Clustering is an important research area for mobile ad hoc networks (MANETs) as it increases the capacity of network, reduces the routing overhead and makes the network more scalable in the presence of both high mobility and a large number of mobile nodes. Routing protocols based on flat topology are not scalable because of their built-in characteristics. However, clustering cause overhead which consumes considerable bandwidth, drain mobile nodes energy quickly, likely cause congestion, collision and data delay in larger networks. This paper uses an implementation of the Dynamic Source Routing (DSR), an flat architecture based and the Cluster Based Routing Protocol (CBRP), a cluster architecture based routing protocol to examine the influence of clustering on the performance of mobile ad hoc networks. This paper evaluates channel utilization and control overhead as a function of number of nodes per sq. km to show the effect of clustering. Simulation results show that in high mobility scenarios, CBRP outperforms DSR. CBRP scales well with increasing number of nodes.

1. Influence of Clustering on the Performance of
MobileAd Hoc Networks (MANETs)
Narendra Singh Yadav
Department of ECE
Malaviya National Institute of Technology
Jaipur, India.
narensinghyadav@yahoo.com
R.P.Yadav
Department of ECE
Malaviya National Institute of Technology
Jaipur, India
rp_yadav@yahoo.com
Abstract— Clustering is an important research area for
mobile ad hoc networks (MANETs) as it increases the
capacity of network, reduces the routing overhead and
makes the network more scalable in the presence of both
high mobility and a large number of mobile nodes. Routing
protocols based on flat topology are not scalable because of
their built-in characteristics. However, clustering cause
overhead which consumes considerable bandwidth, drain
mobile nodes energy quickly, likely cause congestion,
collision and data delay in larger networks. This paper
uses an implementation of the Dynamic Source Routing
(DSR), an flat architecture based and the Cluster Based
Routing Protocol (CBRP), a cluster architecture based
routing protocol to examine the influence of clustering on
the performance of mobile ad hoc networks. This paper
evaluates channel utilization and control overhead as a
function of number of nodes per sq. km to show the effect
of clustering. Simulation results show that in high mobility
scenarios, CBRP outperforms DSR. CBRP scales well
with increasing number of nodes.
Keywords- Mobile Ad hoc Networks; Routing protocols;
Dynamic source routing; Cluster based routing protocol;
Simulation; Performance evaluation
I. INTRODUCTION
Ad hoc wireless network is a reconfigurable network
of mobile nodes connected by multi-hop wireless links
and capable of operating without any fixed infrastructure
support, where each node acts as a router thus relaying
packets for other nodes. The advantages of such a
network are rapid deployment, robustness, flexibility and
inherent support for mobility. Ad hoc networks, due to
their quick and economically less demanding
deployment, find applications in military operations,
collaborative and distributed computing, emergency
operations, wireless mesh networks, wireless sensor
networks and hybrid networks.
Routing protocols for Mobile ad hoc networks can
be classified into two main categories: Proactive or table
driven routing protocols and Reactive or on-demand
routing protocols [1]. A flat architecture exclusively
based on table-driven or on-demand routing approaches
cannot perform well in a large MANET [2–4]. This is
because a flat architecture or topology encounters
scalability problems with increased network size,
particularly with node mobility at the same time. The
reason for this is their built-in characteristics. Proactive
routing is table –driven based and requires control
overhead for building and updating those tables,
containing information about the state of the network.
The control overhead for proactive routing protocols is O
(n2
), where n is the total number of nodes in a network
[5]. Since in reactive routing protocols routes are found
on – demand; incur significant route setup delay which
becomes intolerable in the presence of both a large
number of nodes and mobility. Therefore, both proactive
and reactive routing schemes are not scalable. Clearly,
some sort of a compromise needs to be achieved between
these two schemes.
A major problem for ad hoc wireless networks is the
handling of a large number of nodes. As nodes can join
ad hoc wireless network, contention is more likely and
the open nature of ad hoc wireless network makes it
important that a network remains operational even if
there are more nodes. Most of the previous work [6-13]
is limited on performing simulations for ad hoc wireless
networks with a limited number of nodes (50-100 nodes)
deployed in small geographical areas.
