Trusted Community

A Trust-based Multi-Agent Organisation for Open Systems

Lukas Klejnowski

, Yvonne Bernard

, Gerrit Anders

, Christian M

uller-Schloer

and Wolfgang Reif

Institute of Systems Engineering, Leibniz University Hannover, Hannover, Germany

Institute for Software & Systems Engineering, Augsburg University, Augsburg, Germany

Keywords:

Trust, Multi-Agent Systems, MAS Organisation, Organic Computing, Self-organisation, Desktop Grid.

Abstract:

In this paper, the multi-agent organisation Trusted Community is presented. Trusted Communities are formed

and joined by self-organised by agents with strong mutual trust relations and the purpose to increase their per-

sonal utility. Trusted Communities are maintained by management actions delegated by a designated member

called Trusted Community Manager, having the goal to preserve and optimise the composition and stability

of this organisation. This organisation provides performance beneﬁts for their members by improving interac-

tion efﬁciency, information sharing and cooperation between the agents. In the work presented here, Trusted

Communities are conceptually deﬁned and the application in an open Desktop Grid System is discussed.

1 INTRODUCTION

When realising technical systems based on an

open multi-agent system model, we face challenges

through agents that join and leave the system arbi-

trarily and show various types of behaviours rang-

ing from cooperative to selﬁsh or even adversary (in

terms of having the aim to disrupt the operation of

the system). In previous work, it has been shown that

in systems with these characteristics, trust can be ap-

plied to model the relationships between agents and

that these trust relations can be used to improve the

performance and robustness (towards misconducting

agents) of these systems. In this work, we take a step

further and present an approach where enduring and

mutual trust relations lead to a self-organising process

resulting in a higher form of organisation between

trustworthy agents. This organisation is referred to

as Trusted Community (TC) and is characterised by a

decentralised, yet hierarchically managed, operation

that provides performance beneﬁts for their members

by improving interaction efﬁciency, information shar-

ing and cooperation. The management allows for op-

timisation of the composition and directed actions to

preserve the stability of the organisation. This is es-

sential, as composition and stability of an organisa-

tion consisting of self-interested members can easily

become issues in similar approaches, esp. when leav-

ing agents generate feedback effects.

This paper is organised as follows: Section 2 de-

ﬁnes the system model, Section 3 outlines the applica-

tion. Section 4 lays out the evaluation concept, while

related work is presented in Section 5. The paper is

concluded in Section 6.

2 SYSTEM MODEL

The Trusted Community is an organisation for agents

that have persistent mutual trust relations which allow

to establish a subsystem inside a hosting multi-agent

system as depicted in Figure 1. In this organisation,

Outbound

interactions

Trust

based

interactions

Inbound

(

member

)

interactions

Trusted Community

Open (hosting) system

Unassociated agents

Misconducting

agent

Figure 1: System view on Trusted Communities.

interaction partners are mainly chosen among mem-

bers (we refer to this as inbound or kinship-motivated

interactions). Agents in the examined systems are

self-interested, that is, they will only consider form-

ing and participating in such a subsystem when the

312

Klejnowski L., Bernard Y., Anders G., Müller-Schloer C. and Reif W..

Trusted Community - A Trust-based Multi-Agent Organisation for Open Systems.

DOI: 10.5220/0004332003120317

In Proceedings of the 5th International Conference on Agents and Artiﬁcial Intelligence (ICAART-2013), pages 312-317

ISBN: 978-989-8565-38-9

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

beneﬁt provided is measurably higher than in an unas-

sociated state (ignoring or being unable to join an ex-

istent Trusted Community). A TC thus does not pos-

tulate an explicit group goal, as for example a coali-

tion. To apply a TC in a hosting system, the follow-

ing requirements have to be met: The system needs

to be open, i.e. agents can join and leave at will. All

agents commit to using some standardised mechanics

(production engine) of the (technical) hosting system.

