You are on page 1of 13

Performance Comparison of Ad-hoc

Routing Protocols for Networks


with Node Energy Constraints

EE 360 Class Project


Spring 2000-2001

Anne Aaron and Jie Weng


{amaaron, jweng} @stanford.edu

Abstract
A major issue with ad-hoc networks is energy consumption since nodes
are usually mobile and battery-operated. In this project we compared the
performance of ad-hoc routing protocols in a network where each participating
mobile node has a given battery life and will shut-down when a threshold is
reached. We studied two routing protocols, Dynamic Source Routing (DSR), and
Destination-Sequenced Distance-Vector Routing (DSDV), by comparing the node
termination rate as well as the over-all goodput of the network. From the
simulations we observed that in high node density, DSR performs better than
DSDV. The performance gap can be decreased by smaller number of nodes in the
network or more path loss and shadowing variance.
1 Introduction
A Wireless Ad-Hoc network is a cooperative engagement of a collection
of wireless mobile nodes. Since some receiving nodes may be out of the direct
range of a sending node, intermediate nodes have to act as routers to forward the
data to the receiving nodes. Ad-hoc routing protocols have been developed to
provide the route discovery and maintenance mechanisms for each mobile node in
the network to communicate with all other nodes of the network. Over the past
few years, the research community has developed several protocols and submitted
them as drafts to the IETF’s Mobile Ad-hoc Networking (MANET) group. The
more prominent protocols are the Destination-Sequenced Distance Vector
(DSDV), the Temporally-Ordered Routing Algorithm (TORA), the Ad-Hoc On-
demand Distance Vector (AODV) and the Dynamic Source Routing (DSR). A
brief overview of these protocols is presented in the next section.
Extensive research has been done in comparing the different proposed ad-
hoc routing protocols under varying network scenarios. Routing overhead,
packet delivery ratio, end-to-end delay, path optimality, and throughput are some
metrics commonly used in the comparisons. In [1] , Broch, et. al. extended the
ns-2 simulator to model the behavior of the IEEE 802.11 wireless LAN standard
and the wireless transmission channel. They used their ns-2 extension to compare
the packet delivery ratio, routing overhead and path optimality of DSDV, TORA,
AODV and DSR. Johansson, et. al. [2] extended these simulations by comparing
the throughput and delay of the protocols. Other research, such as that of Perkins,
et. al. [3], focused on only comparing the two on-demand routing protocols which
are AODV and DSR.
One important aspect of ad-hoc networks that was ignored by these studies
is energy-efficiency. In real life systems, energy consumption is a major issue.
For many ad-hoc networks, the nodes are small and portable, imposing stringent
constraints on the battery size and power. In this study, we considered a network
where each mobile node has a limited initial energy. As a node sends, receives or
forwards packets, the energy of the node is decremented accordingly. Once the
energy of the node reaches zero, the node is shut-down (a “node death”) and is
considered terminated by the system.
Since the ad-hoc routing protocol determines which nodes will forward the
packets and the amount of routing overhead each node needs, the type of protocol
definitely affects the energy performance of the system. The protocols affect the
energy dynamics in two ways – first, the routing overhead affects the amount of
energy used for sending and receiving the routing packets, and second, the chosen
routes affects which nodes will have a faster decrease in energy.
Cano and Manzoni [4] have studied the routing energy consumption of the
protocols using the ns-2 simulator. In their paper, they quantified the amount of
energy used for the routing overhead of AODV, DSR, TORA and DSDV under
different scenarios. The simulation results showed that DSR outperforms AODV
and DSDV due to its aggressive approach in promiscuous listening and caching.
TORA had very high consumption due to its aggregated messages for route
discovery and maintenance. The study however, did not look into the effect of the
protocol on the node deaths and its impact on the over-all data transfer of the
system. Furthermore, only the routing overhead energy used by the different
protocols was compared. It seems that the total energy is a better measure since
the routing protocol could also affect the MAC overhead and data packets because
of forwarding and retransmissions.
Other researchers have developed more “energy-aware” protocols for ad-
hoc network routing. Chang and Tassiulas [5] proposed an algorithm to select the
routes and power levels such that the time until the drain out of the batteries of the
nodes is maximized. They proposed that in order to maximize the lifetime of the
network, the traffic should be routed such that the energy is balanced among the
nodes in proportion to their energy consumption, instead of routing to minimize
absolute consumed power. Another proposal, by Ryu and Cho [6], studied a
routing scheme in home ad-hoc networks wherein the packets are routed though
the outlet-plugged devices instead of the battery-powered ones to prolong the life
of the batteries. Finally, Xu et. al. [7] proposed two algorithms, Basic Energy-
Conserving Algorithm (BECA) and Adaptive Fidelity Energy-Conserving
Algorithm (AFECA) to reduce the energy consumption of mobile nodes. BECA
is an algorithm that puts the radio of a node into “sleep mode” for a certain duty
cycle in order to reduce the idle or listening time energy consumption of the node.
AFECA makes use of information of local node density to adjust the length of
time the node is in sleep mode.
In this study, we did not implement any energy-efficiency algorithms, but
instead, compared the relative performance of the more established routing
protocols when there are energy constraints in the network. We used two
performance metrics in our study – the overall goodput of the network and the
node termination rate (refer to Section 5 for details).

