A routing protocol based on node density for ad hoc networks

Ad Hoc Networks 2 (2004) 335–349

www.elsevier.com/locate/adhoc

A routing protocol based on node density for ad hoc networks

Alejandro Quintero *, Samuel Pierre, Benjamin Macab�eo

Department of Computer Engineering, Mobile Computing and Networking Research Laboratory (LARIM),�Ecole Polytechnique de Montr�eal, C.P. 6079, Succ. Centre-Ville, Quebec, Montreal, Canada H3C 3A7

Available online 17 April 2004

Abstract

Ad hoc networks are a type of mobile network that functions without any fixed infrastructure. One of the weak-

nesses of ad hoc networks is that a route used between a source and a destination is likely to break during commu-

nication. To solve this problem, one approach consists of selecting routes whose nodes have the most stable behavior.

Another strategy aims at improving the route repair procedure. This paper proposes a method for improving the

success rate of local route repairs in mobile ad hoc networks. This method is based on the density of the nodes in the

neighborhood of a route and on the availability of nodes in this neighborhood. Theoretical computation and simulation

results show that the data packet loss rate decreased significantly compared to other methods which are well-docu-

mented in the literature. In addition, the time required to complete a local route repair following a failure was sig-

nificantly reduced.

� 2004 Elsevier B.V. All rights reserved.

Keywords: Mobile ad hoc networks; Route repair; AODV

1. Introduction

In recent years, we have witnessed considerable

accomplishments in the design, development, and

deployment of wireless communication networks.

Personal and mobile communications are made

possible by the convergence of several different

technologies, specifically computer networking

protocols, wireless/mobile communication sys-

tems, distributed computing and Internet [6,14,25].

* Corresponding author. Tel.: +1-514-340-3240x4685; fax:

+1-514-340-3240.

E-mail addresses: [email protected] (A. Quin-

tero), [email protected] (S. Pierre), benjamin.maca-

[email protected] (B. Macab�eo).

1570-8705/$ - see front matter � 2004 Elsevier B.V. All rights reserv

doi:10.1016/j.adhoc.2004.03.003

The mixed wired and wireless network topologies

that are becoming so common, including fixed andad-hoc connection types, create the need to ratio-

nally exploit dynamically variable routing as a

function of network conditions [10].

At the same time, a phenomenal growth in data

traffic and a wide range of new requirements of

emerging applications call for new mechanisms for

the control and management of communication

networks [20]. The emergence of real-time appli-cations and the widespread use of wireless and

mobile devices have generated the need to provide

quality of service (QoS) support in wireless and

mobile networking environments [24].

A mobile ad hoc network (MANETs) is a mobile

wireless network composed of several mobile

nodes, likely to communicate among themselves

ed.

336 A. Quintero et al. / Ad Hoc Networks 2 (2004) 335–349

without the intervention of any centralized man-

agement or existing infrastructure. Hence, these

mobile nodes must necessarily be able to cooperate

to allow communication between themselves.

Their main asset resides in the fact that they are

not tributary to fixed installations.Ad hoc wireless networks are self-creating, self-

organizing and self-administering [3]. The distinct

features that characterize wireless ad hoc networks

are [24,26]:

• There is no centralized authority for network

control, routing or administration.

• Frequent and dynamic topological modifica-tions due to mobile hosts changing arbitrarily

their point of connectivity.

• Limited resources, including energy, band-

width, processing capacity and memory.

• Network nodes play multiple roles: source/des-

tination versus router.

• All communications, user data and control

information, are carried over the wireless med-ium, consisting of bandwidth constrained links.

• MANETs are highly heterogeneous environ-

ments due to the diverse nature of communica-

tions technologies employed, as well as the

presence of the different types of nodes.

• Limited survivability and vulnerability to secu-

rity attacks.

Quality of service (QoS) provisioning and

management in ad hoc networks remains a chal-

lenging task because their topologies change

dynamically and arbitrarily with time. Further-

more, channel conditions of wireless mediums are

also time-varying. Hence providing QoS support

for the delivery of real-time audio, video and data

in MANETs presents a number of technical chal-lenges [8]. Providing a complete QoS solution for

the ad hoc networking environment requires the

interaction and cooperation of several compo-

nents: (1) a QoS routing protocol, (2) a resource

reservation scheme, and (3) a QoS capable med-

ium access control layer [24]. Many research efforts

focus either on providing QoS or network man-

agement in ad hoc networks. However, a com-prehensive approach to QoS management in ad

hoc networks, i.e., network management in sup-

port of service differentiation, QoS robustness, and

network survivability, is still lacking [26].

This paper presents a method which improves

the success rate of a local route repair by acceler-

ating the route reparation procedure. Section 2

presents background and related work. Section 3describes the solution proposed to improve route

reparation. Finally, Section 4 presents and ana-

lyzes simulation results.

2. Background and related work

A routing protocol is the mechanism by whichuser traffic is directed and transported through the

network from a source node to a destination node.

The objectives include maximizing network per-

formance from an application point of view, while

minimizing the cost imposed on the network in

terms of capacity. QoS routing is an essential part

of a QoS architecture. It is a routing mechanism

under which paths for flows are determined on thebasis of some knowledge of the resources available

in the network as well as on the QoS requirements

of the flows or connections [24].

Resource reservation is necessary for providing

guaranteed end-to-end performance for multi-

media applications. However, resource reservation

is supported neither in the present Internet nor in

mobile ad hoc networks. Also, the data packetssent by these applications could follow different

paths and reach the destination out of order,

which is not desirable. The current routing pro-

tocols used in IP networks are transparent to any

particular QoS that different flows could require.

As a result, routing decisions are made without

referring to the QoS requirements of the flow. This

means that flows are often routed over paths thatare unable to support their requirements while

alternate paths with sufficient resources exist. This

will increase the call blocking probability. Hence,

the goal of QoS routing algorithms is to find a

path in the network that satisfies the given

requirements [13,19,20].

Internet and wireless networks are based on the

best effort principle, which consists of using linksamong computers in an optimal way, even if this

involves additional delays in data transmission.

A. Quintero et al. / Ad Hoc Networks 2 (2004) 335–349 337

This principle is problematic for new applications

which require a QoS such as multimedia, IP tele-

phony and networked gaming.