The respective performance is measurable via a glob-

ally deﬁned utility function for an agent i of the form

, c

), contrasting a reward function r

with a func-

tion c

quantifying the cost to reach the reward. The

hosting system provides methods to discern trustwor-

thy from untrustworthy agents through interactions,

assigning trust values to each agent and providing rep-

utation information. The following design pattern de-

picted in Figure 2 formalises the requirements on the

underlying agent model:

Agent

ReputationProvider

DirectTrustValue

ReputationValue

-repProv

-knownAgents 0..*0..*

-ratedAgent

0..*

-tcManager

0..*

«provides»

-members 2..*

0..*

-ratedAgent 1

0..*

-repProv 0..*

-agents 0..*

Figure 2: Trusted Community design pattern.

As shown here, a TC is composed of at least 2

member agents of which exactly one is the TC Man-

ager. The agents are able to assign direct trust val-

ues to each other based on the outcomes of their in-

teractions and by contacting a reputation provider,

they make these trust values available to other known

agents. The reputation provider is not further spec-

iﬁed (central entity or realised as broadcast to other

agents to retrieve direct trust values), but the TC itself

is also a reputation provider meaning that members

can share a TC wide reputation.

2.1 Trusted Community Lifecycle

Trusted Communities are formed self-organised by

unassociated agents in a hosting systems. Regarding

TCs, this system can be in one of the following three

phases depicted in Figure 3:

Pre-organisation Phase. The system is in a phase

where all agents are unassociated and no organisa-

tion exists. The agents have to apply safety measures

because of uncertainty about other agents’ behaviour

and constantly rate their interactions according to the

trust mechanics of the system. With time passing,

trustworthy agents develop mutual trust relations. At

Trust and reputation system

TC Maintenance Phase

Pre-Organisation Phase

TC Formation Phase

Agents initiate formation process

Initial management election

TC Manager

initiates

dissolution

Active TC-

Expansion

Strategies

Task

Delegation

Strategies

Member

Control

Strategies

Membership

Evaluation

Strategies

Organisation benefit strategies

Interaction

efficiency

Information

sharing

Cooperation

Figure 3: Trusted Community lifecycle and composition.

this time they consider forming a Trusted Commu-

nity in hope to improve their utility. This decision is

based on an analysis of their current utility U

and the

expected utility predictUtility(i, TC

) when being a

member of the TC j, formalised by formation criteria.

Mutual trust alone does not initiate TC formation, as

this would force the formation of TCs per design. In-

stead we want to apply TCs only when beneﬁcial (in-

crease in U

) and adaptive towards the hosting system

state. When a critical number of agents decide to ini-

tiate the formation process, the next phase is reached.

TC Formation Phase. The formation phase is charac-

terised by negotiations of the potential members about

membership of additional agents. This is necessary as

not all agents have had the same interaction partners.

Finally, in this phase the manager of the TC is elected,

starting the maintenance phase.

TC Maintenance Phase. This is the main phase in

the TC lifecycle. Here, the TC is already assembled,

having a TC manager that can assign roles to mem-

bers and release norms, as well as agents that actively

work together in an organised fashion. As the agents

joined in order to improve their utility U

, they ex-

ecute periodic checks that are related to the initial

analysis predictUtility(i, TC

) in the pre-organisation

phase (but can be performed with more information at

hand). If members discover that their utility U

did not

improve (comparing to their utility when being unas-

sociated), they leave the TC. If a critical number of

members leave, the dissolution of the organisation is

initiated and the pre-organisation phase restored.

Apart from the members of the Trusted Commu-

nity, the unassociated agents continue to interact with

each other and with the TC members (constrained by

interaction rules of the TC). These agents may then

reach a state where they ﬁnd joining the TC beneﬁ-

cial, as well as acceptable from the point of view of

the TC. As we are considering a dynamic system, the

threshold to joining a TC can be reached at different

TrustedCommunity-ATrust-basedMulti-AgentOrganisationforOpenSystems

313

phases of the system: Consider for example an agent

joining late and thus having no trust relations to the

initial members of the TC, thus being unable to be an

initial member itself.

