Dousse, O., Baccelli, F. & Thiran, P., 2003 ’Impact of In-

terferences on Connectivity in Ad Hoc Networks’. IN-

FOCOM.

Fan, P., Sistla, P. & Nelson,P. C., 2008 ’Theoretical analy-

sis of a directional stability-based clustering algorithm

for vanets’. Vehicular Ad Hoc Networks.

Gunter, Y., Wiegel, B. & Gromann, H. P., 2007 ’Medium

Access Concept for VANETs Based on Clustering’.

VTC Fall:2189-2193.

Molina-Gil, J., Caballero-Gil, P. & Caballero-Gil, C., 2010

’A Vision of Cooperation Tools for VANETs’, WoW-

MoM.

Rawashdeh, Z. Y. & Masud Mahmud, S., 2008 ’Media Ac-

cess Technique for Cluster-Based Vehicular Ad Hoc

Networks’. VTC Fall.

Raya, M., Papadimitratos, P. & Hubaux, J.P., 2006 ’Se-

curing Vehicular Communications - Assumptions, Re-

quirements and Principles’, ESCAR.

KNOWLEDGE MANAGEMENT USING CLUSTERS IN VANETS - Description, Simulation and Analysis

175