LOCALISED AD HOC ADDRESS AUTOCONFIGURATION

VIA CLUSTERING IN NOMADIC AD HOC NETWORKS

George Metaxas, David Hutchison and Nicholas J.P. Race

Computing Department, InfoLab21

Lancaster University, Lancaster LA1 4WA, UK

Keywords:

Address Autoconﬁguration, Nomadic Ad hoc Networks, Wireless Networking, IPv6.

Abstract:

Densely populated urban environments provide unique nomadic movement patterns. People tend to move in

groups across predeﬁned paths, with low intra-group movement, giving rise to Nomadic Ad hoc Networks

(NANETs). NANETs, which are a constrained subset of Mobile Ad hoc Networks (MANETs) in terms of

mobility, concentrate on civilian-style usage and are sufﬁciently dynamic in nature to present signiﬁcant chal-

lenges in the design of even the most common tasks such as packet routing or name resolution. An intriguing

issue relates to the automatic conﬁguration of IP addresses to constituent nodes that will allow data exchange.

The underlying physical group structure can be utilised in the network layer as a basis for an address autocon-

ﬁguration mechanism, which would yield low and localised signalling trafﬁc overhead coupled with a subnet

optimisation. This paper proposes and discusses the AALM IPv6 address autoconﬁguration mechanism, which

has been developed to provide the aforementioned characteristics.

1 INTRODUCTION

Nomadic Ad hoc Networks (NANETs), ﬁrst ex-

plored in (Prince, 2005), provide a relatively fresh

view to the generic ﬁeld of Mobile Ad hoc Networks

(MANETs). The inherent nomadic nature constrains

the movement patterns of constituent nodes to a tacit

model found in densely populated urban environ-

ments. People tend to move in groups across pre-

determined paths, pausing at places of interest for an

undetermined amount of time. Intra-group mobility

is considered low, whereas group mobility in respect

to the environment may be high. NANETs present an

environment found in crowded city centres, univer-

sity campuses and shopping centres, which are typ-

ical civilian-style rather than military or emergency

response scenarios.

Although constrained in relation to MANETs,

NANETs are still highly dynamic ad hoc networks

and exhibit several MANET traits, such as lack of a

centralised infrastructure. Consequently, basic tasks

such as packet routing, name resolution and automatic

address conﬁguration require cooperative approaches.

The most widely researched subject is associated with

packet routing. Several experimental standards have

been produced by the IETF MANET Working Group,

which is responsible for engineering such solutions.

Aside from packet routing, address autoconﬁgura-

tion provides an area of signiﬁcant challenges. In

wired and wireless infrastructure networks, this is

supported by infrastructure means, such as address

pool servers. In a MANET, it is impossible and un-

wise to presume the presence of infrastructure sup-

port. Therefore, address autoconﬁguration requires

node cooperation and the design of distributed algo-

rithms capable of rapidly adapting to environmental

conditions. The constrained and specialised nature of

mobility in NANETs, can be exploited to provide bet-

ter solutions and reveal issues, otherwise hidden due

to unconstrained and unnatural movement patterns.

This paper proposes an address autoconﬁguration

mechanism for NANETs, which has been fundamen-

tally designed to operate in partitioned networks, by

localising signalling trafﬁc, through the use of cluster-

ing concepts. The added advantage of this approach

is the subnet optimisation which minimises routing

table size. The remainder of this paper is organ-

ised as follows. Section 2 discusses related work in

the ﬁeld of address autoconﬁguration mechanisms for

MANETs and details various physical layer issues.

Section 3 presents the design of the proposed mecha-

nism. Section 4 provides an evaluation of the mecha-

121

Metaxas G., Hutchison D. and J.P. Race N. (2006).

LOCALISED AD HOC ADDRESS AUTOCONFIGURATION VIA CLUSTERING IN NOMADIC AD HOC NETWORKS.

In Proceedings of the International Conference on Wireless Information Networks and Systems, pages 121-128

 SciTePress

nism through extensive simulations. Finally, Section

5 concludes the paper and discusses future work.

