Content uploaded by Abdelhakim Senhaji Hafid
Author content
All content in this area was uploaded by Abdelhakim Senhaji Hafid on Nov 07, 2015
Content may be subject to copyright.
A New Stability Based Clustering Algorithm (SBCA)
for VANETs
Ahmed Ahizoune, Abdelhakim Hafid
Network Research Lab, University of Montreal, Canada
{ahizouna, ahafid }@iro.umontreal.ca
Abstract- Lately, extensive research efforts have been dedicated
to the design of clustering algorithms to organize nodes in
Vehicular Ad Hoc Networks (VANETs) into sets of clusters.
However, due to the dynamic nature of VANETS, nodes
frequently joining or leaving clusters jeopardize the stability of
the network. The impact of these perturbations becomes worse
on network performance if these nodes are cluster heads.
Therefore, cluster stability is the key to maintain a predictable
performance and has to consider reducing the clustering
overhead, the routing overhead and the data losses. In this paper,
we propose a new stability-based clustering algorithm (SBCA),
specifically designed for VANETs, which takes mobility, number
of neighbors, and leadership (i.e., cluster head) duration into
consideration in order to provide a more stable architecture.
Extensive simulations show that the proposed scheme can
significantly improve the stability of the network by extending
the cluster head lifetime longer than other previous popular
clustering algorithms do.
Index-terms: Vehicular Ad Hoc Network, Clustering Stability,
Communications Overhead, Cluster Residence Time.
I. I
NTRODUCTION
New wireless technologies have the potential to enable
inter-vehicle communications with the purpose of crash
avoidance and transportation system efficiency improvement.
Consequently, the Federal Communications Commission
(FCC) of the U.S. approved the 75MHz bandwidth at 5.850-
5.925GHz band, in year 1999, for Intelligent Transportation
System (ITS). This wireless spectrum is commonly known as
the Dedicated Short-Range Communication (DSRC) allocated
by the regulator to be used exclusively for vehicle to vehicle
(V2V) and vehicle to roadside (V2R) communications. Due to
the high infrastructure cost associated with V2R
communications V2V is considered to be a more economical
and practical approach for safety and non-safety information
delivery.
One of the many challenges in VANETs is the dynamic
and dense network topology. The dynamic topology causes
significant re-routing difficulties and thus congestion, while
the dense network leads to the hidden terminal problem. A
clustered structure can make the network appear smaller and
more stable in the view of each node. By clustering the
vehicles into groups of similar mobility, the relative mobility
between communicating neighbouring nodes will be reduced,
leading to intra-cluster stability; in addition, the hidden
terminal problem can be diminished by clustering [9].
Recent ad hoc network research [1, 9, 10] discussing
cluster-based MACs and routing schemes, motivates the need
for a stable VANET clustering scheme. In this paper, we
propose a new stability-based clustering algorithm (SBCA) for
VANETs; SBCA aims to increase significantly the stability
and reliability in VANETs.
The remainder of this paper is organized as follows. In
Section II, we present related work. Section III describes
SBCA. Section IV presents simulation and performance
analysis. Section V concludes the paper.
II. RELATED
WORK
Figure 1 shows a cluster that is composed of 2 Cluster
Heads (CH), one Cluster Gateway (CG), and 6 Cluster
Members (CM). All the mobile nodes within the radio range
of CH are selected CMs of the cluster. A CG is CM that
belongs to more than one cluster; it acts as the communication
gateway between CHs. Sometimes, an additional state called
Undecided State (US) is used for the initial state of a node.
Fan et al. [3] propose a utility-based clustering scheme; a
vehicle chooses its CH based on the values produced by the
utility function after receiving status information from
neighbouring nodes. The node with the highest utility value is
selected. This approach attempts to improve the performance
of classical clustering algorithms by making them aware of the
vehicle’s movement; however, all nodes attempt to re-evaluate
their conditions (computing utility values) at the same time
which may cause traffic increase and therefore consume more
bandwidth.
Node mobility should play an integral part in cluster
creation in order to achieve stability. In [2], mobility is
addressed during clusters’ collisions; when two CHs come
within range, the winning CH will be the one with both lower
relative mobility and closer proximity to its members. The
algorithm used for cluster formation is based on CBLR [6].
Alternatively, Kayis et al. [15] address mobility by classifying
nodes into speed groups, such that nodes will only join a CH
of similar velocity.
