WHAT CAN CONTEXT DO FOR TRUST IN MANETS?

Eymen Ben Bnina, Olivier Camp, Chùng Ti

ên Nguy

ên

Department of Computer Science, ESEO, Angers, France

Hella Kaffel Ben Ayed

Faculty of Science, Tunis, Tunisia

Keywords:

Trust models, context, ad hoc networks.

Abstract:

The global performance of a mobile ad hoc networks (manet) greatly depends on, both, the cooperation of

participating nodes and the environment in which the nodes evolve. The willingness of a node to cooperate can

be illustrated by the trust assessed to the node. Yet, existing trust models, designed for reliable wired networks,

do not take into consideration possible communication failures between client and server. We believe, that in

the case of ad hoc networks such factors should be considered when computing trust. In this article, we show

how an interaction can be decomoposed in three separate phases : two communication phases for transporting

the request to the server and the response back to the client, and one execution phase which represents the

actual execution of the service by the server. We propose to deﬁne the communication environment using

contextual attributes and to consider this context when assessing trust to a server. We discuss the possible uses

of context in the ﬁeld of trust computation in manets and deﬁne contextual attributes that seem important to

consider when modelling and computing trust.

1 INTRODUCTION

A distributed system is a system consisting of nodes

which cooperate together provide users with services

such as web services, data exchange and sharing of

software. Such an environment is said to be fully dis-

tributed if there is no central component controlling

the whole system. Besides centralised distributed sys-

tems, in which one or more central servers control the

entire system, we can ﬁnd fully distributed systems in

which all nodes are equals and are, together, in charge

of controlling the system. Mobile ad hoc networks

(manets), also called spontaneous networks, are an

example of such fully distributed systems.

Manets are IP networks made up of a collection

of mobile nodes communicating via radio links. They

do not rely on any predeﬁned infrastructure or cen-

tralised administration to operate. Nodes in a manet

may, at any time, leave, enter or roam within the net-

work. The resulting dynamic nature of the network’s

topology, along with the unreliability of the wireless

links, thus require for manet services to be highly

adaptable. For instance, in the case of routing, the

lack of a network infrastructure implies that the ser-

vice is provided in a peer-to-peer fashion and that all

the nodes need to act as collaborating routers, to pro-

vide multi-hop routes between any source and des-

tination. Moreover, the availability of an individual

central node cannot be guaranteed at all times. There-

fore, services cannot rely on a point of centralisation

and should be provided in a distributed and adaptive

manner.

Manets provide an easier way, in comparison with

classical, infrastucture-based networks, to aggregate

large amounts of resources while maintaining a low

system maintenance cost and are thus an interesting

solution when setting up dynamic and ﬂexible appli-

cations. However, these systems also bring up some

new problems. Indeed, in such an environment, when

a node needs to use a given service, it may not know

with which quality of service each server is able to

provide the service. This is particularly true in an

open network in which a node does not knowto which

extent the other nodes are willing to cooperate.

One among the solutions that have been suggested

for such problems, is to control whether a node is, or

not, authorised to enter the network. This way, it can

be considered that the nodes in the network have all

the approval of a trusted central entity and may also

themselves be trusted; according to a policy deﬁned

133

Ben Bnina E., Camp O., Tiê

n Nguyên C. and Kaffel Ben Ayed H. (2008).

WHAT CAN CONTEXT DO FOR TRUST IN MANETS?.

In Proceedings of the Tenth International Conference on Enterprise Information Systems - SAIC, pages 133-141

DOI: 10.5220/0001701001330141

 SciTePress

by the trusted central entity. Yet, such limitation on

the number of cooperating nodes may greatly degrade

the performances of the network. Another solution is

to calculate the trust that the client assigns to each

server and, based on this, decide whether or not to

cooperate with the service provider. Several works

have tackled the problem of computational trust and

different models have been proposed.

The existing trust models are well adapted to sta-

ble and reliable networks, but most are not suitable

for the inherently dynamic nature of manets. Indeed,

it is no point determining the most trustworthy server

for a given service, if the unreliability of the network

does not allow communication with this server. For

this reason, we believe that the factors inﬂuencing the

stability of the network should also be considered dur-

ing the evaluation of trust.