2 Overview of Ad-hoc Routing Protocols


Ad-hoc routing protocols can be characterized in a number of ways as
explained in RFC2501. First of all, they must operate in a distributed fashion
allowing each node to enter and leave the network on its own. The routing
protocol should avoid data looping in the network. Nodes operate in either a
proactive or reactive mode. Similar to wired networks, proactive protocols are
table-driven and maintain routes for the entire network within each node. The
nodes must be in continuous communication about changes in the topology. For
very dynamic topologies, proactive protocols can introduce a large overhead in
bandwidth and energy consumption on the network. Reactive protocols trades off
this overhead with increased delay. A route to a destination is established when it
is needed based on an initial discovery between the source and the destination.
The following is four ad-hoc routing protocols proposed to the IETF.
Destination-Sequenced Distance Vector

Destination-Sequenced Distance Vector (DSDV) is a variation of the


Distributed Bellman-Ford algorithm modified to address problems inherent to ad-
hoc networks, such as time dependent topologies. These modifications reduce the
looping properties that would otherwise be present. Since DSDV is table-driven,
each node maintains a routing table with the next hop entry for each destination
and the metric for the link. In addition, each link has a sequence number
associated with it. This sequence number is periodically incremented by
the destination node for the link. Other nodes then choose the route with highest
sequence number, as that is the least stale route to the destination. If a node
detects that a link has broken, it sets the metric to infinity, and issues a route
update to the other nodes regarding the link status. Other nodes repeat this action
until they receive an update with a higher sequence number to provide it with a
fresh route again.
Due to the lack of synchronization between nodes in the network, a time
delay is imposed to prevent nodes from responding immediately based on a single
potentially disruptive update. This settling time allows for the routing table at
each node to stabilize before it beings issuing route updates to other nodes.
The main advantage to DSDV is that it maintains a loop-free fewest-hop path to
every destination in the network. However, this protocol also contains both
periodic and triggered route updates. While the triggered updates tend to be small
(allowing quick discovery of invalid links), the each node’s periodic update
includes its entire routing table. This means the overhead associated with those
updates grows as O(n2), effectively limiting the number of nodes in the network.

Ad-Hoc On-Demand Routing


AODV is a modification of the DSDV algorithm. When a source node
desires to establish a communication session, it initiates a path-discovery process
to locate the other node. The source node broadcasts a RREQ packet with its IP
address, Broadcast ID (BrID), and the sequence number of the source and
destination. While, the BrID and the IP address is used to uniquely identify each
request, the sequence numbers are used to determine the timeliness of each
packet. Receiving nodes set the backward pointer to the source and generates a
RREP unicast packet if it is the destination or contains a route to the destination
with a sequence number greater than or equal to the destination sequence number
contained in the original RREQ. As the RREP is routed back to the source,
forward pointers are setup by the intermediate nodes in their routing tables. The
deletion of a route would occur if an entry was not used within a specified
lifetime. Link failures are propagated by a RREP message with infinite metric to
the source node where route discovery would again occur. An optional feature of
AODV is the use of hello messages to maintain the connectivity of neighboring
nodes. The hello protocol yields a greater knowledge of the network and can
improve the route discovery process.
The authors of AODV maintain that it is a pure distributed on-demand
approach that minimizes routing table information. However, this also means, that
more route requests are generated. This compounded with the periodic ‘hellos’
can increase routing overhead. The advantage of AODV is its use of destination
numbers and replies to the first arriving RREQ implies that AODV favors the
least congested route instead of the shortest route. Furthermore, another advantage
is that the drafted standard supports both unicast and multicast packet
transmissions.