These mechanics are referred to as lifecycle of

the Trusted Community as we have transitions from

the maintenance phase where a TC exists, to the pre-

organisation phase. In this phase, another TC can

form if the trust requirements and formation criteria

are met again by agents. A TC thus emerges depen-

dent on signiﬁcant criteria and the state of the hosting

system. In the following, we focus on the operation

of a TC in the maintenance phase.

2.2 Organisation Beneﬁt Strategies

Forming/joining a Trusted Community is an act of

self-interest for agents because TCs provide beneﬁts

that increase the utility U

of member agents. Obvi-

ously, the nature of these beneﬁts depends on the un-

derlying application: Still, member agents have the

same utility deﬁnition U

as unassociated agents, ben-

eﬁts must therefore be directed at improving the exact

interactions that unassociated agents have, increasing

the same reward function r

. Despite their application-

dependency, we can afﬁrm that the organisation ben-

eﬁt strategies are part of the following classes:

Interaction Efﬁciency. Interactions that are executable

within the hosting system but proﬁt from being exe-

cuted between agents with high mutual trust, thus TC

members. This refers mainly to subadditive costs or

superadditive beneﬁt interactions. Consider for ex-

ample the search for a suited interaction partner, or

the execution of interactions without the necessity to

apply additional safety measures.

Cooperation. This class represents strategies that can-

not be executed in an environment with uncertainty

about the trustworthiness of the involved interaction

partners, because of safety considerations. Hence,

strategies here are exclusive to member agents. Ex-

amples are the delegation of the execution of funda-

mental tasks to a central entity (e.g. the TC Manager)

or the cooperative detection of collusion (relying on

the absence of colluding agents in the composition).

Information Sharing. In general, sharing informa-

tion is necessary in order to perform other organisa-

tion beneﬁt strategies. Additionally, agents will not

share certain information in the hosting system when

the trustworthiness of the recipient is in doubt. The

TC provides a structure where this information can

be shared and processed safely and in scale (among

TC members). An example of sensitive information,

with access restricted to members, are personal obser-

vations (local world model) which could be abused by

untrustworthy agents.

By providing adequate organisation beneﬁt strate-

gies, the additional overhead of being a member

of a TC (contributing to c

) is marginalised and

the formation of a TC a worthwhile goal for self-

interested agents. In the following, the Trusted Com-

munity Manager is described and the mechanics of

the Trusted Community (which produce the over-

head) are detailed.

2.3 Trusted Community Manager

The Trusted Community Manager (TCM) is the

elected leader of the Trusted Community. As such

it acts as an the active representative of an institution

releasing norms and sanctioning infringement. Main

goals of this entity are to regulate access to the organ-

isation, improve the experience of the members, and

most importantly to preserve the existence of the TC.

The according function blocks (depicted in Figure 3

in the maintenance phase) are composed of mainly

recurring tasks that can be delegated to members or

executed by the TCM itself. Examples of tasks are

the gathering of data through observation or speciﬁc

interactions with non-member agents. In the follow-

ing, the function blocks are detailed:

Active TC Expansion Strategies. The initial compo-

sition of a Trusted Community is based on the form-

ing agents. Due to connectivity and dynamics aspects

of the hosting system, possibly not all suited agents

were involved in the formation process. The task of

the Trusted Community Manager is therefore to ﬁnd,

observe and evaluate potential members with the goal

to optimise the composition of the Trusted Commu-

nity and in order to improve its effectiveness.

Membership Evaluation Strategies. Agents join the

Trusted Community with the expectancy to experi-

ence a higher utility U

. Therefore members periodi-

cally check whether their membership satisﬁes this by

evaluating c

, the overhead (through delegated tasks,

kinship commitment etc.), and r

, here the interaction

efﬁciency. Members leaving the TC as a result of

these examinations are degrading the efﬁciency of the

TC, as they reduce the number of available member

interaction partners and thus r

of remaining mem-

bers. Besides, management tasks delegated by the

TCM have to be distributed among less agents, in-

creasing c

of single members leading to the threat

of a positive feedback effect. The TC Manager is

therefore obliged to try and improve the utility U

a potentially leaving member i. This can e.g. be re-

alised by means of (short-term) reduction of the over-

head of this agent via re-delegation of its management

tasks, increasing its beneﬁts by means of prioritisa-

ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence

314

tion. However, the TCM needs to balance the costs of

preserving the membership of this agent with the ad-

ditional overhead introduced for other members. The

key of this function block is therefore an opportunity

cost analysis for under-performing members.