In this paper,we present a brief review of some ex-

isting trust models and discuss how researchers have

proposed different approaches for computing trust.

We stress the fact that most models do not explicitely

take situational information into account when com-

puting trust and explain why this is penalising in un-

reliable and dynamic contexts such as manets. Thus,

we discuss the notion of context and present how it

can be used in manets during trust computation.

The remain of this paper is organised as follows:

section 2 gives a brief survey of some existing trust

models, and points out approaches proposed by re-

searchers for computing trust. Section 3 presents the

reasons for which, in our opinion, context should be

considered when computing trust in manets and gives

an overview of existing deﬁnitions for context. Sec-

tion 4 gives our perception of context in manets and

proposes a set of contextual attributes that, we think,

should be considered when studying the result of past

interactions when computing trust. We conclude in

section 5 and present our future works.

2 BACKGROUND

Trust is a concept which is frequently used in our

daily lives. In fact, it often guides the social inter-

actions with other individuals. Trust has been stud-

ied in various ﬁelds (philosophy, economics, sociol-

ogy, psychology and, more recently, computer sci-

ence) thus leading to the existence of several deﬁ-

nitions (Duma et al., 2005)(Grandison, 2003) and a

lack of coherence among researchers. (McKnight and

Chervany., 1996). For us, the most appropriate def-

inition is that proposed by Gambetta in (Gambetta,

1988): "Trust (or symmetrically, distrust) is a partic-

ular level of the subjective probability with which an

agent assesses that another agent or group of agents

will perform a particular action, both before he can

monitor such action (or independently of his capacity

ever to be able to monitor it) and in a context in which

it affects its own action". This deﬁnition stresses the

facts that: trust is uncertain, trust is subjective and

trust depends on context.

Several trust models have been described in lit-

erature. The differences between them mainly con-

cern : trust modeling, trust management and decision

making. While trust modeling deals with the repre-

sentational and computational aspects of trust values,

trust management focuses on the collection of evi-

dence and risk evaluation.Concerning decision mak-

ing it is actually a part of trust management. In this

survey, we will concentrate only on trust modeling.

The model deﬁned in (Marsh, 1994) is based only

on direct experiences between trustor and trustee. In

this model, the author introduces three types of trust :

• Dispositional Trust T

refers to the trust of an

agent x regardless of the possible cooperation

partner and the situation.

• General Trust T

(y) describes the trust of x on y

regardless of the situation.

• Situational Trust T

(y, β) describes the trust of

agent x on agent y in situation β. The values of

trust belong to [−1, 1[. These trust values illus-

trate the fact that full distrust exists but not total

trust.

Abdul-Rahman and Hailes (Abdul-Rahman and

Hailes, 2000) present a trust model based on the so-

ciological characteristics of trust. Their model sup-

ports the following properties of social trust: a) Trust

is context-dependent. b) Trust supports negative and

positive degrees of belief of an agent’s trustworthi-

ness. c) Trust is based on prior experiences. Agents

are able to identify repeated experiences with sim-

ilar contexts and with the same agents. d) Rep-

utational information is exchanged between agents

through recommendations. e) Trust is not transitive -

all evaluations of recommendations take into account

the source of the recommendation. f) Trust is sub-

jective -different agents may have different percep-

tions of the same agent’s trustworthiness. g) Trust

is dynamic and non-monotonic -further experiences

and recommendations increase or decrease the level

of trust in another agent. h) Only Interpersonal Trust

(the trust one agent has in another agent directly in a

speciﬁc context) is supported.

This model uses direct trust and recommendor

trust for computing the global trust value assigned by

an agent to the trustworthiness of another. The di-

rect trust that a given trustor agent assigns to another

ICEIS 2008 - International Conference on Enterprise Information Systems

134

trustee agent A, relatively to a given contextC, is rep-

resented by t(A, c,td) where td is the degree of trust

assigned to agent A and may take one of the following

four values: “very trustworthy”, “trustworthy”, “un-

trustworthy”, “very untrustworthy”. In this model,

recommendor trust represents the belief of agent A

concerning the fact that agent B is trustworthy, to a

certain degree, for giving recommendations concern-

ing other agents relatively to a given context c. Rec-

ommendor trust is represented by tr(B, c, rtd) where

rtd is a semantic distance between A’s perception and

the recommendations it received.