Dynamic Source Routing

Dynamic Source Routing (DSR) is one of the more generally accepted ad-
hoc routing protocols. As the name indicates, it utilizes source-based routing
rather than table-based, and it is source-initiated rather than hop-by-hop. When a
node wishes to establish a route, or issues a Route Request to all of its neighbors.
Each neighbor rebroadcasts this Request, adding its own address in the header of
the packet. When the Request is received by the destination or by a node with a
route to the destination, a Route Reply is generated and sent back to the sender
along with the addresses accumulated in the Request header.
The responsibility for assessing the status of a route falls to each node in
the route. Each must insure that packets successfully cross the link to the next
node. If it doesn’t receive an acknowledgement, it reports the error back to the
source, and leaves it to the source to establish a new route. While this process
could use up a lot of bandwidth, DSR gives each node a route cache for them to
use aggressively to reduce the number of control messages sent. If it has a cache
entry for any destination request received, it uses the cached copy rather than
forward the request. In addition, it promiscuously listens to other control
messages for additional routing data to add to the cache.
DSR has the advantage that no routing tables must be kept to route a given
packet, since the entire route is contained in the packet header. The caching of any
initiated or overheard routing data can significantly reduce the number of control
messages being sent, reducing overhead. Using only triggered updates furthers
that same goal.
The primary disadvantages are two-fold. DSR is not scalable to large
networks. The Internet Draft acknowledges that the protocol assumes that the
diameter of the network is no greater than 10 hops. Additionally, DSR requires
significantly more processing resources than most other protocols. In order to
obtain routing information, each node must spend much more time processing any
control data it receives, even if it is not the intended recipient.

Temporally-Ordered Routing Algorithm

Temporally-Ordered Routing Algorithm (TORA) is a distributed protocol


designed to be highly adaptive so it can operate in a dynamic network. For a
given destination, TORA uses a somewhat arbitrary ‘height’ parameter to
determine the direction of a link between any two nodes. As a consequence of this
multiple routes are often present for a given destination, but none of them are
necessarily the shortest route. For a node to initiate a route, it broadcasts a Query
to its neighbors. This is rebroadcast through the network until it reaches the
destination, or a node that has a route to the destination. This node replies with an
Update that contains its height with respect to the destination, which is propagated
back to the sender. Each node receiving the Update sets its own height to one
greater than that of the neighbor that sent it. This forms a series of directed links
from the sender to the destination in order of decreasing height. When a node
discovers link failure, it sets its own height higher than that of its neighbors, and
issues an Update to that effect reversing the direction of the link between them. If
it finds that it has no downstream neighbors, the destination is presumed lost, and
it issues a Clear packet to remove the invalid links from the rest of the network.
An advantage to TORA is that it supports multiple routes to any
source/destination pair. Failure or removal of one node is quickly resolved
without source intervention by switching to an alternate route. Unfortunately,
there are drawbacks to TORA as well. The most glaring being that it relies on
synchronized clocks among nodes in the network. While external time sources are
present (GPS for example), it makes the hardware to support it more costly, and
introduces a single point of failure if the time source became unavailable. TORA
also relies on intermediate lower layers for certain functionality. It assumes, for
example, that link status sensing, neighbor discovery, in-order packet delivery,
and address resolution are all readily available. The solution is to run the Internet
MANET Encapsulation Protocol (IMEP) at the layer immediately below TORA.
This makes the overhead for this protocol difficult to separate from that imposed
by the required lower layer.

3 Methodology
The primary approach for this study was computer simulations. We used
the network simulator ns-2 developed by the VINT research group at University
of California at Berkeley. The Monarch research group at Carnegie Mellon
University extended the ns-2 simulator to include wireless scenarios with mobile
nodes. The more established ad-hoc routing protocols were implemented in the
CMU extension. Subsequent versions of the CMU wireless extension also
included energy models for the mobile nodes and these were utilized for this
study. However, at present, the energy model for the AODV protocol has not yet
been implemented bug-free so this has limited our AODV simulations. From our
own observations and comments of others in the ns-2 news groups, we learned
that the AODV simulation goes unpredictably into infinite loops for certain
scenarios.

4 System Model
We simulated three different protocols (AODV, DSR, and DSDV). Due
to the problem mentioned in the previous section, AODV data were limited and
questionable. As a result, AODV was not reported in the simulation section and
only DSR and DSDV were compared. TORA was not considered because
previous studies have shown that TORA was not energy efficient and does not
scale to large networks.
The source of energy consumption for each node was the transmission and
reception of data packets. Other means of energy consumption such as when the
node is in listening mode or when the node is caching and filtering route
information were assumed to be equal among the two protocols. Each node in the
network is assigned an initial energy value. This energy level is decreased upon
packet transmission and reception. When the energy level reaches zero, the
corresponding node is no longer able to participate in communication.
In addition to the above assumptions we defined the ad-hoc network with
certain attributes. The definition of these attributes was based on scenarios
applicable to energy-constrained networks.