Kwon and Gerla [3] proposed a Passive Clustering
algorithm (PC) for on demand creation and maintenance of the
clustering structure which can avoid potential long setup time
and reduce re-forwarding significantly. PC performs well in a
high mobility network where cluster topology changes
frequently. PC is immune from increased control overhead
due to frequent changes in network topology; it is, however,
dependent on traffic to function.
Similarly, several other existing clustering algorithms in
the literature have been proposed for VANETs considering
cluster stability as the design objective. De souza et al. [11]
present a beacon-based clustering algorithm aimed at
extending the cluster lifetime in VANETs; it uses a new
aggregate local mobility criterion to decide cluster
reorganization. The scheme incorporates a contention method
to avoid triggering frequent reorganizations when two CHs
encounter each other for a short period of time. Shea et al. [12]
use the Affinity Propagation algorithm in a distributed
manner; this algorithm determines clusters that minimize both
relative mobility and distance from each CH to its CMs; the
resulting clusters are stable. Fan et al. [13] present a
theoretical analysis of a directional stability based clustering
algorithm. Rawashdeh et al. [10] propose establishing clusters
to maximize the advance of the relayed information and to
avoid interferences; however, they assume that a CH must
know the exact positions of nodes in the cluster which is
difficult to achieve in real life situations.
Fig. 1- A configuration of clusters
Most implementations of these existing algorithms focus
mainly on how CHs are elected. The communication overhead
for the formation and maintenance of clusters have not been
taken fully into account. There has been few contributions that
assess analytically the communication overhead incurred in
hierarchical routing. In particular, Fan et al. [13] show that the
overhead incurred by DISCA [13] is bound by a constant per
node per time step, avoiding expensive re-clustering chain
reactions; hence, this overhead increases with the number of
nodes.
Since a CH acts as a coordinator in a cluster, if it is absent
for any reason, the clustering architecture has to be
reconfigured; this will significantly increase the message
overhead. However, in our research, we believe that a more
efficient way to form a stable clustering architecture, with
reduced overhead, is that a mobile node should be associated
to a cluster and not to a CH. Indeed, replacing CH is
considered only as an incremental update and does not require
a whole reconfiguration of the cluster structure; this will
definitively increase the lifetime of the clustering architecture.
The resulting clusters are stable and exhibit long average CM
duration, long average CH duration, and low average rate of
CH changes. In this paper, we propose a new stability-based
clustering algorithm protocol (SBCA) aiming to reduce the
communication overhead that is caused by the cluster
formation and maintenance, as well as to increase the lifetime
of the cluster. SBCA makes use of (1) The cluster
configuration protocol that is based on the velocities’
differences among neighboring vehicles to select a primary
CH (PCH) for each cluster; and (2) The election of a
secondary CH (SCH) for each cluster; similarly to the
clustering scheme proposed in [5], SCH works as a backup for
PCH.
III. THE
SBCA
PROTOCOL
DESCRIPTION
In this section, we describe how SBCA forms and
maintains stable clustering architecture able to achieve
stability and thus low communication overhead. SBCA
involves two phases: setup and maintenance. In the cluster
setup phase, nodes in close proximity to each other are
organized into clusters with CHs selected. In the cluster
maintenance step, a secondary CH (SCH) is selected for each
cluster; CHs selected in the setup phase become primary CHs
(PCH). The nodes remain associated with a given cluster (and
not CH as in existing approaches); indeed, when a PCH is no
longer in the cluster, SCH takes over; the cluster structure
does not change but only the node playing the role of CH. This
allows for stable cluster architecture, with low overhead, and
thus better performance.
Fig. 2- Illustrated example. (a) Setup phase; (b) Maintenance phase: SCH
selection; and (c) Maintenance phase: SCH becoming PCH
Before discussing the details of SBCA, let us consider an
example that shows how SBCA provides cluster stability and
thus less overhead. Figure 2 (a) shows a cluster with CH and a
number of CMs; this structure can be created, for example,
using CCP [1]. Figure 2 (b) shows a cluster with two CHs:
primary cluster head (PCH) and secondary cluster head
(SCH); PCH corresponds to CH in Figure 2 (a) and SCH is
elected using SBCA. Figure 2 (c) shows that when PCH is out
of the cluster (e.g., slows down to take an exit), SCH takes the
role of PCH and a new SCH is elected. Thus, the original
cluster (Figure 2 (a)) still exists; the only change concerns a
new CH (PCH). This means that CMs do not have to look for
a new cluster (as in the case of existing protocols, such as CCP
[1]) and thus do not need to generate extra packets (overhead)
to perform this action. The only overhead generated by SBCA,
in this case, is the maintenance and selection of SCH.