2 RELATED WORK

Address autoconﬁguration mechanisms became a re-

quirement in wired networks as a result of the growth

and popularity of the Internet. They are classiﬁed in

two categories, statefull and stateless. The former

requires the presence of a centralised server for dis-

tributing addresses, as in DHCP (Droms, 1997). The

latter achieves address conﬁguration per node atom-

ically, without contacting a centralised server. The

IPv6 Stateless Address Autoconﬁguration (SAA),

deﬁned in (Thomson and Narten, 1998), is such a

mechanism. Nodes are equipped with link local and

global addresses. SAA requires a router advertising

a unique subnet preﬁx in every subnet. A link lo-

cal address is formed by converting the MAC address

of a network interface to a 64-bit identiﬁer and ap-

pending it to the link local preﬁx (FE80::). Address

uniqueness on the link is veriﬁed through a broadcast

message, which initiates the Duplicate Address De-

tection (DAD) process. A reply signals duplication

causing the repetition of the procedure. Otherwise,

the address is considered unique and a global address

is formed by appending the unique interface identiﬁer

to the subnet preﬁx.

Pure ad hoc networks require stateless approaches,

which are classiﬁed in three categories, conﬂict avoid-

ance, conﬂict detection and hybrid. In conﬂict avoid-

ance mechanisms, addresses are acquired through a

conﬁgured neighbour which possesses an unallocated

address pool. In contrast, nodes atomically acquire

addresses in conﬂict detection mechanisms and at-

tempt to detect and resolve possible duplications.

Hybrid solutions provide a fusion of the above ap-

proaches. Several ad hoc address autoconﬁguration

mechanisms are presented in Section 2.1.

At present, software solutions developed for use

in ad hoc networks are impeded by various physical

layer issues. These issues are mainly concerned with

scalability and partition formation. They are inher-

ent in two of the most popular wireless technologies

used for constructing ad hoc networks, IEEE 802.11

and Bluetooth. Section 2.2 discusses some of these

issues.

2.1 Address Autoconﬁguration In

Ad Hoc Networks

An initial attempt to provide a MANET address auto-

conﬁguration mechanism, similar to the IPv6 SAA is

described in (Perkins et al., 2000). An unconﬁgured

node selects a random IPv4 address from a speciﬁc

range, broadcasts a message requesting a route to the

generated address and sets a timer. If the timer expires

and no replies are received, the process is repeated up

to a pre-speciﬁed number of times. Further lack of

replies, is assumed to reveal that the address is unal-

located and can be assigned to the node. Otherwise

the address is in use and the process is repeated.

(Vaidya, 2002) terms the above idea as Strong

DAD and argues that it fails if message delay between

any pair of nodes is unbounded. The proposed Weak

DAD technique allows routing protocols to correctly

deliver packets even in the presence of duplicate ad-

dresses. This is achieved by associating a predeﬁned

unique key with each node, such as an IMEI

, dis-

tributed and stored alongside IP addresses to enable

unique node identiﬁcation when required.

(Zhou et al., 2003) uses a common address pool

which is drawn from the results of a carefully chosen

function that generates sequences with a low proba-

bility of producing the same numbers. The ﬁrst ini-

tialised node chooses the function and the seed to be

used. It is called a prophet as a result of its ability to

predict all generated addresses. Unconﬁgured nodes

request the help of their closest conﬁgured neighbour,

which creates an address using the selected function

and returns it, along with the seed and the function in

use, to the requester.

Another technique called Passive Duplicate Ad-

dress Detection (PDAD) is discussed in (Weniger,

2003) and (Weniger, 2005). It is a passive approach,

involving no explicit signalling trafﬁc transmission,

deriving its decisions from routing protocol trafﬁc in-

vestigation. Nodes are conﬁgured with addresses that

are unlikely to be duplicate, deriving their decisions

through a probabilistic algorithm, which is aided by

an Address Allocation table maintained per node.

A common issue in MANETs is the occurrence

of partitions, which are collections of nodes separat-

ing from or merging into the network. Address au-

toconﬁguration mechanisms handle partitions by as-

sociating them with identiﬁers and providing dupli-

cate address resolution in case of partition merge,

as in (Sun and Belding-Royer, 2003), (Nesargi and

Prakash, 2002), (Zhou et al., 2003) and (Toner and

O’Mahony, 2003). A node is chosen to create, dis-

tribute, maintain and advertise the partition identi-

ﬁer. Partition splits and mergers are identiﬁed by the

lack of partition identiﬁer advertisements or the re-

ception of multiple different advertisements respec-

tively. This approach is cumbersome, generates ex-

cessive signalling trafﬁc and does not resolve address

duplications across multi-hop partitions. Moreover,

partition identiﬁers are unrelated to the network oper-

ation and introduce additional bandwidth and routing

A pre-conﬁgured unique 15-digit number used in GSM

networks

table size overheads.