1. Mobility
The simulations was implemented with light speeds using the random
way-point mobility pattern. In this model, nodes select random way-points within
the roaming area, and travel there with a constant speed randomly chosen from a
uniform distribution, U[0, Vmax]. After reaching its destination, the node waits for
some pause time then moves to the next waypoint. This scenario is applicable to
networks such as conferences, wireless sensors, and emergency situations with
people walking as nodes. We assume low mobility for ad-hoc energy studies
since in a high mobility system the data transmission energy may be negligible
compared to the energy used for the mobility.

2. Traffic Pattern
Traffic was generated using constant bit rate. The network size was kept
constant while varying the number of nodes and the number of sources-
destination pairs within the network. In effect, this model attempts to observe the
relationship between traffic-density and system performance.

3. Physical Layer
The physical layer included a channel with both path loss and shadowing.
Each scenario is characterized by path a loss exponent and a shadowing variance.
This model is governed by the following equation.

Pr
(dB ) = 10 log10 K − 10γ log10 d + ψ dB
Pu do
where K is a unitless constant which depends on the antenna characteristics and
average attenuation, γ is the path loss exponent, and ψdB is the shadowing
variance. Other studies on energy comparisons of ad-hoc routing protocols have
only considered path loss in their physical layer. We believe that this model is
more realistic for our scenario-based simulations.
4. MAC Layer
The MAC layer implements the IEEE802.11 interface with the Distributed
Coordination Function (DCF). The network interface card uses values from the
Lucent WaveLan specifications.

5 Simulation and Results


For our simulation, we constructed a 670 x 670 m square topology. The
number of mobile nodes was varied from 20 to 100. For the random waypoint
model, a pause time of 40 s and a maximum speed of 1.5 m/s was used. Constant
bit rate connections of rate 4 packets per second and 512 bytes per packet were
randomly started at specific times and continued until the end of the simulation
time of 900 s. Two different physical layer parameters were used. The first
scenario using a path loss exponent of 1.6 and shadowing variance of 4 dB
emulated an indoor environment. An outdoor environment was modeled using a
path loss exponent of 2 and shadowing variance of 6 dB.
In comparing the performance of the different protocols, over-all network
goodput and the node termination rate was used as metrics. Goodput is the
number data packets successfully sent and received by the entire network within a
certain period of time. Node termination rate shows the trend of node deaths due
to the drain-out of energy. The trend line showing the number of terminated
nodes over time gives a good picture of the network lifetime of a system for a
given routing protocol. Since the different routing protocols have different
routing mechanisms, they will affect the energy dynamics and node termination
and therefore, will affect the goodput and network lifetime of the system.

5.1 Network Goodput


The over-all network goodput under both indoor and outdoor
environments was compared. Each scenario had a different number of nodes and
a corresponding number of connections. The number of connections (source and
final destination pairs) was about ¾ the number of nodes in the system. Each
point in the plot was averaged over 5 random mobility patterns.
As it can be seen from the plot (see Figure 1), DSR showed higher
goodput compared to the DSDV counterpart. The difference was negligible at 20
nodes but the advantage of DSR over DSDV becomes more prominent as the
number of nodes in the system increases. This was a reasonable result since with
a smaller number of nodes, the continuous routing overhead and routing table
maintenance of DSDV is comparable in size to that of the source routing of DSR.
We can infer that DSDV and DSR consume the same routing energy when the
network is small while DSR is more energy efficient when the network is large.
Another interesting observation from the plot was how the goodput of
both DSDV and DSR is not maximum at the maximum number of nodes and
connections, but instead, at about 60-80 nodes (with about 45-60 connections).
When the number of nodes is large, the source nodes or receiver nodes are more
likely to participate in forwarding, thereby decreasing its energy at a faster rate.
These nodes shutdown and ceased to communicate at an earlier point in time.
Also from the plot it is obvious how the indoor environment allowed more
goodput than the outdoor environment due to less variance in shadowing and path
loss. We can also see how the performance gap of DSR and DSDV is larger in
indoor environments.