Member Control Strategies. This function block re-

gards the necessity to observe and inﬂuence the be-

haviour of member agents. The requirement to ob-

serve member behaviour despite their (proven) trust-

worthiness stems from the fact that we are dealing

with self-interested agents and dynamic systems, a

combination that can lead to various forms of mis-

conduct. Consider for example agents with a strategy

that lets them behave trustworthy until membership in

a TC is accomplished, only to start to defect, reaping

beneﬁts without committing themselves to any tasks.

The TC Manager is given the capability to observe

the behaviour of member agents, be it at random, pe-

riodically or in case of suspicion. This can be realised

by utilising a light-weight version of the reputation

mechanism of the hosting system. Again, the over-

head of introducing safety measures needs to be care-

fully balanced in order to preserve the effectiveness

of the TC. Finally, when detecting misconduct, the

TCM will sanction the agent, the exclusion from the

TC being the ultimate form of punishment.

Task Delegation Strategies. All function blocks are

composed of tasks and the Trusted Community Man-

ager cannot execute all these tasks itself - this would

reduce its utility U

and imply leaving the TC, intro-

ducing a high membership ﬂuctuation and render the

TC unmanageable. Consequently, the TCM has to

apply a task delegation model with the requirement

of fairness, as an unbalanced delegation would lead

to under-performing agents and thus to leaving mem-

bers. Besides being fair, the task delegation needs

to be adaptive to short-term relief of single agents as

mentioned in the description of the membership eval-

uation function block. The supervision of task execu-

tion is subject to member control strategies.

The Trusted Community Manager is a key ele-

ment in the Trusted Community concept. On the one

hand, it represents a higher-ranking level among the

otherwise equal members of the Trusted Community,

yet on the other hand, the TCM is just an agent partic-

ipating in the Trusted Community in order to increase

its own utility. This is fundamental to the concept in

order to avoid the introduction of an external element

with goals that are beyond justiﬁcation. With regard

to the self-interest of the agent being TCM, compen-

sation is necessary for the aforementioned tasks exe-

cuted by the TCM to balance its r

and c

. The most

general form of compensation is to assign all sys-

tem interactions that do not aim at managing the TC

(but generate beneﬁt) to other member agents - these

agents then act on behalf of their manager. Finally,

in consideration of the open nature of the system, a

Trusted Community Manager has to be elected anew

whenever the current TCM either leaves the Trusted

Community (utility considerations) or the system as

such. This election is performed analogously to the

initial election at formation phase.

3 APPLICATION SCENARIO

In this section we describe how Trusted Communities

can be deployed in an exemplary hosting system from

the Desktop Grid Computing domain: We assume an

open, distributed and volunteer-based Desktop Grid

System in the tradition of systems like XtremWeb

(Fedak et al., 2001), to which we refer as Trusted

Desktop Grid (TDG). The system is designed with-

out central control (p2p) and the applications regarded

produce bag-of-task jobs, i.e. they are composed of

tasks (work units) that are independent of each other.

A system like this is suited for scenarios where most

clients run applications that produce grid jobs and

thus are in high demand of computing resources. Ad-

ditionally, we are considering agents that are in charge

of the grid client on the machines and make decisions

on behalf of their users (especially about resource se-

lection). Due to the open nature of the system, we

have to deal with agents that show various types of be-

haviour (from altruistic to untrustworthy) in order to

achieve their self-interested goal of scheduling their

own jobs as efﬁciently as possible. According to the

taxonomy of (Choi et al., 2008) and taking the re-

source perspective, we therefore classify the poten-

tially participating agents of this Desktop Grid Sys-

tem as: egoistic, volatile, distributed over the internet,

dynamic, faulty and heterogeneous.