In order to compute the trust that an agent A as-

signs to another agent B, the authors in (Castelfranchi

and Falcone, 1998) propose a trust model based on a

cognitive approach. They assert that the reasons that

make A ask for a service from B result from a set of

mental beliefs. For that, their model is based on the

following three beliefs: 1) Competence belief: A be-

lieves that B can actually do the task; 2)Dependence

belief: A believes that B is the necessary or the best

agent to rely on for doing this task; 3)Disposition be-

lief: A believes that B actually will do the task. The

latter belief is articulated by two other beliefs which

are: i)Willingness belief: A believes that B has de-

cided and will perform action α that is related to the

task; ii)Persistence belief: A believes that B is stable

in its intentions of doing α.

Thus, trust in this model is represented as a set of

mental attitudes which allow agent A to believe that

another agent B will respond to its requests.

The authors in (Yu and Singh, 2001) present a

model which doesn’t combine information relative to

direct and indirect interactions. Recommendations

are only used if an agent has never interacted directly

with the agent for which trust is computed. An ex-

ample of such a situation would be if the trustee has

only recently joined the network. Each node stores

the information concerning direct interactions as a set

of values that reﬂect the quality of these interactions.

This model only considers the most recent experi-

ences and deﬁnes an upper and lower thresholds that

deﬁne the limit between what are considered QoS as-

cribed to trustworthy agents, QoS with no clear clas-

siﬁcation and QoS ascribed to non trustworthy agents.

By applying the Dempster-Shafer theory of evidence

on the stored information an agent is able to compute

the probability with which the trustee belongs to one

of the above three groups.

In (Nguyen and Camp, 2007), the authors propose

a model that uses, both, direct experiences and rec-

ommendations. They are given as input to a Bayesian

network, and the computing of trust values is based

on a probabilistic approach using Bayes’ theorem of

conditional probability. This model considers that the

future behavior of an agent depends on its past be-

haviors and bayesian networks are very efﬁcient for

manipulating the associated conditional probabilities.

In this model, the trust assigned by agent A to a

server S for providing service S∗ with a quality q is

deﬁned as the probability that S will provide S∗ to A

with the given quality q and is calculated consider-

ing the results of the past interactions with service S∗

provided by S.

In this section, we have reviewed some of the ex-

isting trust models and pointed out approaches that re-

searchers rely on for computing trust. However these

models do not use situational information which we

believe very important in dynamic environments. In

the following section, we will discuss why contextual

information, speciﬁcally in the case of manets, is im-

portant when computing trust. We will also show how

such information may change the trust an agent has in

a server for completing a given task.

3 WHY IS CONTEXT

IMPORTANT FOR TRUST IN

MANETS ?

For their computing needs, agents in a manet often

have to rely on services provided by others. The

fully distributed nature of a manet does not allow for

a trusted central entity to manage these interactions

and this thus has to be taken in charge of by each

agent. The server providing the required service may

be out of the client’s emission range. In such a case

the client must rely on intermediate nodes to trans-

mit its request, and the server should also use a hop

by hop approach to reply to the client. For the client

to consider its interaction successful, the following

three steps must be accomplished successfully : The

request reaches the server, the server executes the re-

quest, the server’s response reaches the client. This is

represented in ﬁgure 1.

On the contrary, if the interaction fails, client C

cannot determine which one of the steps, detailed

Server (S)

Client (C)

3) Response

1) Request

2) Execution

Figure 1: Interaction between agents in a manet.

WHAT CAN CONTEXT DO FOR TRUST IN MANETS?

135

above, has failed. Moreover,C does not have a global

vision of the network and thus does not know the

route taken by its request, nor does it know the in-

termediate nodes that helped in the transmission of

the response. In fact, the only information the client

holds concerning the routes to and from the server are

those found in its own routing table - ie ; the ﬁrst hop

and the number of hops to destination. This, of course

is only true if we do not search for such information

in the implementation dependent data manipulated by

the routing protocol. We choose to only consider the

routing table to stay independent from any speciﬁc

routing protocol.