Figure 1 – Goodput of DSR and DSDV for Different Number of Nodes

5.2 Node Termination Rate


The results for the node termination rate were consistent with the goodput
results. As it can be seen from the plots (see Figure 2 and 3), DSDV nodes died at
earlier times than DSR nodes. The plot for DSR with 80 nodes in an indoor
environment showed that by the end of the simulation time none of the nodes
died. However, for the exact scenario, all the nodes terminated by the simulation
end when the DSDV protocol was used. This is due to DSDV’s table-driven
mechanism in which routing tables are flooded by each node throughout the
network. This induced a large overhead on the energy consumed. The trend line
of DSDV was steep, implying that the nodes die at close to the same time. This is
also attributed to the dominant factor of the overhead which is evenly distributed
among all the nodes. However, for DSR, a staircase-like plot can be seen. Each
step, or group of node termination, corresponds to a specific connection
(including the sender, receiver and forwarding nodes).
The two plots, one for 80 nodes and the other for 100 nodes, showed the
effect of number of nodes on the performance of the network. A comparison of
the two plots showed that the 100 nodes scenario causes the nodes to terminate
earlier as compared to the 80 nodes, thus, decreasing the goodput. Also, the time
gap between the node terminations for DSR and DSDV is larger for indoor
scenarios, which is consistent with the goodput gap.

Figure 2 – Node Termination Rate of DSR and DSDV, 100 Nodes


Figure 3 – Node Termination Rate of DSR and DSDV, 80 Nodes

5.3 Effect of Node Density with Fixed Traffic Pattern


In this scenario, we used the same traffic pattern (same number of source
destination pairs) and varied the number of nodes participating in the network.
The plot for goodput showed that for DSR, the total number of packets
successfully received remained constant. We infer that the goodput does not
increase in spite of the additional possible routes because the goodput is
constrained by the lifetime of the source and destination nodes. The goodput does
not decrease because the additional nodes do not proactively exchange routing
information. On the other hand, DSDV showed a linear decrease in goodput as
the number of nodes increased. We believe that the downward trend of goodput
for DSDV is caused by the table-driven overhead proportional to the number of
nodes in the network.
Figure 4 – Goodput of DSR and DSDV for Different Number of Nodes but Same
Traffic

6 Summary and Conclusions


Results from this paper showed that network lifetime is a significant issue
for the performance of a multi-hop ad-hoc network. For light node density,
DSDV is comparable with DSR. However, DSR outperforms DSDV at high node
density. This is due to the tremendous overhead incurred by DSDV when
exchanging routing tables. The performance gap is especially significant in
indoor environments.
Many extensions can be applied to this study. It is obvious that this is not
a complete study of all the major protocols. A stable and loop-free AODV
implementation can be added to the performance comparison. In addition to
providing better energy consumption models that also account for caching and
processing of data packets, we believe that this energy-constraint model can also
be beneficial to the performance analysis of energy-aware protocols.
Furthermore, when such an energy-constraint exist, research can be devoted to a
fair-queuing mechanism that balances between the node’s own packets and the
packets that need to be forwarded.
7 References
[1] Per Johansson, Tony Larsson, Nicklas Hedman, and Bartosz Mielczarek.
Routing protocols for mobile ad-hoc networks – a comparative performance
analysis. In proceedings of the 5th International Conference on Mobile
Computing and Networking (ACM MOBICOM ’99), August 1999, pages 195-
206.

[2] J. Broch, D.A. Maltz, D. B. Johnson, Y-C. Hu, and J. Jetcheva. A performance
comparison of Multi-hop wireless ad-hoc networking routing protocols. In
Proceedings of the 4th International Conference on Mobile Computing and
Networking (ACM MOBICOM ’98), October 1998, pages 85-97.

[3] Charles Perkins, Elizabeth Royer, Samir Das, and Mahesh Marina.
Performance of two on-demand Routing Protocols for Ad-hoc Networks. IEEE
Personal Communications, February 2001, pages 16-28.

[4] Juan-Carlos Cano and Pietro Manzoni. A Performance Comparison of


Energy Consumption for Mobile Ad-Hoc Network Routing Protocols. Proceedings
of the 8th International Symposium on Modeling, Analysis and Simulation of
Computer and Telecommunication Systems, 2000, pages 57 –64.

[5] Jae-Hwan Chang and Leandros Tassiulas. Energy conserving routing in


wireless ad-hoc networks. INFOCOM 2000. Proceedings of the Nineteenth
Annual Joint Conference of the IEEE Computer and Communications Societies.
Volume: 1 , 2000, page 22 –31.

[6] Jung-hee Ryu and Dong-Ho Cho. A new routing scheme concerning energy
conservation in wireless home AD-HOC networks. IEEE Transactions on
Consumer Electronics, Volume 47 Issue 1, February 2001, page 1 –5.

[7] Ya Xu, John Heidemann, and Deborah Estrin. Adaptive Energy-Conserving


Routing for Multihop Ad-hoc Networks. USC-ISI Research Report 527, October
12, 2000.

You might also like