In this paper, we present performance comparison of
two on – demand source routing protocols proposed for
ad hoc wireless networks. In particular, the main goal is
the evaluation of the channel utilization and control
overhead of the protocols by focusing on large and dense
network. We will show that, in some scenarios it could
be convenient using simple flat topology based routing
protocols like DSR rather than cluster based CBRP. To
compare the protocol behaviors, simulation results
performed with NS-2 (Network Simulator -2) are given.
Network Simulator2 (NS-2) is an object-oriented,
discrete event driven network simulator developed at UC
Berkely written in C++ and OTcl. More details about
NS-2 can be found in [14].
The rest of the paper is organized as follows: Section
2 provides an overview of the routing protocols used in
the study. The simulation environment and performance
metrics are described in Section 3 and then the results are

2. presented in Section 4. Finally Section 5 concludes the
paper.
II. OVERVIEW OF DSR AND CBRP
As each protocol has its own merits and demerits,
none of them can be claimed as absolutely better than
others. Two mobile ad hoc routing protocols – the
Dynamic Source Routing (DSR), the flat architecture
based On-Demand source routing protocol and the
Cluster Based Routing Protocol (CBRP), the cluster
architecture based On-Demand source routing protocol
are selected for study.
A. Dynamic Source Routing protocol (DSR)
The Dynamic Source Routing Protocol [15] is an on-
demand routing protocol designed to restrict the
bandwidth consumed by control packets by eliminating
the periodic table-update messages required in the table-
driven approach. It is beacon-less and hence does not
require periodic hello packet transmissions, which are
used by a node to inform its neighbors of its presence.
The key distinguishing feature of DSR is the use of
source routing. That is, the sender knows the complete
hop-by-hop route to the destination. These routes are
stored in a route cache. The data packets carry the source
route in the packet header.
When a node in the ad hoc network attempts to send
a data packet to a destination for which it does not
already know the route, it uses a route discovery process
to dynamically determine such a route. Route discovery
works by flooding the network with route request
(RREQ) packets. Each node receiving an RREQ
rebroadcasts it, unless it is the destination or it has a
route to the destination in its route cache. Such a node
replies to the RREQ with a route reply (RREP) packet
that is routed back to the original source. RREQ and
RREP packets are also source routed. The RREQ builds
up the path traversed across the network. The RREP
routes back to the source by traversing this path
backward. The route carried back by the RREP packet is
cached at the source for future use.
B. Cluster Based Routing Protocol (CBRP)
In CBRP [16] the nodes of a wireless network are
divided into clusters. The diameter of a cluster is only
two hops and clusters can be disjoint or overlapping.
Each cluster elects one node as the clusterhead,
responsible for the routing process. The head of a cluster
knows the addresses of its members. Clusterheads
communicate with each other through gateway nodes. A
gateway is a node that has two or more clusterheads as
its neighbors when the clusters are overlapping or at least
one clusterhead and another gateway node when the
clusters are disjoint
The routing process works in two steps. First, it
discovers a route from a source node S to a destination
node D, afterwards it routes the packets. When a source
has to send data to destination, it floods route request
packets (but only to the neighboring cluster-heads). On
receiving the request a clusterhead checks to see if the
destination is in its cluster. If yes, then it sends the
request directly to the destination else it sends it to all its
adjacent cluster-heads. The cluster-heads address is
recorded in the packet so a cluster-head discards a
request packet that it has already seen. When the
destination receives the request packet, it replies back
with the route that had been recorded in the request
packet. If the source does not receive a reply within a
time period, it backs off exponentially before trying to
send route request again.
III. SIMULATION AND PERFORMANCE METRICS
A. Simulation model
Network Simulator2 (NS-2) a object-oriented,
discrete event driven network simulator developed at UC
Berkely written in C++ and OTcl, particularly popular in
the ad hoc networking research community is use for the
simulations. The traffic sources are CBR (continuous bit
– rate). The source-destination pairs are spread randomly
over the network. The node movement generator of ns-2
is used to generate node movement scenarios. The
movement generator takes the number of nodes, pause
time, maximum speed, field configuration and
simulation time as input parameters. The propagation
model is the two ray ground model [17]. Each simulation
scenario is run for enough time to reach and collect the
desired data at steady state. Several runs of each
simulation scenario are conducted to obtain statistically
confident averages. Simulation parameters are listed in
table 1.