2.2 Physical Layer Issues

One of the major issues inhibiting the wide deploy-

ment of MANETs are the inherent physical layer is-

sues in the most popular underlying networking tech-

nologies. These issues result in the creation of parti-

tions and impact the scalability of the resulting net-

works. Partitions are mainly manifested when several

nodes become separated from the rest of the network,

as a result of physical movement or due to wireless

channel conditions. The existence of partitions pro-

hibits proper operation of distributed algorithms due

to the lack of routing paths. Even though partitions

can be treated as separate networks, the presence of

mobility and freedom of movement leads to partition

merging and splitting.

Wireless channel conditions are the most impor-

tant contributer to partitioning, apart from physical

node movement. This is especially evident in the

IEEE 802.11 protocol. (Lundgren et al., 2002) reports

the problem of gray zones, which are areas where

nodes are able to exchange routing information but no

data. The problem of unidirectional links discussed

in (Prakash, 1999), can also lead to partitions, espe-

cially if such a link exists between key routing paths

in the network. Even the number of people impacts

the channel quality as discussed in (Mathur et al.,

2004).

Ad hoc networks are also plagued by scalability

problems impeding their multihop ability, as a re-

sult of various physical layer issues. (Tschudin et al.,

2003) and (Tschudin et al., 2004) discuss the Ad hoc

Horizon problem which exists in IEEE 802.11. It dic-

tates that beyond two to three hops, routing informa-

tion becomes useless, route maintenance and discov-

ery inhibit network operation and TCP performance

becomes unacceptable. Bluetooth is known for its

difﬁculty in scaling due to oversights in its speciﬁ-

cation, as discussed in (Guerin et al., 2002), regard-

ing the scatternet formation procedure used for creat-

ing larger Bluetooth networks. A number of research

studies propose different algorithms for this purpose,

such as for example (Zaruba et al., 2001) and (Law

and Siu, 2001). However, it is doubtful whether Blue-

tooth is scalable, due to the increased overhead re-

quired for scatternet creation, as discussed in (Mikl

et al., 2000) and (Vergetis et al., 2005).

3 THE AALM MECHANISM

Partitioning is a natural tendency of MANETs and

NANETs due to physical movement and wireless

channel conditions. Especially in the case of

NANETs, the existence of partitions is a highly likely

occurrence due to the underlying nomadic behav-

iour. However, current address autoconﬁguration pro-

posals treat partitioning as a special case and con-

sider generic MANET movement scenarios, which

are mostly unrealistic and do not reﬂect civilian-style

scenarios. The Ad hoc Address autoconﬁguration

Localisation Mechanism (AALM) is a novel ap-

proach in the ﬁeld. It has been designed around IPv6,

reuses the ICMPv6 signalling mechanisms and at-

tempts to tailor the SAA procedure for use in parti-

tioned NANETs.

Partitioning is fundamental to the design of

AALM, which combats this issue and provides a scal-

able, low overhead solution. This is achieved through

the use of Network layer clusters, or groups as they

are referred to in this paper, which are at most two

hops in diameter. This technique also achieves in-

creased scalability by reducing the required routing

table size, as groups are considered to be small scale

subnets. Address uniqueness is highly localised in

small regions, up to a maximum of six hops. Con-

sequently, the impact of address duplications and par-

tition creations is localised to a very small network

part, thereby minimising the signalling trafﬁc over-

head. This localisation of signalling trafﬁc governs

all aspects of address generation and maintenance.

All packets are transmitted with a hop count of one

unless otherwise stated.

3.1 Assumptions

Several assumptions are required in the design of the

AALM mechanism. The underlying environment is

assumed to be constrained in terms of mobility and

node movement, following the nomadic model, where

nodes move in groups and intra-group mobility is

considered low. NANETs concentrate on civilian-

style scenarios found in settings such as city centres.

Therefore, it is assumed that group movement with

respect to the environment and other groups is con-

sidered low. Low mobility is similar to the walking

speed of people. Constituent devices are considered

to be small mobile devices, such as mobile phones,

PDAs or laptops.