According to Gambetta’s deﬁnition of trust, the

trust C has that it will obtain a reply to its request

to S depends on: the trust it has that the request will

reach S, the trust it has that S will correctly execute

the service and the trust it has that the reply will be

transmitted back.

Most of the existing trust models only concentrate

on the value of trust assigned to S in properly exe-

cuting the service and choose a collaborator among

the servers with the highest such value of trust. We

believe that such an approach is insufﬁcient, in the

case of manets, and that the correct transmission of

request and response should also be considered. How-

ever, due to the lack of information available con-

cerning the exact participants and their behaviours in

both these transmissions, we consider that an expe-

rience/recommendation based trust model is not ad-

equate for the routing service. Instead, we prefer

to consider that the proper transport of request and

response both depend on the environment through

which those messages are exchanged and propose to

consider context relevant information when comput-

ing trust and deciding of a service provider.

Many deﬁnitions have been given for the concept

of context. While in (Salber et al., 1999) authors qual-

ify context as an environment or situation, many re-

searchers use the deﬁnition given in (Abowd et al.,

1999): ”Any information that can be used to charac-

terise the situation of entities (i.e whether a person,

place or object) that are considered relevant to the in-

teraction between a user and an application, including

the user and the application themselves”. In (Schilit

et al., 1994), the authors propose to organise the con-

cept of context in three categories:

• User Context: user proﬁle, location, people

nearby, social situation, activity, health condi-

tions, agenda settings, etc.;

• Execution Context: network trafﬁc, status of the

device, availability of resources, communication

costs, nearby resources, etc.;

• Environment Context: weather, light, noise

level, temperature, time, etc.

This categorisation ﬁts well with our vision of an in-

teraction in a manet. Indeed, the routing context inﬂu-

encing the data exchanges between client and server

clearly belongs to the environment context. Also, the

execution context reﬂects situational factors that may

inﬂuence the actual execution of the service by the

server. As for user context, we will see that it can be

used to describe the user proﬁles of both partners of

the interaction.

The trust models presented in (Castelfranchi and

Falcone, 1998), (Marsh, 1994), (Yu and Singh, 2001)

and (Nguyen and Camp, 2007) are not context aware

and compute trust values independently from context.

Concerning the model presented in (Abdul-Rahman

and Hailes, 2000), authors consider trust in the pres-

ence of virtual communities and leave context man-

agement open to let developers deﬁne their own con-

text. In our opinion, contextual information should be

recorded together with every experience and consid-

ered during the trust computation phase. In this work,

we point out the contextual informations that, in our

opinion, should be considered and discuss on how the

context can be used when computing trust in ad hoc

networks.

This section has presented the main reason for

which we consider a trust model for manets should be

context-aware and has brieﬂy presented works deﬁn-

ing the concept of context. We must now isolate

the contextual attributes that may be interesting when

considering the mobile, unstable and open environ-

ment of manets.

4 WHICH CONTEXT IS

IMPORTANT FOR TRUST IN

MANETS ?

Even though, to the best of our knowledge, few trust

models take into account the routing context (Toivo-

nen et al., 2006), some consider basic contextual in-

formation when collecting experiences (direct or con-

tained in recommendations). For instance, the model

proposed in (Nguyen and Camp, 2007) records the

time at which each experience occurs and gives less

consideration to older interactions than it does to

more recent ones. We believe this information is

particularly relevant in the case of a manet in which

agents can have varying efﬁciencies according to their

battery power or even suddenly disappear from the

network.

Another contextual attribute, usually implicitely,

ICEIS 2008 - International Conference on Enterprise Information Systems

136

taken in charge by some models is the number of

interactions that have occurred between client and

server when trust in the server is calculated. Indeed,

the trust values computed by probability based mod-

els gain precision as the number of interactions with

the server increases. Now, the number of interactions

could be considered as an information relative to the

execution of a service by a given server and thus be

part of the execution context; This is a ﬁrst step to-

wards context awareness.

Even though the above two contextual attributes

seem important, they have no inﬂuence on the trans-

mission of messages between client and server. We

should now decide which contextual attributes are rel-

evant to these steps of the interaction.

The routing context associated to an experience

should contain representationsfor the factors that may

inﬂuence the quality of exchanges between client and

server.