Gelenbe et al. [12] present the concept of Cog-

nitive Packet Networks (CPN) in which reliability,

security, scalability and QoS are key issues. Theypropose packet switching networks in which intel-

ligent capabilities for routing and flow control are

concentrated in the packets, rather than in the

nodes and protocols. Networks which contain

Cognitive Packets (CP) are called Cognitive Packet

Networks. Cognitive packets route themselves.

They are assigned goals before entering the net-

work, and pursue these goals adaptively. They learnto avoid congestion and to avoid getting lost or

being destroyed. Cognitive packets learn from their

own observations about the network and from the

experience of other packets with whom they ex-

change information via mailboxes. Cognitive

packets rely minimally on routers, so that network

nodes only serve as buffers, mailboxes and proces-

sors. Each cognitive packet starts with a givenrepresentation of the network from which it then

progressively constructs its own Cognitive Map of

network state and uses it to make routing decisions

[10–12]. Routing packets in ad hoc networks is a

challenge because of the constantly changing

topology of the networks triggered by node

mobility. Thus, when two nodes travel apart, they

may no longer have a direct link between them [32].The concept of CPN shows great potential for

application inmobile networks in general and in the

area of ad hoc networking in particular, since PCNs

rely minimally on existing routing infrastructures

and are highly adaptive to network conditions.

Some protocols can be classified like a Best Effort

protocols, which minimize the network costs in

terms of capacity. Best effort protocols can be di-vided into two classes: proactive and reactive pro-

tocols. Proactive protocols keep track of routes for

all destinations in the network. These protocols have

the advantage that communications with arbitrary

destinations experience minimal initial delays. Reac-

tive protocols acquire routing information only

when it is needed. The reactive class of protocols, as

opposed to proactive protocols, reduces the workrequired to maintain routes as ad hoc network

topologies may change dynamically and arbitrarily.

Destination Sequenced Distance Vector

(DSDV) is a best effort protocol designed specially

for MANETs [21]. It belongs to the class of pro-

active protocols and uses a version of the Bellman–

Ford distributed algorithm adapted to ad hoc

networks. Each mobile station maintains a routingtable which contains routing information for each

node in the network:

• The number of hops necessary to reach a node.

• The sequence number corresponding to a desti-

nation node. This number is used to distinguish

old routes from more recent ones. Hence, this

helps prevent the formation of routing loops.

When a mobile station A initiates communica-

tion with a mobile station B, A can easily reach B

through the information included in its routing

table. In order to update a node’s routing table in

a context of dynamic topological changes, each

node in the network periodically sends an update

message with routing information to its neighbors.Updates depend on two criteria: periodicity of

updates and events occurrence, such as the arrival

of a new node in the neighborhood. Its goal is to

make it possible for a mobile host to locate any

other unit in the network at any time. These up-

dates can be carried out either in a complete or an

incremental way. The main weakness of this pro-

tocol is the enormous bandwidth consumed byrouting traffic control. Furthermore, DSDV is

slow; a mobile unit must wait to receive routing

information to update the corresponding entry in

its routing table.

Ad hoc On Demand Vector (AODV) represents

an improvement of DSDV [22,23]. In fact, in a

nutshell, it takes the advantages of DSDV while

limiting bandwidth consumption. It functions on

demand, i.e., it builds a route to a destination only

if a source node needs to reach it. Each mobile

host operates as a router, and routes are obtained

as needed (i.e., on demand) with little or no reli-

ance on periodic advertisements. AODV provides

loop-free routes even while repairing broken links.

Since the protocol does not require global

periodic routing advertisements, the demand forthe overall available bandwidth is substantially

less than in the protocols that require such

338 A. Quintero et al. / Ad Hoc Networks 2 (2004) 335–349

advertisements. Nevertheless, it can still maintain

most of the advantages of basic distance-vector

routing mechanisms. Although AODV does not

depend specifically on particular aspects of the

physical medium across which packets are dis-

seminated, its development was largely motivatedby limited range broadcast media such as those

utilized by infrared or radio frequency wireless

communications adapters. Routing tables are or-

ganized as to optimize the response time to local

movements and provide quick response times for

requests regarding the establishment of new routes

[23].

Johnson and Maltz [17] proposed a DynamicSource Routing (DSR) protocol for ad hoc wire-

less networks. To send a packet to another host,

the sender constructs a source route in the packet’s

header, with the address of each host in the net-

work through which the packet should be for-

warded in order to reach the destination host.

Then, the sender transmits the packet over its

wireless network interface to the first hop identi-fied in the source route. When a host receives a

packet, if this host does not constitute the final

destination of the packet, it simply transmits the

packet to the next hop identified in the source

route of the packet’s header. Once the packet

reaches its final destination, it is delivered to the

network layer software on that host [17].

Several papers discuss QoS routing in ad hocnetworks [1,9,16,18,27,35]. Quality of service

(QoS) consists of a collection of characteristics or

constraints between a source and a destination

that a connection must guarantee during the

communication to meet the requirements of an

application [2]. QoS routing is a procedure that

identifies the routes, between a source node and its

destination node, which obey the constraints re-quired by the source application and selects be-

tween these routes the one to be used. QoS routing

protocols must work in conjunction with resource

management mechanisms to establish paths

through the network that meet end-to-end QoS

requirements, such as delay, jitter, available

bandwidth, packet loss rate, hop count and path

reliability. Furthermore, a routing algorithmwhich takes QoS into account must also deal with

the maintenance of the routes. Indeed, route fail-

ure probabilities are high in MANETs mainly due

to the mobility of nodes. Wang and Crowcroft [30]

showed that if QoS contains at least two metrics

(e.g., delay, delay jitter, cost), then QoS routing is

an NP-complete problem.

With the increasing number of applicationsrequiring QoS, the success of mobile ad hoc net-

works relies heavily on their ability to provide

routing protocols that take into account QoS.

However, providing QoS in such a dynamic envi-

ronment is not an easy task. Indeed, ad hoc net-

works impose multiple constraints that need to be

addressed: the wireless medium is unreliable, the

topology is entirely dynamic and the computingcapacity of the mobile units is very limited.

Moreover, the available bandwidth is too precious

to authorize heavy use of control messages. Fi-

nally, QoS routing is a distributed problem.