Each device must be equipped with one wireless

ad hoc interface and one wireless infrastructure inter-

face

. The interface operating in infrastructure mode

is assumed to be assigned a globally unique IPv6 ad-

dress which remains unchanged. Furthermore, each

ad hoc interface is associated with a unique MAC

address. For the remainder of this paper, the abbre-

viations NSOL, NAD, RAD and RSOL specify the

ICMPv6 Neighbour Solicitation, Neighbour Adver-

For example IEEE 802.11 and GPRS, found in some

mobile phone models

Figure 1: The link and extended link concepts.

tisement, Router Advertisement and Router Solicita-

tion messages respectively.

3.2 Founding Concepts of AALM

AALM deﬁnes two neighbour sets, the link and the

extended link. A link is: the set of all one hop neigh-

bours of a node, whereas an extended link is: the

set of all one hop neighbours of a node plus their

one hop neighbours. The link deﬁnition resembles

the corresponding deﬁnition used in wired networks.

An extended link represents a node’s two hop neigh-

bourhood. Nodes acquire unique link local addresses

that are cooperatively maintained in their extended

link. Precise knowledge of the extended link is not

required, but provides a safety zone for link local ad-

dress uniqueness. Figure 1 shows the difference be-

tween the two constructs.

The SAA speciﬁcation requires nodes to defend

themselves in case of address duplication in their link.

Reception of a message on a speciﬁc multicast ad-

dress, after the completion of the SAA procedure, per-

mits the receiver to defend its address by transmitting

an appropriate NAD. Nodes using the AALM mecha-

nism are also required to respond to any potential du-

plications, but also defend their link neighbours. This

is necessary due to a node’s partial perception of the

network topology. Maintaining address uniqueness in

a node’s link requires neighbour cooperation, as their

link is part of the defended node’s extended link. All

address defence messages are transmitted to the All

Nodes in Link Multicast (FF02::1) address.

There are two address types assigned by the AALM

mechanism, link local and site local. Link local ad-

dresses are of limited scope, valid for one hop com-

munications only. Their uniqueness is maintained in

a node’s extended link through regular beaconing and

neighbour cooperation. This cooperation is enforced

by transmitting all beaconing, address acquisition and

maintenance related messages to the IPv6 broadcast

address, for inspection by all neighbours. All such

messages are transmitted with a hop count of one.

Upon acquiring link local addresses, group forma-

tion is initiated to equip nodes with site local ad-

dresses. AALM uses the term group rather than clus-

ter to differentiate from clustering algorithms, which

are usually implemented at the MAC layer. Each

group is associated with a 32-bit Group Identiﬁer

(GID) that is embedded in the site local addresses of

group members. GID uniqueness is maintained across

neighbouring groups. There are three group member

types, a leader, gateways and ordinary nodes. Groups

are centred around a leader and all group members are

one hop neighbours of a leader. Therefore, all group

members are at most two hops away. Nodes record

neighbour information in their Neighbour Cache and

group information in their Preﬁx and Router Caches.

3.3 Link Local Address Acquisition

The link local address acquisition procedure, is simi-

lar to the DAD procedure of SAA, although there are

several distinct features associated with the AALM

mechanism. The 64-bit interface identiﬁer is ran-

domly generated, rather than being associated with

the MAC address, for enhancing privacy. The NSOL

carrying the generated address is transmitted to the

All Nodes in Link Multicast address (FF02::1) to

promote cooperation among neighbours in the con-

ﬁguration procedure. Receiving nodes must record

this information in their Neighbour Cache, unless a

duplication is detected. The initial NSOL is transmit-

ted exactly once and the node must also set a timer

to signal the end of the DAD procedure. This event

is followed by the initialisation of a pseudo-periodic

beaconing service to advertise a node’s link local ad-

dress, which in AALM terms has been acquired by

the node that is now considered to be conﬁgured.

MAC address comparison is used for resolving du-

plications, with the lowest one favoured. The pre-

ferred node is also determined by whether either of

the offending nodes has acquired the address. If both

or none have acquired the address, only MAC address

comparison is performed. Otherwise, the node which

is conﬁgured or presumed to be conﬁgured is allowed

to maintain ownership of the address, irrespective of

MAC address comparison. Furthermore, if address

duplication occurs after group formation, involving a

leader and a non leader node, the leader is favoured.

All signalling and data trafﬁc containing link local ad-

dresses are searched in the Neighbour Cache for du-

plications.