TABLE I. SIMULATION PARAMETERS
Parameters value
Network size 50, 100, 150, 200 and 250
nodes
Area 2000m x 500m
Traffic model CBR
Traffic sources 30% and 70%
Packet size 512 bytes
Packet rate 4 packets/s
Max. speed 20m/s
Transmission range 250m
Bandwidth 2 Mb/s
Node movement model Random way point
B. Performance metrics
In order to compare the performance of cluster
architecture based, CBRP and flat architecture based,
DSR this paper focus on the following performance
metrics for evaluation:
• Channel utilization capacity: This metric gives
the fraction of channel capacity used for data
transmitted by the network and is computed as

3. ( ) BWSET
SZPR
CU
∗
∗
=
o
o
where PR is the number of data packets received
by the destination nodes, SZ is the size of the
data packets, SET is simulation end time and
BW is the nominal channel bandwidth.
• Control Overhead: The total number of non data
packets transmitted by the protocol. In DSR,
routing overhead (ROH) that is the total number
of routing packets transmitted during the
simulation is the main source of control
overhead. For CBRP, the control overhead is the
sum clustering overhead (COH) and routing
overhead. Clustering overhead is the number of
clustering messages sent by each node in cluster
formation and cluster maintenance operation. It
is an important measure for the scalability of a
protocol. If a protocol requires sending many
control packets, it will most likely cause
congestion, collision and data delay in larger
networks.
IV. RESULTS
The simulation results are shown in the following
section in the form of line graphs. Graphs show
comparison between the two protocols on the basis of the
above-mentioned metrics as a function of node density in
high mobility and stationary scenarios with 30%, 70%
traffic sources.
A. Channel utilization
Fig. 1 shows the channel utilization comparison
among CBRP and DSR with different number of nodes
per sq. km. and with 30%, 70% CBR traffic sources in
high mobility scenarios.
0
2
4
6
8
10
12
14
50 100 150 200 250
Number of nodes per sq. km
Channelutilization(%)
CBRP, 30% sources
DSR, 30% sources
CBRP, 70% sources
DSR, 70% sources
Figure 1. Channel utilization vs Number of nodes per sq.km in high
mobility scenarios.
With 30% CBR sources, the channel utilization of
both the protocols improve with the number of nodes.
Up to N = 100 nodes per sq. km both the protocols show
comparable channel utilization. As the number of nodes
per sq. km increase, CBRP performs better than DSR
and with N = 250 CBRP clearly outperforms DSR at a
factor of ~ 1.5. However this is not true for 70% traffic
sources. Both the protocols show improvement in
channel utilization but up to certain N after that the
performance degrades. For DSR, the channel utilization
improves at a factor of 2 from N = 50 (4.61%) to N =
100 (9.59%) nodes per sq. km after that the performance
shows a little improvement up to N = 150, where the
performance is maximum (9.96%). The performance
degrades on increasing number of nodes beyond 150.
The CBRP depicts the same behavior and attains the
maximum channel utilization at N = 200 (12.36%) after
that the performance degrades slowly. Both have nearly
comparable performance up to N = 100 and with
increasing nodes density CBRP performs better than
DSR.
0
2
4
6
8
10
12
14
16
18
20
50 100 150 200 250
Number of nodes per sq. km
Channelutilization(%)
CBRP, 30% sources
DSR, 30% sources
CBRP, 70% sources
DSR, 70% sources
Figure 2. Channel utilization vs Number of nodes per sq.km in
stationary scenarios.
Fig. 2 shows the channel utilization comparison
among CBRP and DSR with different number of nodes
per sq. km. and with 30%, 70% CBR traffic sources in
stationary scenarios. With 30% sources, both the
protocols show improvement in channel utilization with
increasing number of nodes. The performance improves
from 2% at N = 50 to 11% at N = 250 for both the
protocols. With 70% CBR sources the channel utilization
improves from ~ 5% at N = 50 to ~ 17% at N = 200 for
both protocols. But with further increasing number of
nodes CBRP shows slight improvement in performance,
whereas for DSR the channel utilization degrades.