4.1 Hop Count

Works presented in (Hekmat and Mieghem, 2003)

show that the probability of success of a communi-

cation between agents is highly dependent on the dis-

tance, in terms of hop count, between the participants;

the higher the hop count, the lower the probability of

success. For this reason we consider the hop count be-

tween client and server to be a crucial information for

caracterising the context of an interaction. This infor-

mation is freely available in the routing table, for any

node with which a communication is possible. How-

ever, the client can only retrieve, the distance to the

server, from its routing table; It cannot determine the

length of the return trip taken by the server’s response.

Indeed, existing routing algorithms, do not necessar-

ily use the same route from one node to another, as

they do than they do when returning. Nevertheless,

we can assume that the return trip is similar, in terms

of number of hops and therefore that the length of the

route from the client to the server also characterises

the travelling of the response.

From our point of view, hop count between client

and server is the ﬁrst contextual information to be

considered in manets.

4.2 Mobility

The performance in terms of communication, of a

manet is closely related to the capacity of its rout-

ing algorithm to adapt to the mobility of the nodes.

Nevertheless, however efﬁcient routing is in dealing

with mobility, communication between very mobile

nodes, or through a very dynamic network will be less

reliable than between stationnary nodes, or in a sta-

ble environment. Finding out a metric for mobility in

manets is a challenge. There exists few researchs that

have focused on such a topic. In (Boleng et al., 2002),

the authors have used link stability as a metric for mo-

bility to show that mobility has a direct effect on end

to end delay and data packet deliveryratio. Authors in

(Ghassemian et al., 2005) have used a similar metric,

deduced from the frequency of link state change and

link connectivity duration. It thus seems natural to in-

clude information concerning mobility in the routing

context. Moreover, intuitively, it seems to us gener-

ally more efﬁcient for an agent to choose the most

stable server in order to have the greatest probability

that its requets reach the destination server.

What is mobility, how can it be deﬁned and mea-

sured? Is it useful to consider the mobility of the en-

tire network or should we only study that of the path

between client and server? All these questions should

be answered before considering mobility as a part of

the routing context.

Mobility in ad hoc networks is a topic of research

that has been considered through various angles : the

speciﬁcation of mobility aware routing protocols, the

effect of mobility on the performances of routing pro-

tocols, the deﬁnition of mobility models for simula-

tion, the deﬁnition of metrics to measure the mobility

of a given simulated network. Here, our aim is, for

any node of an operating network, to be able to de-

termine both the mobility of any other node and the

general mobility of the network.

This, of course, should be done using information

held by each node concerning the others ; namely the

routing table.

We can generally deﬁne mobility as the behav-

ior of an object (entity, person, thing, etc) that has a

changing position over time. However, in our case, it

is important to point out that, what we refer to as mo-

bility is in fact relative mobility. If two nodes move

in the same direction and with equal speed, they can

both be considered stable with respect to one another.

However, if the other agents in the network are static

then the two nodes have very high mobility relatively

to the rest of the network. Thus, two measures of

mobility may be considered: individual mobility of

nodes and global mobility of the network. In our case,

global mobility of the network should be considered

because it gives an idea on the mobility of the sur-

rounding environment. This general measure, how-

ever, does not reﬂect the individual mobility of each

node because it is an average measure and thus sta-

ble nodes can not be discovered using it. Thus, in

our opinion, it is also useful to consider the individual

mobility of nodes. In such a case, it will be easier for

WHAT CAN CONTEXT DO FOR TRUST IN MANETS?

137

a client agent, given a set of servers with different in-

dividual mobility measure, to choose the one it judges

most appropriate to respond to its needs. We propose

to use the same metrics to measure both general net-

work and individual node mobilities. In fact, we will

consider that the global mobility of the environment

is the average of the mobilities of all reachable nodes.

Before, examining the information that may re-

ﬂect the mobility of a node or of the network as a

whole, it should be noted that, since we do not con-

sider that the nodes are equiped with a globally ac-

cessible location service (GPS for instance), a node is

only able to partially capture its relative mobility with

respect to other nodes and the mobility of the path be-

tween it and all reachable destination nodes. More-

over, the disappearence (respectively reappearence)

of a node from the routing table can be the conse-

quence of its relative mobility. Even though, such a

situation may also be the result of the switching off

or on of the device, we choose to consider that this

information is a hint concerning this particular node’s

mobility .