The idea is to provide QoS in a closely knit two-

step process: first, the routing protocol detects the

routes that can fulfill the desired QoS, and, second,

it reserves these routes. Furthermore, it is neces-sary to integrate route maintenance in a QoS

routing protocol so that it can deal with route

breakdowns during communication. The current

literature on the topic suggests reusing the tech-

nologies engineered for traditional networks in

order to preserve certain compatibility, but also to

simplify as much as possible those technologies

and to adapt them to the requirements of ad hocnetworks [7,15,31,34].

Sinha et al. [29] proposed a core-extraction

distributed ad-hoc routing algorithm (CEDAR),

which can react effectively to the dynamics of

ad hoc networks. It includes three main compo-

nents: core extraction, link state propagation and

route computation.

Chen and Nahrstedt [5] presented a ticket-basedprobing algorithm for QoS routing in ad hoc

networks. The idea is to use tickets to limit the

number of candidate paths. When a source node

wants to find QoS paths to a destination, it sends a

message that contains a certain number of tickets.

The number of tickets contained in the message is

a function of the difficulty of finding a QoS path in

the network. When a message is sent on more thanone path, the tickets are divided between the

messages on each path so as the total number of

A. Quintero et al. / Ad Hoc Networks 2 (2004) 335–349 339

tickets in the network is equal to the sum of tickets

on each path.

Wen-Hwa et al. [31] proposed a TDMA-based

bandwidth reservation protocol for QoS routing in

MANETs. They assume a simpler TDMA model

that uses a single common channel shared by allhosts and consider the bandwidth reservation

problem in such environment. A route discovery

protocol that is able to find a route with a given

bandwidth is proposed.

Chen [4] presented a routing algorithm initiated

by the source, a reactive protocol. This protocol

uses local state maintained in each node. Infor-

mation about the global state of the network orthe topology is not necessary. When a source

wants to reach a destination node with given

constraints of service, it initiates a route discovery

procedure by diffusing a request for connection to

all of its neighbors. This connection request is

similar to the one used by AODV.

Hongxia and Hughes [16] proposed an adaptive

QoS routing scheme based on the prediction of thelocal performance in ad hoc networks. It is

implemented using a link performance prediction

strategy. Lower layer parameters are translated

into link state information and used to estimate

the integrated QoS performance in each local area.

Predicting node location is another feasible way

to enhance QoS in mobile networks [28]. However,

this is not suitable for most indoor environments,and only a few existing protocols provide routing-

based location [33].

3. Routing protocol for ad hoc networks

This section presents some of the mechanisms

that can be used to improve QoS communicationin MANETs. The proposed routing protocol is

based on a routing algorithm initiated by the

source that takes into account QoS in terms of

bandwidth consumption.

3.1. Route repair

If a failure occurs during communication be-tween two nodes, two scenarios can be used to

repair the route:

• A global route repair initiated from the source of

communication. This solution is used in most

routing protocols, even if it requires significant

time and consumes much bandwidth.

• A local route repair initiated from a node in theneighborhood of the link where the failure oc-

curred. This local route repair has the advanta-

ges of being fast and consuming little

bandwidth.

Global route repair: If a link failure were de-

tected during the transmission of a packet from

source A to destination B, the node that detectedthe failure would return an error message to the

source. Then, the source initiates a new route

discovery phase to find a new path between A and

B. This phase requires much time to complete and

overloads the network with many routing mes-

sages. This results in bandwidth waste that is

detrimental to the overall performance of the

network.Local route repair: When a node detects a link

failure, it does not systematically send an error

message to the source. First, it attempts to repair

the route itself. It only sends an error message to

the source if this first attempt fails. Consider the

case presented in Fig. 1. Step (a) represents the

route before failure detection. In step (b), Node D

detects a link failure between D and G. Accordingto the local route repair procedure of step (c), D

diffuses a route request (RREQ) packet which is

propagated across the network. When it receives

this packet, J returns a route reply (RREP) packet

to D. This packet is forwarded by I, G and F.

When D finally receives it, the route is repaired

and communication can carry on.

3.2. A new protocol for route discovery

Our objective aims to insure the selection of the

most easily reparable route among those extracted

from the route discovery phase. To achieve this

goal, we recommend taking into account the nat-

ure of the neighborhood of the nodes composing

the network, and in particular the density of nodesand their availability. The reparation of a route in

the case of failure can be carried out through local

route repair.

BA

C

D

E

F

G

H

I

J

BA

C

D

E

F

G

H

I

J

B

A

C

D

E

F

G

H

I

J

BA

C

D

E

F

G

H

I

J

RREQ

RREQRREQ

RREQRREP

RREP

RREPRREP

(a)

(c)

(b)

(d)

Fig. 1. Local route repair: (a) initial route, (b) ruptured initial route, (c) local repair procedure and (d) new repaired route.

340 A. Quintero et al. / Ad Hoc Networks 2 (2004) 335–349

We use the availability parameter to establish

the ability of Node A to replace Node B. The

availability of a node depends on the nature of the

node (laptop, PDA, etc.), the number of packets

that are forwarded by the node on its communi-

cation channels as well as their capacity. To sum-

marize, we can say that the availabilitycorresponds to the available bandwidth of a node

for each of its communication channels. Since each

node needs to be aware of the availability of all of

the other nodes in its neighborhood, our algorithm

provides periodic availability information ex-

changes within the nodes of a neighborhood.

We define the density of a node k as the number

of direct neighbors of k (that is, the number ofnodes in the k radio range) whose available

bandwidth is higher than that required by the

connection. The density parameter is completely

specified by a node and the bandwidth associated

with that node. Thus, upon receiving a route re-

quest, each node evaluates its density by deducting

the number of neighbors whose availability is

higher than that required by the connection.Strangulation occurs when the density of a node in

the route is equal to two (that is, a node for which

the redundancy is null as the only neighbors of the

node are its predecessor and successor nodes on

the route). The higher the instances of strangula-

tion, the more difficult the restoration through

local route repairs.