DSR has a lower channel utilization than CBRP in
high mobility scenarios and same or less (N = 250 with
70% sources) in stationary scenarios. The performance
degradation in channel utilization is due to packet drops
by the routing algorithm after being failed to transfer
data in the active routes. The packet drops are due to
network partitioning, link break, collision and congestion
in the ad hoc network.

4. B. Control overhead
Fig. 3 shows the control overhead comparison among
CBRP and DSR with different number of nodes per sq.
km. and with 30%, 70% CBR traffic sources in high
mobility scenarios. With 30% sources, the control
overhead for both the protocols increase with number of
nodes. For DSR overhead increases from 1027 packets at
N = 50 to 40552 packets at N = 250 (nearly 40 times),
whereas in CBRP overhead increases just 6 times from N
= 50 (3364 packets) to N = 250 (19804 packets). For
70% traffic sources both protocols experience significant
growth in control overhead for DSR (nearly 33 times)
and for CBRP (just 8 times) from N = 50 to N = 250 per
sq. km. For DSR the control overhead increases at factor
of 2 when traffic sources are increased from 30% to 70%
whereas CBRP shows significant overhead only when
the number of nodes are more than 100 per sq. km.
Fig.3 shows two interesting things, first DSR
performs better than CBRP with less number of nodes (N
= 100) but with number of nodes more than 100 CBRP
clearly outperforms DSR. Second, for CBRP the control
overhead is independent of traffic sources for less
number of nodes (N = 150).
0
10
20
30
40
50
60
70
80
90
50 100 150 200 250
Thousands
Number of nodes per sq. km
Controloverhead(packets)
CBRP, 30% sources
DSR, 30% sources
CBRP, 70% sources
DSR, 70% sources
Figure 3. Control overhead vs Number of nodes per sq.km in high
mobility scenarios.
Fig. 4 shows that the control overhead for both the
protocols increase with increasing number of nodes per
sq. km. and traffic sources in stationary scenarios. With
30% sources the control overhead for CBRP is much
more than that of DSR, of the order of 6 or more for less
number of nodes (up to N = 100) and of the order of 4
for more number of nodes (N more than 100). But with
70% sources, the control overhead for CBRP is more
than 4 times that of DSR for less number of nodes ( up to
150) after that the overhead for DSR increases sharply
and at N = 250 CBRP has just 2000 control packets more
than that of DSR. CBRP shows less dependency on
number of traffic sources whereas DSR experiences
significant dependency on traffic sources when the
number of nodes is more than 150 per sq. km
0
2
4
6
8
10
12
14
16
18
20
50 100 150 200 250
Thousands
Number of nodes per sq. km
Controloverhead(packets)
CBRP, 30% sources
DSR, 30% sources
CBRP, 70% sources
DSR, 70% sources
Figure 4. Control overhead vs Number of nodes per sq.km in
stationary scenarios.
Fig. 5 shows the routing overhead comparison among
CBRP and DSR with different number of nodes per sq.
km. and with 30%, 70% CBR traffic sources in high
mobility scenarios. The routing overhead for both the
protocols increase with increasing number of nodes and
traffic sources. With 30% sources, DSR shows more
routing overhead than CBRP ranging from 3 to 9 times
of CBRP with increasing number of nodes from 50 to
250 per sq. km. DSR shows the same behavior with 70%
sources, but the range is from 2 to 6 times of CBRP.
Both the protocols show increase in routing overhead
with increasing traffic sources and this dependence is
more in case of CBRP with number of nodes is more
than 150 per sq. km.
0
10
20
30
40
50
60
70
80
90
50 100 150 200 250
Thousands
Number of nodes per sq. km
Routingoverhead(packets)
CBRP, 30% sources
DSR, 30% sources
CBRP, 70% sources
DSR, 70% sources
Figure 5. Routing overhead vs Number of nodes per sq.km in high
mobility scenarios.
Fig. 6 shows the routing overhead for both the
protocols with increasing number of nodes 50, 100, 150,
200, 250 per sq. km and with 30%, 70% traffic sources
in stationary scenarios. With 30% sources, the routing
overhead for DSR is 2 to 4 times more than that of
CBRP when the number of nodes is increased from 50 to
250 per sq. km. Both the protocols show the same
behavior with 70% traffic sources. Also both the

5. protocols show increase in routing overhead with
increasing traffic sources from 30% to 70%.