To measure mobility of nodes, we choose to con-

sider the following information gathered from the

routing table: the appearances and disappearances of

nodes in the routing table, the number of hops to all

reachable nodes and the ﬁrst hop to each destination.

If we consider the situations depicted in ﬁgure 2, in

which S

represents the position of server S at time

(i ∈ [0, 3] and t

< t

i+1

), we notice that the move-

ment of S between t

and t

can be read in C’s routing

table as a change in the distance to S; the movement

between t

and t

is read by C as a change in the ﬁrst

hop of the route to S; and movement between t

and

is seen as the disappearance of S from C’s routing

table.

Yet, this capture is only partial as the available in-

formation in the routing table (the existence of the

destination node, the hop count and the ﬁrst hop to

destination) is not sufﬁcient to detect all mobile nodes

Figure 2: Detected mobility.

Figure 3: Undetectable mobility.

or routes. In fact, any change in a route, past the ﬁrst

hop, will be unoticeable if it does not modify the hop

count. Figure 3 in which S

, S

and S

represent ,re-

spectively, the position of server S at times t

, t

and

< t

) represents such a situation.

From C’s point of view S is always 4 hops away

and always has the same ﬁrst hop to destination.It

is thus not considered mobile. For these reasons, it

would be useful, when studying mobility, to know

all intermediate nodes between C and S. This infor-

mation is available from the routing table if C and S

are neighbours (in this case there are no intermediate

nodes between C and S).Otherwise, the exact route

can only be discovered by examining the routing ta-

bles on the intermediate nodes.

To estimate the mobility of the nodes in the net-

work, we propose to consider the routing table at reg-

ular time intervals and, for each reachable destination

node, to measure the following :

• Path Stability: The percentage of time during

which each entity has been in the routing table

since the begining of the time interval,

• Link Stability: The percentage of time during

which each entry has been as a neighbour in the

routing table since the begining of the time inter-

val,

• Distance Stability: The average number change in

hop count to each entry since the begining of the

time interval,

• First Hop Stability: The average number of

changes in the ﬁrst hop since the begining of the

time interval.

4.3 Density

Density in ad hoc networks is another information

that may be considered when choosing an appropri-

ate server with which to cooperate. On the one hand,

ICEIS 2008 - International Conference on Enterprise Information Systems

138

if the network is dense, there will be more possibili-

ties to ﬁnd the best route to each destination. On the

other hand, in a scarce network, less potential routes

will be available for each destination. The movement

of the nodes may have, as an effect, to break links in

a route. In such a case, the redundancy of routes in

a dense network will often allow to ﬁnd a new route

to reach the destination node. On the contrary, such a

movement might result, in interruption of all commu-

nications with the destination if node density is not as

high.

Mobility and density are thus closely related to

one another, and it is clear that some conﬁgurations

are more favorable than others. For instance, it seems

clear that a dense and static network will be more efﬁ-

cient in ﬁnding stable routes to destination nodes than

a scarce and very mobile network. However, the case

of stable and scarce networks and of mobile and dense

networks need to be studied with attention.

As it is the case for mobility, different granulari-

ties can be given to the measure of density: the global

density of the network, composed of all reachable

nodes, can be considered or density can be calculated

on a per hop basis.

We are now studying the effects of both mobility

and density on the efﬁciency of communications in a

manet in order to determine precisely how these con-

textual attributes should be measured and how they

can be considered in trust computation. This will be

the subject of a future paper.

4.4 Server and Client Proﬁles

Number of hops, mobility and density are contextual

attributes that are part of the routing context; they

inﬂuence the quality of the communication between

client and server. In the classiﬁcation described in

section 4, we have chosen to consider the routing con-

text as a part of the environment context. We will now

discuss how the proﬁles of both client and server may

also be considered as contextual information and in-

cluded in what is deﬁned by (Schilit et al., 1994) as

the user context.