The discovery phase: We use the route discovery

phase as described in the AODV protocol for

which we add provisions for the availability and

density parameters. These two parameters need to

be taken into account in order for our protocol toprovide QoS. Thus, in our protocol, the source

initiates the routing process upon receiving a con-

nection request. Then, it sends a route request for

this connection to all of its neighbors. The nodes

that receive the message for the first time and that

fulfill the QoS requirements propagates the request

message towards the destination only after having:

• incremented by one unit the number n of nodes

of the route in the message of route request;

• evaluated the density of the neighborhood

(nodes whose availability is sufficient to enable

them to replace a node in case of failure);

• updated the fields average density and strangu-

lation number of the route in construction in

the RREQ packet that is propagated towardsthe destination. The update of the average den-

sity is expressed as follows:

Dm ¼ Dmðn� 1Þ þ Dn

; ð1Þ

BA

C

D

E

F

G

H

I

J

RREP

RREQ

BA

C

D

E

F

G

H

I

J

RREP

RREQ

Fig. 2. Local route repair.

B

A

C

D

E

F

G

H

I

J

Routage

RERR

Fig. 3. Global route repair.

A. Quintero et al. / Ad Hoc Networks 2 (2004) 335–349 341

where Dm is the average density along the route,

D is the density, and n the number of nodes in

the route.

The request message is gradually propagatedtowards the destination following the aforemen-

tioned scenario. Finally, when it arrives at its

destination, the destination node initiates a

countdown and records all of the incoming mes-

sage requests. Each of these messages corresponds

to a detected route between the point of failure and

the destination. At the end of the countdown, the

destination selects the easiest route to repair andthen sends a confirmation packet to reserve the

needed resources.

To select a route, the parameters contained in

each message request that arrived at the destina-

tion are necessary. It is important to note that a

route containing long sequences of high density

nodes will be easier to repair with the local route

repair procedure than a route that does not holdthat property. The repair fitness F associated with

a given route is expressed as

F ¼ Dm

nþ 1þ eð5=2� 1=2nÞ ; ð2Þ

where e is the number of strangulations on the

route. We select the route whose repair fitness is

the highest among all routes.

We now present a theoretical justification to

improve on the protocol described above. The

mean duration of the route repair procedure for

a given route and a given link failure was evalu-ated.

Duration of a local route repair: Let s be the

propagation delay of a notification message from a

node to another. Let Trl;i be the duration of a local

route repair at node i, where i is the distance be-

tween the source of the communication and the

node that is responsible for the route repair (Fig.

2). Trl;i can be expressed as

Trl;i ¼ 2sðn� iÞ: ð3Þ

Duration of a global route repair: Let Trg;i be theduration of a global route repair started at node i(Fig. 3). Trg;i can then be expressed as follows:

Trg;i ¼ s½iþ 2n�: ð4Þ

Duration of a route repair for any node i: Let Tibe the duration of a route repair for node i and let

pi be the success probability for a local route repairat node i. We have

pi ¼ 1� en: ð5Þ

Thus, the probability of a successful local route

repair decreases with the strangulation number,

although it is independent of the position i wherethe local route repair was initiated.

Since we first proceed by systematically initiat-

ing a local route repair, and if this route repair is

unsuccessful, we initiate a global route repair, we

can easily deduce that

Ti ¼ pi � Trl;i þ ð1� piÞ � ½Trl;i þ Trg;i�: ð6ÞMean duration of a route repair: we assume that

the failure probability of a node is independent of

its position on the route. If we let T be the mean

duration of a local route repair, we now have

T ¼ sn

Xn�1

i¼0

2ðnh

� iÞ þ enðiþ 2nÞ

i; ð7Þ

thus

T ¼ s½ðnþ 1Þ þ eð5=2� 1=2nÞ�: ð8Þ

342 A. Quintero et al. / Ad Hoc Networks 2 (2004) 335–349

We notice that T increases as e, the number of

strangulations and n, the length of the route, in-

crease. This observation is confirmed by the defi-

nition of the repair fitness stated in Eq. (3). Indeed,

the repair fitness associated with a route is in-versely proportional to these two parameters.

If a node detects that the link between itself and

its successor is broken for a given route, it may be

useless, even a waste of time, to proceed to a local

route repair. For example, imagine that the density

of the aforementioned node is two or less, meaning

that the only neighbors of the node are its prede-

cessor and successor nodes on the route. Theprobability of success of a local route repair initi-

ated by the node that detected the link failure is

almost null. Indeed, given its density, we know the

node does not have any neighbors in its neigh-

borhood, i.e., in its transmission range, that can be

used to relay the communication to the destina-

tion. In this case, we entrust the source with the

route repair procedure which is simply a globalroute repair.

4. Implementation and results

The implementation of AODV used under

OPNET Modeler was developed in March 2001 by

the National Institute of Standards and Technol-ogy (NIST) according to version 08 of the Internet

draft published by the Mobile Ad Hoc Network-

ing Working Group (MANET-WG) of the Inter-

net Engineering Task force (IETF). It is of

primary importance to specify that the version of

AODV used already supported local route repairs.

We implemented the following two modifications:

• selection of the most robust route;

• systematic global route repair if the density is

insufficient to ensure a successful local route re-

pair.

4.1. Implementation of the protocol

The protocol includes four types of packets thatcan be exchanged between the topology nodes:

RERR, REEQ, RREP and DATA.

RERR: This type of packet is used to indicate

errors. RERR packets are sent to the source of a

communication to signal that the route to the

destination is broken and that the local repair

mechanism, built in the protocol, has failed to

restore the route. When a source node receives thiskind of packet, it immediately invalidates the sta-

tus of the route corresponding to the destination,

interrupts packet transmission to that destination,

and places them in a waiting queue.

RREQ: These packets are generated when a

source node needs to transmit data to a certain

destination, although active routes to that desti-

nation are unavailable. The source broadcastsRREQ packets to all of its neighbors which in

turn, broadcast them to all their neighbors until

either a route to the destination node is found or

the packet TTL field expires.

RREP: When a RREQ packet finally reaches

the destination node, a RREP packet is generated

and sent back to the node that initiated the route

discovery process by reversing the route that hasjust been found. Upon receiving such a packet, the

source node can start transmitting data.

DATA: These packets simply carry the data

from the source node to the destination node.

4.2. The finite state machine model of the protocol

A finite state machine is defined by states andtransitions between those states. Two types of states

are possible: blocking and non-blocking states.