The favoured node is not defended immediately,

unless it is also the one detecting the duplication. In

all other cases an appropriate reply packet is created

and queued for future transmission. The queueing in-

terval is random for each node, but does not exceed

one hundred milliseconds. Reception of a similar re-

ply packet, causes the receiving node to cancel its

queued transmission if it is in agreement with the pro-

vided information. Otherwise, it re-evaluates its deci-

sion and proceeds to update the contents of its queued

packet, but not the scheduled transmission time. This

procedure minimises the number of defence packets

transmitted and reduces wireless channel contention

and the number of dropped packets. Assuming two

conﬂicting nodes, only one packet is necessary to re-

solve a duplication.

3.4 Group Formation and Site Local

Address Acquisition

The group formation procedure is based on the

Lowest-ID and Least Cluster Change (LCC) clus-

tering algorithms, discussed in (Ephremides, 1988)

and (Chiang, 1997) respectively. Lowest-ID is used

in leader election, whereas LCC is used as a means

to increase group stability by not destroying groups

whenever a member with a lowest identiﬁer from the

leader becomes a group member. As previously men-

tioned, the Preﬁx and Router Caches record informa-

tion about groups. The former stores preﬁxes adver-

tised by group leaders or gateways, whereas the latter

stores leaders addresses. Prior to the group formation

procedure, nodes investigate their Preﬁx and Router

Caches to identify groups in the vicinity, in which

case they associate themselves with such a group, be-

coming members. Otherwise, group formation is ini-

tiated by investigating the contents of the Neighbour

Cache for identifying the node possessing the low-

est MAC address, which is designated as a potential

leader. All other nodes set a timer and await for a

leader to make its presence known. This is required

in case a potential leader does not claim its leader-

ship due to the existence of a node with an even lower

MAC address in its link.

The potential leader picks a random GID, by ﬁrst

investigating the contents of the Preﬁx Cache, and

generates its leader speciﬁc site local address as:

FEC0:0000:HIGH-GID:LOW-GID::1

. To verify the

GID uniqueness, the newly introduced Duplicate

GroupIdentiﬁer Detection (DGID) procedure is ini-

tiated. A NSOL sourced from the leader’s link lo-

cal address is transmitted to the All Nodes In Link

Multicast address, containing the leader speciﬁc site

local address. The hop count of the message is set

to one and an option specifying the leader’s wireless

infrastructure address, which uniquely distinguishes

every node in the network, is included. A timer is set

to signal the end of the DGID procedure.

Neighbouring nodes receiving the NSOL investi-

gate their Preﬁx Caches for the included preﬁx. If

HIGH-GID and LOW-GID represent the high and low

order 16 bits of the GID respectively

one is found, a duplicate GID may have been de-

tected which is resolved by comparing the associated

wireless infrastructure addresses, with the lowest be-

ing favoured. In GID duplications it is also important

to check if they involve established groups. If both

groups are established or are not established, then

comparison of the leaders wireless infrastructure ad-

dresses resolves the duplication. Otherwise, the es-

tablished group is favoured. The result of the resolu-

tion is transmitted by either RADs or NADs, depend-

ing on whether the detecting node is a leader or not.

Only leaders are allowed to immediately transmit a re-

ply. All other nodes are required to queue their packet

for a random time interval of up to one hundred mil-

liseconds and follow a similar procedure to link local

address defence, to transmit or cancel the packet ac-

cordingly.

If a potential leader receives a NAD signifying du-

plication, it picks a new GID and restarts the DGID

procedure. However, if a RAD is received, an es-

tablished leader is in the link of the potential leader

which must stop the group formation procedure and

join the existing group. If no messages are received by

the potential leader and its timer expires, the GID is

assumed unique, no other leaders are found in the link

and the group can be established. Its NAD beaconing

service is replaced by pseudo-periodically transmit-

ted RADs. All such packets are transmitted from the

leader’s link local address and contain options signi-

fying the preﬁx advertised, the leader’s link layer ad-

dress and its wireless infrastructure address.

All nodes receiving the group creation NSOL mes-

sage insert appropriate entries in their Preﬁx and

Router Caches and generate a site local address. This

is achieved by appending the interface identiﬁer from

their link local address to the preﬁx included. The

resulting address is associated with a tentative status,

as the group is not yet established. Permanent status

is only assigned after reception of a valid RAD from

the same leader which advertised the address preﬁx.