In a distributed system, nodes may be organised

in virtual communities (Abdul-Rahman and Hailes,

2000) and prefer to cooperate with nodes in the same

community or in partner communities. Our trust

model deﬁnes two types of communities : static com-

munities that reﬂect an underlying administrative or-

ganisation (we could for instance consider a ”stu-

dent” community, an ”administrative” community, a

”teacher” community and a ”research” community

in a university’s network) and dynamic communities

which each node deﬁnes based on the results of the

interactions it has had with the others and thus on the

trust it has in them in providing a given service with

a certain level of quality. Concerning dynamic com-

munities, they can be used to regroup well behaving

servers together or, contrarily, misbehaving servers ;

for example, a node could dynamically deﬁne, both,

the community of servers in which it has ”very high”

trust for providing service S∗ and the community of

servers in which it has ”very low” trust in providing

service S∗. The trust to put on an interaction with

a member of the community is deﬁned by the dy-

namic community itself. Any node, whether server or

client, may thus belong to several communities, and

we consider that such information should be consid-

ered as proﬁle information that may be useful in pre-

dicting the result of an interaction. In the case of static

communities a trust policy can be deﬁned to describe

inter-community trust relation. The static communi-

ties to which belong the server and the client may also

be considered as contextual information since they

may affect the quality with which the service is pro-

vided

4.5 First Hop

The contextual information we have considered up to

now in the routing context affect the quality of an in-

teraction, however cooperative the nodes of the net-

work are. In fact, mobility, density and the number of

hops to a destination are factors that affect the quality

of communication regardless of the willingness of the

nodes to properly route trafﬁc. Thus, the contextual

attributes presented above do not take into account the

presence of selﬁsh or malicious nodes in the network.

Such nodes may, by refusing to relay data, have a dis-

astrous effect on communication.

Rather than considering routing as any other ser-

vice and using the model to compute the trust in

routers, we propose to also use situational informa-

tion to take into account such malicious nodes.

Whenever two agents wish to communicate with

one another, unless they are neighbours, they must

rely on intermediate nodes and especially on a ﬁrst

hop. This information may seem very partial in the

case of long routes. Yet, we believe that it is an im-

portant contextual attribute to be added to the routing

context.

In this section we have presented contextual at-

tributes that seem interesting to consider when com-

puting trust. These contextual arguments can be or-

ganised according to the following taxonomy repre-

sented in ﬁgure 4. We have also showed in this sec-

tion that certain context attributes (hop count, mobil-

ity and density) helped dealing with the general be-

WHAT CAN CONTEXT DO FOR TRUST IN MANETS?

139

Figure 4: Taxonomy of context.

havior of manets when only in the presence of coop-

erative nodes, other attributes (proﬁle informations)

took advantage of a possible underlying organisation

and others stills (hop count) may help in avoiding non

cooperative or malicious agents.

Contextual information should be considered

both, when recording experiences and during the trust

computing process. An experience should thus carry

the contextual information that reﬂects the context in

which it has occurred. Moreover, to evaluate the trust

we have in a server S that, in a speciﬁc context C,

we will receive its response to a request, we should

transpose the set of all previous experiences with S

into context C. Through this transposition, we evalu-

ate, for each experience, the quality it would have had

if it had occurred in the current context. The set of

transposed experiences can now be used as an input

to the trust computing process, to calculate the trust

assigned to S in context C.

5 CONCLUSIONS AND FUTURE

WORKS

Trust is a concept that must be studied with attention

in distributed environments based on the exchange

of services between partners. In this paper we have

brieﬂy presented some of the models proposed by

research for computing trust. These models do not

explicitely take contextuals information into account

when assessing trust. We have discussed that in the

particular case of manets, the context, and particu-

larly the routing context, should be considered in the

trust computation process. We have identiﬁed a set

of contextual attributes that may affect the result of

an interaction with a service provider. We propose to

use these contextual attibutes together with the other

parameters that characterise each experience when as-

signing trust to a cooperation partner. These attributes

may greatly inﬂuence the way in which a client agent

perceives the quality of a given server. It may thus

be more efﬁcient to request a service form a provider

with low trust in a favourable context than to one with

higher trust in a disadvantegeous context. Yet, before

considering the proposed context variables, an ade-

quate metric, for measuring their values should be de-

ﬁned. Even though the metrics used to measure some

of the proposed variables seem quite straightforward

(this is the case for the number of hops between both

partners), other factors are much more tricky to mea-

sure and the deﬁnition of a metric for these variables

is not clear (this is the case for mobility). Moreover,

while some variables may be considered individually,

others are so closely related to each others that they

should be considered together (this is the case for den-

sity and mobility). We are now studying the inﬂu-

ence of the identiﬁed context variables on the qual-

ity of communications in order to deﬁne metrics for

these attributes. A second step will be to study how

these contextual attributes should be considered by a

context-aware trust model for manets.