A. Setup Phase
In the SBCA setup phase, the main activities consist of the
cluster structure creation and the election of CHs. The
operation of SBCA in this phase represents an adapted version
of CCP [1].
Initially, every network node is in Undecided State (US). If
a node receives an invite-to-join (ITJ) message (in this case a
neighboring CH exists; CH sends invite-to-join (ITJ) messages
every tj time units ), it will check the received signal strength
denoted Pr; in the case where Pr is bigger than some
predefined threshold Pth, the node sends a request-to-join
(RTJ) message, to the neighboring CH, including the node’s
ID and the network address to the advertising CH; upon
receipt of the corresponding ACK, the node becomes CM of
this cluster. If a node stays in US more than tj time units (i.e.,
does not receive ITJ message during [t,t+tj]), then the node
becomes CH.
A node remains CM (of a cluster) as long as it receives ITJ
message from its CH every tj time units. If it does not receive
ITJ message during [t, t+2tj], it considers its association, with
its CH, is lost and switches to US. If a CM receives an ITJ
message (with Pr bigger than Pth) from another neighboring
CH, then it switches to CG after sending RTJ and receiving
the corresponding ACK.
A CH broadcasts ITJ messages every tj time units; upon
receipt of a RTJ message, it sends an ACK and adds the
requesting node to its cluster-member list. If CH does not hear
from one of its CMs during [t, t+3tj], it removes this CM from
its cluster-member list. A CH switches to US, if its cluster-
member list becomes empty. If a CH is in the transmission
range of another CH (i.e., both will receive ITJ messages from
each other), only one of them will keep its CH role while the
other(s) will become CM; the CMs of the cluster (whose CH
has just become CM) will switch to US; they will change their
role according to the procedure explained earlier (e.g., they
will all become CMs of the new CH if they are in the
transmission rage of this CH). The decision of CH to give up
or not its role is based on a weighted factor CH* (see Equation
1); this factor represents the minimum of the difference
between the sum of velocity differences between the CH and
its neighboring nodes and the number of neighboring nodes of
this CH. The CH that will keep its role corresponds to CH that
produces the minimum value of this difference.
CH∗
= arg
CHmin α∗∑V
CH − V
i
∈
−β∗
N
neighbors#; (1)
where α+β= 1, i is a CM of the cluster headed by CH,
V
CH is the speed of CH, V
i is the speed of i, and N
neighbors
is the number of neighboring nodes of CH.
B. Maintenance phase
The objective of the maintenance phase is to achieve
stability and reliability (less packet losses and thus better
packet delivery) of the cluster structure produced in the setup
phase. The basic idea is to use two CHs: (a) primary CH
(PCH): it is elected in the setup phase; and (b) secondary CH
(SCH): it is elected in the maintenance phase.
Once a PCH is elected during the setup phase, it generates
a unique identifier, Cluster_ID, for the cluster. Cluster_ID is
computed using the following hash function:
Cluster_ID = Hasht ⊕ PCH_ID (2)
where t is the current time and PCH_ID is the unique
identifier of the primary cluster head.
PCH periodically selects SCH among its CMs; the node
that produces the minimum sum, of velocity difference
between PCH and CM and distance between PCH and CM
(see Equation 3), is selected SCH.
SCH∗
= arg
imin α∗ V
PCH − V
i + β∗ D
i#; (3)
where α+β= 1, i is a CM of the cluster headed by PCH,
V
PCH is the speed of PCH, V
i is the speed of i, and D
i is the
distance between PCH and i.
If PCH can no longer be a CH (e.g., leaving the cluster by
taking a highway exit), it will order SCH to switch to PCH and
change its own state to CM; it will eventually change to
undecided state when it no longer receives ITJ messages. The
new PCH will keep the same identifier (Cluster_ID) as the
previous PCH; thus, the cluster structure will remain intact
(CMs of the cluster do not have to reorganize in new cluster(s)
as the case of existing protocols); indeed, no re-clustering is
needed and thus no re-clustering overhead is generated. The
new PCH will also select a new SCH.
In VANETS, nodes are highly mobile; thus, the
determination of when PCH needs to order SCH to take the
role of PCH is a challenge. We propose to use mobility
prediction techniques to allow PCH to predict, ahead of time,
when it will move out of the cluster; thus, it will have enough
time to communicate with SCH. Mobility prediction is not
trivial in MANET [5]; the random way point mobility models
[7] are widely applied. Fortunately, with the roadway topology
constraints in VANET, we can use the existing driver
behaviour model [8] to make a better mobility prediction; this
will increase the probability of a better communication
between PCH and SCH during the switching process and thus
better reliability (in terms of packet losses and packet
delivery). However, in the case where PCH moves out of the
cluster without notifying SCH (e.g., bad mobility prediction),
SCH will elect itself PCH after T time units without receiving
ITJ messages from its PCH; we believe that such cases will be
rare.