Nodes receiving RADs from multiple leaders become

gateways and create site local addresses per group.

One group is selected as the primary group of each

node.

AALM attempts to maintain GID uniqueness

among neighbouring groups. There are two types of

neighbouring groups, direct and indirect. Two groups

are direct neighbours of each other when there exists

a gateway node, in other words a one hop neighbour

of both leaders. In contrast, two groups are indirect

neighbours when they have no gateways among them,

but there are member nodes in each group which

are one hop neighbours of each other. Neighbour-

ing groups of a group are responsible for maintain-

ing GID uniqueness by prohibiting any of their neigh-

bours to use the same GID. This process guarantees

site local address uniqueness across a maximum of

Figure 2: Average lifetime of link local addresses.

six hops. It is augmented by the newly introduced

Group Multicast Information Exchange Mecha-

nism (GMIEM) which uses a Group Multicast Ad-

dress (GMA), as well as a newly deﬁned ICMPv6

Group Information option. Each group leader main-

tains knowledge about neighbouring groups proac-

tively and is able to resolve possible GID duplica-

tions.

The Group Information option is a new ICMPv6

option for the exchange of group information. It con-

tains a GID and the associated wireless infrastructure

address of the leader, together with a ﬂag signify-

ing whether the information belongs to the transmit-

ter’s primary group. These options are inserted by

all group members in their beacons. Each leader in-

cludes the complete contents of its Preﬁx Cache in

all transmitted RADs. This provides group members

with an up-to-date view of all known neighbouring

groups and are qualiﬁed for resolving possible GID

duplications.

Member nodes passively register to the GMA of

each group they are members of. The GMA is a

multicast address of the form: FF02:FF02:HIGH-

GID:LOW-GID::1. All trafﬁc transmitted to the

GMA is received by some of the members of the as-

sociated group. It is provided as a means to ﬁlter out

certain nodes from receiving packets. Only a leader

may be able to reach all group members. The GMA

is used in the GMIEM, to assist member nodes in in-

forming their group neighbours and especially their

respective leaders about newly discovered groups and

propagate this information through the packet queue-

ing mechanism.

Nodes regularly monitor GMA trafﬁc to identify

information identical to their queued packets, result-

ing in the cancellation of their own packets. Dupli-

cations are resolved through wireless infrastructure

address comparison of the conﬂicting group leaders,

Figure 3: Average lifetime of primary site local addresses.

Table 1: Summary of Simulation Scenarios.

No of Groups Total Nodes

7 7 to 70 increments of 7

11 11 to 110 increments of 11

15 15 to 150 increments of 15

with the lowest one being favoured. A reply is imme-

diately transmitted, if the transmitter is a leader, and

forwarded to the affected group as a NAD.

4 EVALUATION

The AALM mechanism has been evaluated through

extensive simulations using the ns-2 simulator. The

characteristics investigated are the amount of sig-

nalling trafﬁc generated, the lifetime of addresses and

the impact of increasing numbers of physical groups

and nodes in the simulation. The simulated scenarios

are produced by the rpgm tool, kindly provided by the

authors of (Camp et al., 2002), which implements the

Reference Point Group Mobility Model. The exact

node and group conﬁguration parameters of the simu-

lations are shown in Table 1. For each node and group

increment, twenty scenarios are generated and simu-

lated with 100 different seeds. The simulator has been

conﬁgured with the standard IEEE 802.11 implemen-

tation, using the Two Ray Ground propagation model.

There exists no other trafﬁc in the network apart from

AALM signalling trafﬁc. The simulated area is 300

metres wide by 300 metres long and the transmission

range is set to 60 metres. The simulated scenario time

is ﬁxed at 300 seconds. The allowed address space

used, is constrained. The interface identiﬁer values

are restricted from 0 to 4 times the number of nodes

in the simulation, which is also applied to GIDs. Fi-

Figure 4: Average total number of packets transmitted per

simulation.

nally, the beaconing interval for all nodes is set to

2 seconds, apart from leaders which advertise their

presence every 1.5 seconds.

The results acquired regarding the lifetime of link

local addresses are highly positive. As shown in Fig-

ure 2 the average lifetime is high, with a lowest value

of about 282 seconds. The graph demonstrates a min-

imal decrease in the link local address lifetime as

the number of nodes per group increases. Increased

numbers of physical groups in the simulation result

in slightly decreased lifetime values. The decreased

trends observed are mainly related with the number of

physical groups, rather than nodes per group. Insert-

ing more groups in the network increases the number

of nodes in the simulation which are beyond a node’s

extended link. Therefore, the existence of a duplicate

address somewhere in the simulation area, which is

not resolved, is more probable.