When the machine enters a non-blocking state, a

transition occurs automatically and the machine

immediately exits that state. On the other hand, a

transition from a blocking state can only be trig-

gered by a specific event defined by the designers of

the machine. As Fig. 4 clearly shows, the finite statemachine modeling the AODV protocol revolves

around nine states amongst which, only one is a

blocking state.

The initial state of the machine, or Init, initial-

izes the state variables of each of the mobile units

in the network. Furthermore, the Init state also

specifies the first few interruptions that will be

used to broadcast HELLO messages to a node’sneighbors. The only blocking state of the machine

(Idle), corresponds to a state where the system is

Fig. 4. Finite state machine of the AODV protocol.

A. Quintero et al. / Ad Hoc Networks 2 (2004) 335–349 343

idle. Each transition to a non-blocking state nec-

essarily passes through the Idle state. The Den-

sity_Timeout state, shown on the top of Fig. 4,

periodically calculates the density state variable.

The density of a node is evaluated by couting thenodes in its neighborhood whose availability is

greater than 100. The availability of a node can be

found in the neighbors table which is another state

variable. Once the density is evaluated, the neigh-

bors table is re-initialized to take into account only

the nodes who broadcasted their availability in the

specified time interval. Before exiting the Den-

sity_Timeout state, an interrupt that will triggerthe next density evaluation is programmed.

Our protocol implementation includes a mech-

anism that periodically exchanges the nodes

availability between the nodes of a neighborhood

and calculates the resulting density for each node.

The state variables associated with each mobile

unit in the network are

• Availability: since AODV does not provide a

bandwidth reservation mechanism, the update

of the availability variable does not depend on

the congestion present in the network. This

variable is parametrized individually for each

node.

• Neighbor_availability: the Neighbor_availability

variable gathers in a table the availability

broadcasted by each node of the neighborhood.Only the latest (since the last update of the den-

sity variable) availability information is stored

in the table.

• Density: the variable is updated periodically

(every 20 s). The time interval chosen to update

the density is a trade off between the quality of

information collected and the overhead im-

posed on the network by exchanging messagesbetween nodes of a neighborhood.

Route selection: The state machine has a non-

blocking state that is responsible for selecting a

route at the end of the route discovery process. A

transition to this state automatically occurs once

the waiting period that follows the reception of the

first RREQ packet at the expired destination. Thiswaiting period is proportional (with a waiting

coefficient x) to the time taken by the first RREQ

packet to propagate from source to destination.

While this waiting period increases the time taken

344 A. Quintero et al. / Ad Hoc Networks 2 (2004) 335–349

by the route discovery process to find a feasible

route, it may also improve the success rate of a

local route repair. Given that Tr is the time taken

to initialize the route discovery process (in a con-

figuration where there is no route selection), it is

possible to modify the duration of the routingprocess using Eq. (9):

Trx ¼ Tr � ð1þ xÞ: ð9ÞPreventing local route repair from a low density

node: Our modified protocol attempts to execute a

local route repair only if the density is strictlygreater than two (density > 2). If the node has a

lower density, a RERR packet is sent back to

the source to indicate that a global route repair

should be undertaken. This condition is added

in AODV’s route repair mechanism: aodv_link_

repair_attempt. If the density of the node which

initiated the local route repair is too low, the route

repair procedure is bound to fail. In this situation,it is preferable to turn over directly to the source

after the detection of a link failure and to proceed

to a global route repair. Hence, we accelerate the

restoration of the route by avoiding the delays

associated with a failed local route repair.

4.3. Example

Here is a detailed example to emphasize the

reasons why the initial protocol was subject to

Fig. 5. The topology us

such modifications. Consider the case of the

topology presented in Fig. 5.

Update of the density: For every HELLO_

INTERVAL time interval, each of the 19 nodes of

the network diffuses a special kind of RREP

message where the destination node is the source.The diffusion of RREP messages is not simulta-

neous for all of the 19 nodes. Each node diffuses its

RREP messages at a random time, thus avoiding a

situation where the network is saturated due to the

simultaneous diffusion of RREP by all of the

nodes.

These RREP messages are diffused by a node to

announce its presence to all its possible neighbors.When the message is created, the node assigns its

availability field the value of its availability state

variable at that moment. When a node receives

such a RREP message, it updates its neigh-

bor_availability[N] table. A new interruption is

programmed for every DENSITY_INTERVAL

time intervals. A node which enters this state

counts the nodes for which the availability storedin its availability table is higher than 100, and as-

signs the density the value it found. Before leaving

this state, we program the next interruption and

reinitialize each entry of the neighbor_availabil-

ity[N] table.

To clearly explain the procedure presented

above, let us illustrates how it works by using the

ed in simulations.

A. Quintero et al. / Ad Hoc Networks 2 (2004) 335–349 345

scenario shown in Fig. 5. Mobile node 6 receives

RREP messages sent by its five neighbors: 14, 16,

13, 11 and 8. It extracts the availability informa-

tion of these nodes and assigns it to the corre-

sponding neighbor_availability[N] table entry.

During the interruption used to calculate thedensity of each node, the node enters in the

DENSITY_TIMEOUT state and determines that

five nodes in the neighbor_availability[N] table

have availability fields higher than 100. The den-

sity of Node 6 is thus equal to 5.

Initial route search: the source node (mobile_

node_0_X) generates data messages in its appli-

cation layer for an application running on thedestination node (mobile_node_5_X). At the net-

work layer of the source, the first data packet is

queued as no route is available for the destination

in its routing table. Then, the source initiates a

route discovery procedure for the destination by

diffusing a RREQ message with the source and

destination fields respectively assigned 0 and 5.

The average density and strangulation number areassigned the local node’s density. The nodes that

receive the RREQ message for the first time up-

date their local database for the destination 0 (by

recording the identity of the next node in the

route) and further propagate the RREQ after

having updated the average density and incre-

mented the strangulation number according to

their local density.Thus, Node 6 (mobile_node_6_X) receives a

RREQ message for destination 5 that has been

forwarded by Nodes 17 and 14. In this RREQ

message, the average density field has value 3 while

the strangulation number has value 1. Node 6 adds

in its database an entry for the destination 0 by

indicating that the next node for this destination is

14. Then, it updates the average density field byaffecting it the value 3.5 and the strangulation

numbers field by assigning it the value 1. Finally,

the RREQ message can finally be sent.