IV. SIMULATION
RESULTS
In this section, we present our simulation and analysis
evaluation of our proposed protocol.
A. Simulator Setup
All simulations are run using NS-2 version 2.33. Table I
lists the various IEEE 802.11p parameters settings configured
in the simulator.
TABLE I.
PARAMETERS
’
VALUES USED IN THE SIMULATION
B. Scenario description
The mobility model used in the simulations is the freeway
mobility model with four highway lanes, all in the same
direction. When vehicles arrive at the end of the highway, they
wrap around from the beginning position of the same lane of
the highway. The scenario setup is shown in Figure 1. Each
node in the simulation is restricted to only travel within its
lane. The velocity of each node is temporally restricted based
on the node’s previous velocity. A safety distance is
maintained so that a node cannot exceed the velocity of the
node in front of it if they are within the safety distance.
A velocity range is specified for the nodes and the vehicle
acceleration is set to 10% of the maximum velocity. In these
simulations, we compare the performance of the cluster
configuration protocol (CCP) proposed in [1] against our
proposed SBCA protocol; CCP has been chosen because it
provides better CH stability, better cluster structure stability
and low communication overhead compared to existing
protocols. We consider the results for three different cases in
terms of vehicle densities, including low, medium and high;
these values are on average 50 (low density), 100 (medium
density) and 150 (high density) vehicles/km/4 lanes. For each
traffic density, the average preferred vehicle speed varies
uniformly between 25 and 35 m/s. In each scenario (with a
distinct traffic density), we ran the simulations ten times to
obtain the mean value of the final performance metric.
C. Simulation Results
To evaluate our proposed scheme we considered the
following metrics: (1) Average cluster lifetime: the time a
node remains associated with a given cluster; (2) Clustering
overhead: the total clustering related messages (i.e., RTJ, ITJ
and corresponding ACK) divided by the total messages that
are transmitted; this is an important metric because it shows
how efficient the scheme is in reducing clustering
communication overhead; and (3) Packet delivery ratio, the
number of packets successfully delivered divided by the total
number of packets generated.
Figure 3 shows that the average cluster lifetime, when
using SBCA, is considerably bigger than the average cluster
lifetime when using CCP. Therefore, we can conclude that,
when using SCBA, the cluster topology is more stable, since
CMs are associated with the cluster and not CH; in the case of
CCP, if the CH is absent for any reason, the cluster will
collapse and a new cluster will be formed. However, with
SCBA, when PCH can no longer be a CH, SCH takes over the
CH responsibility and thus the cluster leadership will be
changed from PCH to SCH; the CMs being heard by their
SCH would continue their cluster residency within SCH when
it has become a PCH. Hence, the cluster will still survive and
its cluster membership lifetime will be extended. Figure 3 also
shows that when increasing the number of nodes from 50 to
150, the cluster lifetime is gradually improved by three to four
times respectively. Therefore, the larger the network is, the
more stability our protocol provides compared to the cluster
configuration protocol.
Fig. 3- Average cluster lifetime vs. density (number of nodes)
Fig. 4- Overhead vs. density (number of nodes)
Figure 4 shows that the average clustering overhead
increases with the number of nodes. This is expected because
increasing the number of nodes increases the number of
clusters created and generally decreases CH duration.
Therefore, a node will join more clusters during its lifetime
generating more clustering related messages. Obviously, the
overhead generated by SBCA is far less than the overhead
generated by CCP; this is expected, since a CH leaving a
cluster does not cause necessarily re-clustering, when using
SBCA; in this case, SCH takes over the role of cluster head of
the cluster; therefore, a node does not need to change clusters
and thus does not need to exchange more clustering related
messages. We observe, from Figure 4, that CCP generates
more than twice the overhead generated by SBCA.
Figure 5 shows that SBCA performs far better than CCP in
terms of communication efficiency. Although the performance
generally decreases when the number of nodes in the network
increases, SBCA outperforms handily CCP. This can be
explained by the fact that SBCA has more stable clusters, thus
resulting in fewer route interruptions and thus few packet
losses and subsequent packet retransmissions.