Figure 3 shows the results obtained regarding the

average lifetime of primary site local addresses, in

other words the address associated with a node’s pri-

mary group. It is evident that increased numbers of

nodes per group results in decreased address lifetime

values. The same applies to increased numbers of

physical groups. Another important observation is

that as the number of physical groups increases, the

differences in the results obtained for increased num-

bers of nodes per group are decreased. For example,

for 15 physical groups, the average primary site lo-

cal address lifetime for 1 node per group is about 30

seconds, whereas it is 20 seconds for 10 nodes per

group, a difference of 10 seconds. On the other hand,

for 7 groups the corresponding values are 60 and 36,

which is a difference of 24 seconds. The obtained

values are suitable for HTTP-type trafﬁc, where the

amount of data exchanged is relatively low. They

are unsuitable for multimedia-type trafﬁc and large

Figure 5: Average number of packets transmitted per node.

data exchanges, due to the requirement for a mech-

anism capable of maintaining connections after ad-

dress changes. Although such mechanisms exist in

wired networks (MIPv6), which have been somewhat

adapted to operate in an ad hoc context, there exists

no mechanism at present to operate in a pure ad hoc

networking environment. The lifetime of primary site

local addresses is dependent on group and link local

address lifetime. Whenever a link local address is

removed, for any reason, all site local addresses are

also removed. Furthermore, upon group destruction,

nodes must pick a new primary site local address, if

another group is present, or initiate the group forma-

tion process. The achieved results point to reduced

group lifetimes. This is one aspect of the proposed

mechanism which requires further investigation and

improvement.

The graphs shown in Figure 4 show the signalling

trafﬁc footprint of the mechanism. Clearly as the

number of nodes and nodes per group increases, so

does the number of transmitted packets. This increase

is almost linear in nature. It is observed that increas-

ing the total number of nodes in the simulation by 40

results in about 7000 extra packets. Figure 5 shows

the average number of packets transmitted per node.

In this ﬁgure, it is interesting to note that for the most

part, increasing the number of nodes per group, re-

sults in an increased average number of packets trans-

mitted per node. The only exception is for 1 node

per group. The initial high values are attributed to

the greater number of leaders in respect to ordinary

nodes per group, which transmit packets at a lower

interval (1.5 seconds for leaders versus 2 seconds for

ordinary nodes). Furthermore, increasing the number

of groups results in slightly increased values. The av-

erage number of packets transmitted per node and the

one hop nature of these packets indicates the scalabil-

ity of the mechanism.

5 CONCLUSION

This paper has presented a novel address autoconﬁg-

uration mechanism for NANETs, which are a sub-

set of MANETs and are associated with civilian-style

scenarios. The mechanism has been fundamentally

designed to address the problem of partitioning, util-

ising underlying environmental characteristics to im-

pose a Network layer subnet structure. Furthermore,

the structure is maintained through signalling trafﬁc

localisation, thereby reducing the effects of partition-

ing and probable address duplications.

The evaluation of the proposed AALM mechanism,

through extensive simulations, has demonstrated its

effectiveness to facilitate short lived HTTP-type traf-

ﬁc, in regards to signalling trafﬁc overhead and link

local addresses lifetime. It is an efﬁcient approach

for assigning addresses usable for one hop data ex-

change, although requires further improvement re-

garding long-lived connectivity in respect to the life-

time of site local addresses. Future work should also

concentrate on improving the design of AALM in or-

der to provide a more generic solution applicable to a

wider range of MANETs.

REFERENCES

Camp, T., Boleng, J., and Davies, V. (2002). A survey of

mobility models for ad hoc network research. Wireless

Communications and Mobile Computing, 2(5):483–

502.

Chiang, C. C. (1997). Routing in Clustered Multihop,

Mobile Wireless Networks with Fading Channel. In

IEEE Singapore International Conference on Net-

works (IEEE SICON ’97), pages 197–211.

Droms, R. (1997). RFC 2131: Dynamic Host Conﬁguration

Protocol.

Ephremides, A. (1988). A design concept for reliable mo-

bile radio networks with frequency-hopping signaling.