The first message that arrives at the destination

is the one that was forwarded by Nodes 1, 2, 3 and

4. A pointer on this message is stored in the first

element of the RREQ_received[0] table. We also

evaluate the repair fitness of the correspondingroute using the formula established in the previous

section. We obtain 0.095. This grade is stored in

the first element of the repair_fitness[0] table.

Then, the Nb_RREQ_received[0] variable is

incremented, which makes it possible to point to

the following element in the table if a new RREQ

message arrives. A countdown, whose value de-

pends on the delay taken by the packet to reachthe destination, is launched. During this interval,

the destination stores each new RREQ packet that

reaches it in the next elements of the table fol-

lowing the same process. Hence, a second RREQ

packet that was forwarded by Nodes 17, 14, 6, 8,

12 and 7 is processed in the same manner by the

destination. The length of the detected route is

equal to 7. The repair fitness associated to thisroute is: 0.28 (average density: 3.625 and 2 stran-

gulations).

At the end of the countdown period, an inter-

ruption is generated. We determine, among the

two repair fitness value stored in the repair_fit-

ness[0] table, the one with the highest value. We

identify the number of the corresponding route: 2.

The pointer stored in RREQ_received allows thegeneration of the RREP message which will be

sent back to the source (mobile_node_0_X). Each

intermediate node that receives the RREP message

updates its local database towards Destination 5

by recording the next hop in the route towards

Node 5. When the RREP message arrives at the

source, the queued data packets can be sent to the

destination along this new route.Local route repair: Let us now have a look at

the route repair procedure in the event that Node

12 fails. Node 8, which precedes Node 12 on the

route, detects its failure (at the MAC layer) as it

does not receive the acknowledgement from 12

that necessarily follows the reception of a data

message. A NACK packet is then returned to the

network layer. Node 8 queues the correspondingdata packet and attempts a local route repair. In-

deed, its density is higher than two. It starts by

passing the current status of the route towards 5 to

UNDER_REPAIR in the routing table so that

new data packets that arrive at later time are

queued. Node 8 makes provision for the case

where the local route repair does not succeed by

initializing an interruption timer. Then, it sendsRREQ packets towards 5 following the same

principle as the one used for the initial routing, but

346 A. Quintero et al. / Ad Hoc Networks 2 (2004) 335–349

with the notable difference that this time the

Density_Mode field is switched off. The messages

are thus propagated through the network without

taking into account the average density and the

strangulation number. The first RREQ that arrives

at destination triggers the generation of a RREPaddressed to Node 8. This packet is returned back

to Node 8. Once the packet arrives at Node 8, the

status of the corresponding route is switched back

to ACTIVE and the queued packets are sent to the

destination along the new route.

4.4. Experiments and results

In our approach, we want to benefit from the

heterogeneity of the distribution of the mobile

units in the network by taking into account the

density parameter. We consider that it would be

useless to test our improvements on a statistically

homogeneous random configuration of mobile

units. Nevertheless, we carried out several simu-

lations in such configurations to conclude that, forthese configurations, our contribution does not

increase (nor decrease) the performance of the

protocol compared to its traditional implementa-

tion. The duration of simulations was set to four

minutes each. To test our assumptions, we plan the

‘‘departure’’ of a given node at a precise moment;

many nodes may depart during a simulation, each

at a predetermined moment. We initiated a nodedeparture from the network by moving it out of

the range of every other mobile unit in the net-

work. Then, we analyze the network evolution,

Fig. 6. ETE delay––node departure

more specifically the ensuing route repair proce-

dure.

We observed the evolution of the network fol-

lowing the departure of different mobile nodes.

For each simulation, we consider the following

three cases:

Case 1: AODV without taking into account the

density parameter and without local route

repair.

Case 2: AODV with local route repair (original

AODV protocol).

Case 3: Our improved AODV protocol.

The results presented correspond to average

results on a series of 100 simulations. We do not

consider the cases where the source and/or desti-

nation nodes fail, since this case cannot be medi-

ated through a route repair procedure. We assume

that each node in the network has the same failure

probability.

Fig. 6a shows the end-to-end (ETE) delays forCases 1 and 2, and Fig. 6b shows the ETE delay

for Case 3. Table 1 summarizes the results for the

node departure. Table 1 also indicates that a local

route repair was undertaken successfully. We can

further note that this local route repair lasted

0.00782 s, compared to the 0.018 s needed for the

total route repair obtained in Case 1. Hence, we

deducted that our protocol clearly improves thetime required to complete a route repair, com-

pared to Case 1. Table 2 summarizes the global

results.

: (a) cases 1 and 2, (b) case 3.

Table 1

Numerical results for node departures

Case 1 Case 2 Case 3

Number of lost packets 0 0 0

Duration of the initial routing (s) 0.00855 0.00855 0.022013

Duration of local route repair (s) 0 0 0.00782

Total duration for route repair (s) 0 0 0.00782

Average ETE delay before route repair (s) 0.0095 0.0095 0.014

Average ETE delay after route repair (s) 0.0095 0.0095 0.016

Average route length (nodes) 5 5 8.261543

Table 2

Global results

Case 1

(AODV without local

repair)

Case 2

(AODV with local

route repair)

Case 3

(AODV with route

selection)

Number of lost packets 0.6 1.8 0

Duration of the initial routing (s) 0.00878 0.008632 0.0205

Total duration of the route repair (s) 0.039 0.22561 0.0058

Average ETE delay before route repair (s) 0.0095 0.0095 0.014

Average ETE delay after route repair (s) 0.119 0.0113 0.0146

Average route length (nodes) 5.63 5.47 7.52

A. Quintero et al. / Ad Hoc Networks 2 (2004) 335–349 347

We notice in this table that Case 3 (i.e., the caseutilizing our improved version of AODV) cancels

completely the packets loss rate normally associ-

ated with the departure of a node in the network.

This result puts forward the increased reliability of

our routing protocol as compared to the cases

where local repair is not implemented nor realized

with the selection of the most robust route (Cases 1

and 2). The departure of a given node of the net-work generates an average loss rate of 0.6 data

packet for case 1 and 1.8 data packets for Case 2.