0
2
4
6
8
10
12
14
16
50 100 150 200 250
Thousands
Number of nodes per sq. km
Routingoverhead(packets)
CBRP, 30% sources
DSR, 30% sources
CBRP, 70% sources
DSR, 70% sources
Figure 6. Routing overhead vs Number of nodes per sq.km in
stationary scenarios.
Fig. 7 shows the fraction of routing and clustering
overhead of control overhead for CBRP with different
number of nodes per sq. km. and with 30%, 70% CBR
traffic sources in high mobility scenarios. With 30%
sources the main source of control overhead is clustering
overhead, which is more than 75% of control overhead.
Routing overhead is well below 25% of control
overhead. With 70% sources the same holds true up to N
= 150 where fraction of COH is more than two times that
of ROH but the fraction of ROH is more than that of
COH with increasing number of nodes more than 150
per sq. km
0
10
20
30
40
50
60
70
80
90
100
50 100 150 200 250
Number of nodes per sq. km
COH,ROH(%)
COH, 30% sources ROH, 30% sources
COH, 70% sources ROH, 70% sources
Figure 7. Fraction of routing and clustering overhead vs Number of
nodes per sq.km in high mobility scenarios.
Fig. 8 shows the fraction of routing and clustering
overhead of control overhead for CBRP with different
number of nodes per sq. km. and with 30%, 70% CBR
traffic sources in stationary scenarios. With 30% traffic
sources, the fraction of COH is more than 89% whereas
the fraction of ROH is always less 11%. With increasing
traffic sources to 70% the fraction of COH decreases.
0
10
20
30
40
50
60
70
80
90
100
50 100 150 200 250
Number of nodes per sq. km
COH,ROH(%)
COH, 30% sources ROH, 30% sources
COH, 70% sources ROH, 70% sources
Figure 8. Fraction of routing and clustering overhead vs Number of
nodes per sq.km in stationary scenarios.
Fig. 9 shows the comparison of clustering overhead
with increasing number of nodes per sq. km for CBRP.
The clustering overhead is proportionally increased by
increasing the number of nodes and is independent of
traffic sources. This is logically true as clustering
overheads are the control packets sent by each node
periodically for cluster formation and cluster
maintenance. A control packet are of two types- hello
packets sent periodically by each node at hello_interval
seconds, proportional to the number of nodes and second
is triggered hello packets generated in cluster
maintaining events, depends upon the mobility.
0
2
4
6
8
10
12
14
16
18
50 100 150 200 250
Thousands
Number of nodes per sq. km
clusteringoverhead(packets)
COH, 30% in high mobility scenarios
COH, 70% in high mobility scenarios
COH, 30% in stationary scenarios
COH, 70% in stationary scenarios
Figure 9. Clustering overhead vs Number of nodes per sq.km
CBRP has lower routing overhead than DSR in high
mobility and stationary scenarios. This is because CBRP
only broadcasts route requests to cluster heads. Gateway
nodes receive the route requests as well but they forward
them to the next cluster-heads. This largely reduces the
route discovery packets which in turn reduces NRL.
CBRP has more control overhead than DSR in stationary
and for low number of nodes in high mobility scenarios.
This is because of clustering overhead.

6. V. CONCLUSIONS
This paper compared the performance of DSR and
CBRP, two on-demand source routing protocols for ad
hoc wireless networks. DSR and CBRP both use on-
demand route discovery, but with different flooding
behavior. In particular, DSR use network-wide flooding
for route discovery and does not depend on any periodic
hello message or timer-based activities. CBRP, on the
other hand, only broadcasts route requests to cluster
heads, largely reducing the network traffic. Hello
messages are the integral part of CBRP and the size of a
hello message may be large as it contains the neighbor
table and cluster adjacency table of the sender. As a
result, while CBRP uses hello messages to establish
clusters and in turn reduce the flood in route discovery,
the hello message itself is another kind of overhead. The
general observation from the simulation is that in static
scenarios, due to similar performance any, either DSR or
CBRP, can be used in large ad hoc wireless networks but
in high mobility scenarios, CBRP outperforms DSR.