NASA STI/Recon Technical Report N, 89:17772–+.

Guerin, R., Kim, E., and Sakar, S. (2002). Bluetooth Tech-

nology: Key Challenges and Initial Research. In Con-

ference on Network and Distributed Simulations.

Law, C. and Siu, K. Y. (2001). A Bluetooth scatternet for-

mation algorithm. In IEEE Symposium on Ad Hoc

Wireless Networks 2001.

Lundgren, H., Nordstr

om, E., and Tschudin, C. F. (2002).

Coping with communication gray zones in IEEE

802.11b based ad hoc networks. Technical Report

2002-022, Department of Information Technology,

Uppsala University.

Mathur, R., Klepal, M., McGibney, A., and Pesch, D.

(2004). Inﬂuence of People Shadowing on Bit Error

Rate of IEEE802.11 2.4GHZ Channel. In 1st IEEE

International Symposium on Wireless Communication

Systems, ISWCS04.

Mikl

os, G., R

acz, A., Tur

anyi, Z., Valk

o, A., and Johansson,

P. (2000). Performance aspects of Bluetooth scatternet

formation. In MobiHoc ’00: 1st ACM international

symposium on Mobile Ad hoc Networking & Com-

puting, pages 147–148, Piscataway, NJ, USA. IEEE

Press.

Nesargi, S. and Prakash, R. (2002). MANETconf: Con-

ﬁguration of Hosts in a Mobile Ad Hoc Network. In

Proceedings of IEEE INFOCOM.

Perkins, C., Royer, E., and Das, S. (2000). IP Address Au-

toconﬁguration for Ad Hoc Networks. Internet Draft,

IETF.

Prakash, R. (1999). Unidirectional links prove costly in

wireless ad hoc networks. In DIALM ’99: 3rd Interna-

tional Workshop on Discrete algorithms and methods

for mobile computing and communications, pages 15–

22, New York, NY, USA. ACM Press.

Prince, D. (2005). Dynamic Service Deployment through

Consensus Negotiation in Programmable Ad hoc Net-

works. PhD thesis, Lancaster University.

Sun, Y. and Belding-Royer, E. M. (2003). Dynamic address

conﬁguration in mobile ad hoc networks. UCSB Tech-

nical Report 2003-11.

Thomson, S. and Narten, T. (1998). IPv6 Stateless Address

Autoconﬁguration. RFC 2462.

Toner, S. and O’Mahony, D. (2003). Self-organising node

address management in ad-hoc networks. Lecture

notes in Computer Science 2775, pages 476–483.

Tschudin, C. F., Gold, R., Rensfelt, O., and Wibling, O.

(2004). LUNAR: a Lightweight Underlay Network

Ad-hoc Routing Protocol and Implementation. In Next

Generation Teletrafﬁc and Wired/Wireless Advanced

Networking (NEW2AN’04), St. Petersburg.

Tschudin, C. F., Lundgren, H., and Nordstr

om, E. (2003).

Embedding MANETs in the Real World. In Personal

Wireless Communications (PWC), pages 578–589.

Vaidya, N. H. (2002). Weak duplicate address detection in

mobile ad hoc networks. In MobiHoc ’02: 3rd ACM

international symposium on Mobile ad hoc network-

ing & computing, pages 206–216, New York, NY,

USA. ACM Press.

Vergetis, E., Guerin, R., Sarkar, S., and Rank, J.(2005). Can

Bluetooth succeed as a large-scale ad hoc networking

technology? IEEE Journal on Selected Areas in Com-

munications (JSAC) Special Issue, 23:644–656.

Weniger, K. (2003). Passive Duplicate Address Detection

in Mobile Ad Hoc Networks. In IEEE Wireless Com-

munications and Networking Conference (WCNC).

Weniger, K. (2005). PACMAN: Passive Autoconﬁguration

for Mobile Ad hoc Networks. IEEE Journal on Se-

lected Areas in Communications (JSAC) Special Issue.

Zaruba, G., Basagni, S., and Chlamtac, I. (2001). Blue-

trees - scatternet formation to enable bluetooth based

ad hoc networks. In IEEE International Conference

on Communications(ICC).

Zhou, H., Ni, L. M., and Mutka, M. W. (2003). Prophet

address allocation for large scale MANETs. Ad Hoc

Networks, 1(4):423–434.