REFERENCES

Abdul-Rahman, A. and Hailes, S. (2000). Supporting trust

in virtual communities. In HICSS ’00: Proceedings

of the 33rd Hawaii International Conference on Sys-

tem Sciences-Volume 6, page 6007, Washington, DC,

USA. IEEE Computer Society.

Abowd, G. D., Dey, A. K., Brown, P. J., Davies, N., Smith,

M., and Steggles, P. (1999). Towards a better under-

standing of context and context-awareness. In HUC

’99: Proceedings of the 1st international symposium

on Handheld and Ubiquitous Computing, pages 304–

307, London, UK. Springer-Verlag.

Boleng, J., Navidi, W., and Camp, T. (June 2002.). Metrics

to enable adaptive protocols for mobile ad hoc net-

works. In in Proceedings of the International Con-

ference on Wireless Networks (ICWN ’02), Las Vegas,

Nev, USA,, pages 293–298.

Castelfranchi, C. and Falcone, R. (1998). Principles of

trust for mas: Cognitive anatomy, social importance,

and quantiﬁcation. In Proceeding of the International

Conference on Multi-Agent Systems(ICMAS’98).

Duma, C., Shahmehri, N., and Caronni., G. (August 2005.).

Dynamic trust metrics for peer-to-peer systems. In In

Proc. of 2nd IEEE Workshop on P2P Data Manage-

ment, Security and Trust.

Gambetta, D. (1988). Can We Trust Trust? Basil Blackwell.

Ghassemian, M., Friderikos, V., and Aghvami, A. H.

(September 2005). On mobility metrics applied for

ad hoc network protocol evaluation. In The 7th

IFIP International Conference on Mobile and Wire-

less Communications Networks (MWCN), Marrakech,

Morocco.

ICEIS 2008 - International Conference on Enterprise Information Systems

140

Grandison, T. W. (2003). Trust management for internet

applications. PhD thesis, Departement of computing,

University of London.

Hekmat, R. and Mieghem, P. V. (October 2003). Degree

ditribution and hopcount in wireless ad hoc networks.

In Proceedings of the 11th IEEE International Confer-

ence on Networks (ICON 2003), pages 603–609, Syd-

ney, Australia.

Marsh, S. P. (1994). Formalising Trust as a Computational

Concept. PhD thesis, University of Stirling.

McKnight, D. H. and Chervany., N. L. (1996). The mean-

ings of trust. Technical report, Carlson School of

Management,University of Minnesota,United States.

Nguyen, C. T. and Camp, O. (2007). A bayesian network

based trust model for improving collaboration in mo-

bile ad hoc networks. International IEEE Conference

on Computer Sciences - RIVF(07).

Salber, D., Dey, A. K., and Abowd, G. D. (1999). The

context toolkit: aiding the development of context-

enabled applications. In CHI ’99: Proceedings of the

SIGCHI conference on Human factors in computing

systems, pages 434–441, New York, NY, USA. ACM

Press.

Schilit, B., Adams, N., and Want, R. (December 1994).

Context-aware computing applications. In Proceed-

ings of IEEE Workshop on Mobile Computing Systems

and Applications.

Toivonen, S., Lenzini, G., and Uusitalo, I. (2006). Context-

aware trust evaluation functions for dynamic reconﬁg-

urable systems. In Proceedings of the WWW’06 Work-

shop on Models of Trust for the Web (MTW’06), Edin-

burgh, Scotland, UK.

Yu, B. and Singh, M. P. (2001). Towards a probabilistic

model of distributed reputation management. In Pro-

ceedings of the Fourth Workshop on Deception, Fraud

and Trust in Agent Societies, Montreal, Canada, pp.

125-137.

WHAT CAN CONTEXT DO FOR TRUST IN MANETS?

141