However, we notice that the duration of the

initial routing procedure is systematically and sig-

nificantly longer for Case 3. On average, this initial

routing time is increased by 137% compared to

Cases 1 and 2. That consists in the main disad-

vantage of our improvements. Longer time is nec-essary to establish an initial route as a reparable

route is selected. Furthermore, the average dura-

tion of a local route repair caused by the departure

of a node in the network (Case 3) is far better than

in Cases 1 and 2. Indeed, our local route repair

procedure is on average 6.7 times shorter than in

Case 1, and 38 times shorter than in Case 2.

Another weakness of our improved protocol isthe average length of the initial route. We notice

that the average length of a route in Case 3 is 7.52nodes while the average length of a route is

approximately 5.5 nodes long for the two other

cases.

Finally, the ETE average delay of data packets

is longer in our improved version of AODV than

in the original one. For Case 3, the ETE delay

before a failure is 0.014 s compared with 0.0095 s

for the two other cases. On the other hand, we seethat this delay does not increase significantly after

the occurrence of a failure and remains relatively

stable at about 0.0146 s, compared to the delays

observed for Cases 1 and 2, which are 0.0119 and

0.0113 s respectively. Thus, we can conclude that

the ETE delay obtained in Case 3 is more stable

than that the ones observed in the other two cases.

5. Conclusion

The routing method presented in this paper aims

to improve QoS management in MANETs by

taking into account the density of a node, defined as

the number of mobile units available in the radio

range of the node. Our approach was based on athorough analysis of the available mechanisms and

348 A. Quintero et al. / Ad Hoc Networks 2 (2004) 335–349

tools that take into account quality of service in ad

hoc networks. Then, we introduced the concept of

density and described how the network could ex-

ploit this information to improve the QoS offered.

The method can be subdivided into two parts: a

selection mechanism that chooses the most robustroute available, and a mechanism to forecast the

failure of a local route repair before it occurs.

The route selection mechanism described aims

at selecting among several routes, the one whose

maintenance is the easiest to realize. The simula-

tion results confirmed our hypothesis: the duration

of a route repair after a failure is improved. In

addition, our mechanism drastically reduced thedata packet loss rate.

We also highlighted a situation where an at-

tempt to repair a route locally could be useless and

even harmful for the network: if the density

around the node which initiated the local route

repair is too low, the route repair procedure is

bound to fail. In this situation, it is preferable to

turn over directly to the source after having de-tected the failure of the link and to proceed to a

global route repair. To validate our work, we

implemented our improvements of the AODV

protocol in Opnet Modeler. Then, we tested our

protocol for a given configuration. The results

obtained are encouraging: the data packet loss rate

is strongly reduced compared to the initial version.

In addition, the time required to complete a localroute repair following a failure decreased signifi-

cantly.

References

[1] G. Aggelou, R. Tafazolli, QoS support in 4th generation

mobile multimedia ad hoc networks, in: Second Interna-

tional Conference on 3G Mobile Communication Tech-

nologies, 2001, pp. 412–416.

[2] S. Chakrabarti, A. Mishra, QoS issues in ad hoc wireless

networks, in: IEEE International Conference on Commu-

nications, February 2001, pp. 142–148.

[3] S. Chakrabarti, A. Mishra, QoS issues in ad hoc wireless

networks, IEEE Communications Magazine 39 (2) (2001)

142–148.

[4] S. Chen, Routing support for providing guaranteed end-to-

end quality of service, Ph.D. dissertation, Univ. of Illinois

at Urbana-Champaign, 1999.

[5] S. Chen, K. Nahrstedt, Distributed quality-of-service

routing in ad hoc networks, IEEE Journal on Selected

Areas in Communications 17 (8) (1999) 1488–1505.

[6] R. Chunhung, L. Jain-hing, QoS routing in ad hoc wireless

networks, IEEE Journal on Selected Areas in Communi-

cations 17 (8) (1999) 1426–1438.

[7] S. Das, A. Mukherjee, S. Bandyopadhyay, K. Paul, D.

Saha, Improving quality-of-service in ad hoc wireless

networks with adaptive multi-path routing, in: IEEE

Conference on Global Telecommunications GLOBE-

COM’00, vol. 1, 2000, pp. 261–265.

[8] D. Dharmaraju, A. Roy-Chowdhury, P. Hovareshti, J.

Baras, INORA––a unified signaling and routing mecha-

nism for QoS support in mobile ad hoc networks, in:

International Conference on Parallel Processing Work-

shops, 2002, pp. 86–93.

[9] E. Elmallah, H. Hassanein, H. AboElFotoh, Supporting

QoS routing in mobile ad hoc networks using probabilistic

locality and load balancing, in: IEEE Conference on

Global Telecommunications GLOBECOM’01, vol. 5,

2001, pp. 2901–2906.

[10] E. Gelenbe, M. Gellman, P. Su, Self-awareness and

adaptivity for quality of service, in: Eight IEEE Interna-

tional Symposium on Computers and Communications

ISCC’03, 2003, pp. 3–9.

[11] E. Gelenbe, R. Lent, A. Nunez, Traffic balancing via smart

packets, in: 3rd IEEE Workshop on IP Operations and

Management, 2003, pp. 15–21.

[12] E. Gelenbe, R. Lent, A. Montuori, X. Zhiguang, Cognitive

packet networks: QoS and performance, in: Proceedings of

the 10th IEEE International Symposium on Modeling,

Analysis and Simulation of Computer and Telecommuni-

cations Systems, MASCOTS, 2002, pp. 3–9.

[13] D. Ghosh, V. Sarangan, R. Acharya, Quality-of-service

routing in IP networks, IEEE Transactions on Multimedia

3 (2) (2001) 200–208.

[14] D. Goodman, The wireless Internet: promises and chal-

lenges, IEEE Computer 33 (7) (2000) 36–41.

[15] A. Gupta, D. Sanghi, QoS support in mobile ad-hoc

networks, in: IEEE International Conference on Personal

Wireless Communications, 2000, pp. 340–344.

[16] S. Hongxia, H. Hughes, Adaptive QoS routing based on

prediction of local performance in ad hoc networks, in:

IEEE Conference on Wireless Communications and Net-

working, vol. 2, 2003, pp. 1191–1195.