CBRP scales well with increasing number of nodes per
sq. km.
REFERENCES
[1] E. M. Royer and C. K. Toh, “A review of current routing
protocols for ad hoc mobile wireless networks,” IEEE
Personal Communications magazine, April 1999, pp. 46–55.
[2] P. Gupta and P. R. Kumar, “The Capacity of Wireless
Networks,” IEEE Trans. Info. Theory, vol- IT 46.2, Mar. 2000,
pp. 388-404.
[3] X. Y. Hong, K. X. Xu and M. Gerla, “Scalable Routing
Protocols for Mobile Ad Hoc Networks,” IEEE Network, July-
Aug. 2002, pp. 11-21.
[4] X. Y. Hong, K. X. Xu and M. Gerla, “An Ad Hoc Network with
Mobile Backbones,” Proc. IEEE ICC 2002, vol. 5, Apr.-May
2002, pp. 3138-43
[5] E. M. Belding-Royer, “Hierarchical Routing in Ad Hoc Mobile
Networks,” Wireless Commun. And Mobile Comp., vol. 2, no. 5,
2002, pp. 515-32.
[6] J. Broch, D. A. Maltz, D. Johnson, Y. C. Hu and J. Jetcheva, “A
Performance Comparison of Multi-Hop Wireless Ad Hoc
Network Routing Protocols,” in Proceedings of ACM/IEEE
MOBICOM 1998, October 1998, pp. 85-97.
[7] S. R. Das, C. E. Perkins and E. M. Royer, “Performance
Comparison of Two On-Demand Routing Protocols for Ad Hoc
Networks,” in Proceedings of the IEEE INFOCOM 2000, March
2000, pp. 3-12.
[8] Hong Jiang, J. J. Garcia-Luna-Aceves, “Performance
Comparison of Three Routing Protocols for Ad Hoc Networks”,
Computer Communications and Networks, 2001, Proceedings,
Tenth International Conference on, 2001; Page(s): 547-554.
[9] Samir R. Das, Charles E. Perkins, Elizabeth M. Royer and
Mahesh K. Marina. ”Performance Comparison of Two On-
demand Routing Protocols for Ad hoc Networks.” IEEE
Personal Communications Magazine special issue on Ad hoc
Networking, February 2001, pp. 16-28.
[10] P. Johansson, T. Larsson, N. Hedman, B. Mielczarek and M.
Degermark, “Scenario-based Performance Analysis of Routing
Protocols for Mobile Ad-hoc Networks,” in Proceedings of the
ACM/IEEE MOBICOM 1999, August 1999, pp.195-206.
[11] S. R. Das, R. Cestaneda, J. Yan and R. Sengupta, “Comparative
performance evaluation of routing protocols for mobile ad hoc
networks,” in proceedings of the IC3N 1998, October 1998, pp.
153-161.
[12] Narendra Singh Yadav and R.P.Yadav, “The Effects of Speed on
the Performance of Routing Protocols in Mobile Ad Hoc
Networks”, International Journal of Electronics, Circuits and
Systems, vol.1, no. 2, 2007, pp. 79-84.
[13] Narendra Singh Yadav and R.P.Yadav, “Performance
comparison and analysis of table-driven and on-demand routing
protocols for Mobile Ad Hoc Networks”, International Journal of
Information Technology, vol. 4, no. 2, 2007 , pp. 101-109.
[14] K. Fall and K. Vardhan, The Network Simulator (ns-2).
Available: http://www.isi.edu/nsnam/ns
[15] D.Johnson, D.Maltz, Y.Ho, “The dynamic source routing
protocol for mobile ad hoc networks”, IETF Manet working
group, Internet RFC 2026.
[16] Mingliang Jiang, Jinyang Li and Y.C.Tay, “Cluster Based
Routing Protocol”, August 1999 IETF Draft.
http://www.ietf.org/internet-drafts/draft-ietf-manetcbrp-spec-
01.txt
[17] T. S. Rappaport, Wireless Communications, Principles &
Practices. Prentice Hall, 1996, ch. 3, pp. 70-74.