Fig. 5- Packet Delivery ratio vs. density (number of nodes)
V. CONCLUSION
In this paper we proposed a new clustering algorithm
protocol (SBCA) based on stability, for use in VANETs. The
SBCA protocol involves two phases, setup and maintenance.
In the cluster setup phase, nodes in close vicinity with each
other are organized into clusters. In the cluster maintenance
step, all nodes remain associated with these clusters,
maintaining the hierarchical structure formed within the
dynamic environment, thus providing stable cluster
architecture and minimizing cluster maintenance costs. The
basic idea of SBCA protocol is the election of a secondary
cluster head (SCH) for each cluster. SCH works as a backup
for PCH, which was selected in setup phase and the future
leader of the cluster. The aim of SCH is to provide clustering
stability as well as routing reliability. In SBCA, SCH reforms
the cluster very quickly, thus reducing the overhead needed to
select new cluster heads. The simulation results show that the
proposed protocol can significantly improve the cluster
residence time, for each node, reducing the overhead and thus
improving the performance/reliability (in terms of packet
delivery).
REFERENCES
[1]
H. Su and X. Zhang, "Clustering-Based Multichannel
MAC Protocols for QoS Provisionings Over Vehicular Ad
Hoc Networks," IEEE Transactions on Vehicular
Technology, vol.56, pp. 3309-3323, 2007.
[2]
Y. Gunter, B. Wiegel, and H. P. Grossmann, “Cluster-
based medium access scheme for vanets,” Intelligent
Transportation Systems Conference, IEEE, pp. 343–348,
2007.
[3]
P. Fan, J. G. Haran, J. Dillenburg, and P. C. Nelson,
“Cluster-based framework in vehicular ad-hoc networks,”
Lecture Notes in Computer Science, vol. 3738, pp. 32–42,
2005.
[4]
T. J. Kwon, M. Gerla, V. K. Varma, M. Barton, and T. R.
Hsing, “Efficient flooding with passive clustering-an
overhead-free selective forward mechanism for ad
hoc/sensor networks ,” in Proc. of the IEEE, Vol. 91,
Issue 8, pp.1210 - 1220, 2003.
[5]
Mohammed S. Al-kahtani, Hussein T. Mouftah.
SERC/LC3R: A new paradigm for cluster-based routing
in MANETs. In Proceedings of BROADNETS, pp. 469-
478, 2005.
[6]
R.A. Santos, R.M. Edwards, A. Edwards, ”Cluster-based
location routing algorithm for vehicle to vehicle
communication”, Radio and Wireless Conference, pp. 39-
42, 2004.
[7]
W. Su, S.J. Lee and M. Gerla, ”Mobility prediction and
routing in adhoc wireless networks,” International Journal
of Network Management, Volume 11, No. 1, pp. 3-30,
2001.
[8]
K. Nagel, M. Schrekenberg, "A cellular automaton model
for freeway traffic", J. Phys. I 2, pp. 2221-2229, 1992.
[9]
A. Ahizoune, A. Hafid, R. Ben Ali, “A Contention Free
Broadcast Protocol for Periodic Safety Messages in
Vehicular Ad-Hoc Networks,” 35th Annual IEEE
Conference on Local Computer Networks and
Workshops, (LCN), pp. 48-55, 2010.
[10]
Z. Rawashdeh and S. Mahmud, “Media access technique
for cluster-based vehicular ad hoc networks,” Vehicular
Technology Conference,(VTC), IEEE 68th, pp. 1–5,
2008.
[11]
E. de Souza, I. Nikolaidis, and P. Gburzynski, "A New
Aggregate Local Mobility (ALM) Clustering Algorithm
for VANETs." Proceedings of ICC, Cape Town, South
Africa, 2010.
[12]
C. Shea, B. Hassanabadi, and S. Valaee, “Mobility-based
clustering in VANETs using affinity propagation,” in
IEEE Globecom, 2009.
[13]
P. Fan, P. Sistla, and P. C. Nelson, Theoretical analysis of
a directional stability-based clustering algorithm for
vanets. Vehicular Ad Hoc Networks, 2008.
[14]
S. Kuklinski and G. Wolny, “Density based clustering
algorithm for VANETs,” 5th International Conference on
Testbeds and Research Infrastructures for the
Development of Networks & Communities and
Workshops, pp. 1-6, 2009.
[15]
O. Kayis and T. Acarman, “Clustering formation for inter-
vehicle communication,” Intelligent Transportation
Systems Conference, IEEE, pp. 636–641, 2007.