[17] D. Johnson, D. Maltz, Dynamic source routing in ad hoc

wireless networks, in: Computer Communications Review,

Proceedings of SIGCOMM ’96, 1996, pp. 153–181.

[18] S.B. Lee, G.S. Ahn, X. Zhang, A.T. Campbell, INSIG-

NIA: An IP-based quality of service framework for mobile

ad hoc network, Journal of Parallel and Distributed

Computing, Special issue on Wireless and Mobile Com-

puting and Communications (2000) 374–406.

[19] S. Nelakuditi, S. Varadarajan, Z. Zhi-Li, On localized

control in QoS routing, IEEE Transactions on Automatic

Control 47 (6) (2002) 1026–1032.

A. Quintero et al. / Ad Hoc Networks 2 (2004) 335–349 349

[20] A. Orda, A. Sprintson, Precomputation schemes for QoS

routing, IEEE/ACM Transactions on Networking 11 (4)

(2003) 578–591.

[21] C. Perkins, P. Bhagwat, Highly dynamic destination-

sequenced distance-vector routing for mobile computer,

in: ACM Conference on Communications Architectures,

1994, pp. 234–244.

[22] C. Perkins, E. Royer, S. Das, Ad hoc on demand distance

Vector routing, IETF, Internet Draft, March 2000.

[23] C. Perkins, E. Royer, Ad hoc on demand distance vector

algorithm, in: IEEE Workshop on Mobile Computing

Systems and Applications WMCSA ’99, 1999, pp. 90–

100.

[24] D. Perkins, H. Hughes, A survey of quality-of-service

support for mobile ad hoc networks, Wireless Communi-

cations and Mobile Computing 2 (2002) 503–513.

[25] S. Pierre, Mobile computing and ubiquitous networking:

concepts, technologies and challenges, Telematics and

Informatics 18 (2001) 109–131.

[26] K. Phanse, L. DaSilva, Addressing the requirements of

QoS management for wireless ad hoc networks, Computer

Communications 26 (12) (2003) 1263–1273.

[27] G. Raju, G. Hernandez, Q. Zou, Quality of service routing

in ad hoc networks, in: IEEE Conference on Wireless

Communications and Networking, vol. 1, 2000, pp. 263–

265.

[28] S. Shah, K. Nahrstedt, Predictive location-based QoS

routing in mobile ad hoc networks, in: IEEE International

Conference on Communications, vol. 2, 2002, pp. 1022–

1027.

[29] P. Sinha, R. Sivakumar, V. Bharghavan, CEDAR: a core-

extraction distributed ad hoc routing algorithm, in: IEEE

Infocom’99, vol. 1, New York, 1999, pp. 202–209.

[30] Z. Wang, J. Crowcroft, QoS routing for supporting

resource reservation, IEEE Journal on Selected Areas in

Communications 14 (7) (1996) 1228–1234.

[31] L. Wen-Hwa, T. Yu-Chee, S. Kuei-Ping, A TDMA-based

bandwidth reservation protocol for QoS routing in a

wireless mobile ad hoc network, in: IEEE International

Conference on Communications, vol. 5, 2002, pp. 3186–

3190.

[32] S. Woo, S. Singh, Scalable routing protocol for ad hoc

networks, Wireless Networks 7 (2001) 513–529.

[33] L. Xiao, J. Wang, K. Nahrstedt, The enhanced ticket-

based routing algorithm, in: IEEE International Confer-

ence on Communications, vol. 4, 2002, pp. 2222–2226.

[34] H. Xiao, W.K.G. Seah, A. Lo, K. Chua, A flexible quality

of service model for mobile ad hoc networks, in:

IEEE Vehicular Technology Conference, 2000, pp. 445–

449.

[35] H. Xiaoyan, K. Xu, M. Gerla, Scalable routing protocols

for mobile ad hoc networks, IEEE Network 16 (4) (2002)

11–21.

Alejandro Quintero received the engi-neer degree in computer engineeringfrom Los Andes, Colombia, in 1983.In June 1989, and in 1993, he receivedthe diploma of advanced studies, andthe Ph.D. degree in computer engi-neering, respectively, from the JosephFourier University, Grenoble, France.He is currently an associate professorat the Department of ComputerEngineering of �Ecole Polytechnique deMontreal, Canada. His main researchinterests include mobile computingand 3rd generation wireless infra-

structures. He is the co-author of one book, as well as over 40other technical publications including journal and proceedingspapers. He is a member of ACM and IEEE CommunicationsSociety.

Samuel Pierre received the B.Eng.degree in civil engineering in 1981 from�Ecole Polytechnique de Montr�eal,Qu�ebec, the B.Sc. (1984) and M.A.Sc.(1985) degrees in mathematics andcomputer science from the UQAM(Montr�eal), the M.Sc. degree in eco-nomics in 1987 from the Universit�e deMontr�eal, and a Ph.D. in ElectricalEngineering in 1991 from �Ecole Poly-technique de Montr�eal. He is currentlya Professor of Computer Engineeringat �Ecole Polytechnique de Montr�ealwhere he is Director of the Mobile

Computing and Networking Research Laboratory (LARIM)and NSERC/Ericsson Industrial Research Chair in Next-gen-eration Mobile Networking Systems.

Dr. Pierre is the author of four books, co-author of twobooks and six book chapters, as well as over 240 other technicalpublications including journal and proceedings papers. He re-ceived the Best Paper Award of the Ninth International Work-shop in Expert Systems & their Applications (France, 1989),a Distinguished Paper Award from OPNETWORK’2003(Washington, USA). One of these co-authored books,T�el�ecommunications et Transmission de donn�ees (Eyrolles,1992), received special mention from Telecoms Magazine(France, 1994). He is a Fellow of Engineering Institute ofCanada. His research interests include wireline and wirelessnetworks, mobile computing, performance evaluation, artificialintelligence, and electronic learning. He is an Associate Editorof IEEE Communications Letters and IEEE Canadian Review.He also serves on the editorial board of Telematics and Infor-matics published by Elsevier Science.

Benjamin Macab�eo received M.A.Sc.degrees in Electrical Engineering in2003 from �Ecole Polytechnique deMontr�eal. His research interests in-clude wireless networks and ad-hocnetworks.