In previous work, for example (Bernard et al.,

2011), we have shown that by using a trust and repu-

tation mechanism combined with techniques from Or-

ganic Computing, cf. (M

uller-Schloer and Schmeck,

2011), we are able to isolate uncooperative agents

and minimise their inﬂuence on the system perfor-

mance in systems like this. The result of the isolation

process was an implicit Trusted Community (iTC), a

loose coupling of interaction partners without global

membership function. ITCs, despite being successful,

leave room for improvement, as unassociated agents

miss essential opportunities to increase their utility,

because they have to account for the general uncer-

tainty. By applying the MAS organisation Trusted

Community (as detailed in section 2), we plan to ac-

count for these opportunities and further raise the util-

TrustedCommunity-ATrust-basedMulti-AgentOrganisationforOpenSystems

315

ity of the agents by providing them beneﬁts through

interaction efﬁciency, information sharing and coop-

eration.

To allow for a self-organised formation of a TC in

a suited hosting system, we need to deﬁne agent util-

ity, formation criteria and organisation beneﬁt strate-

gies. The remaining mechanics (lifecyle, mainte-

nance etc.) of the Trusted Community are generic.

In the Trusted Desktop Grid, agents use each others’

spare resources to process jobs generated by user ap-

plications. Their goal is to schedule single work units

on available worker agents (resource selection), such

that they minimise the time it takes to receive valid

results. This is formalised in Desktop Grid metrics

like ﬂow time, makespan and turnaroundtime, cf. for

example (Zhou and Lo, 2006). On the other hand,

overhead is introduced through communication with

other agents and working for other agents, because al-

though a user volunteers its machine, we assume that

from the user perspective it is best to receive good

performance for own jobs without having to commit

resources. A utility function U

for the TDG thus in-

corporates these aspects in r

and c

Formation criteria for initiating TC formation are

drawn from suboptimal agent states. These are for ex-

ample indicated by a low submit/work-ratio, meaning

that an agent has a high overhead due to commitment,

or a high average number of communication acts nec-

essary to ﬁnd a suited worker that accepts own work

units. Being a member of a Trusted Community is

likely to improve these states, as TCs provide scala-

bility and kinship commitment to members, therefore

TC formation is aspired.

Finally, the most fundamental aspects to deﬁne

when applying TCs are the organisation beneﬁt strate-

gies. Figure 4 depicts these for the TDG Scenario:

Interaction

efficiency

Information

sharing

Cooperation

Omit safety

measures

(esp. replication)

Observed world

model (situation

description)

Utility

Coordinated

processing of

jobs

Shared trust and

reputation

management

Collusion

detection

Kinship-driven

interactions

realisation

Figure 4: Exemplary TDG organisation beneﬁt strategies.

In the TDG, interactions between agents are

mainly restricted to the processing of each others’

work units, negotiations about the respective terms

and the exchange of information necessary for the

identiﬁcation of suited partners (esp. reputation). For

TC members these interactions are more efﬁcient be-

cause of scalability and kinship commitment. A par-

ticular interaction efﬁciency beneﬁt is that no work

unit replication needs to be performed, as all work

units can be assigned to members and thus valid re-

sults can be expected.

Information to be shared inside a TC is sensi-

tive to trust reasoning. In the TDG, sharing the ob-

served world model or own utility of an agent with un-

trustworthy agents could be abused by misconducting

agents that adapt their strategies accordingly. Con-

sider for example an agent that gets to know the exact

resource selection criteria applied by an other agent.

A misconducting agent could actively adapt to ap-

pear as inappropriate worker in order to avoid being

requested to process a work unit and commit its re-

sources.

Unlike interaction efﬁciency strategies, coopera-

tion strategies provide interaction opportunities not

readily executable by unassociated agents in the host-

ing system. In the Trusted Desktop Grid, centralised

scheduling is a good motivation for TCs: Centralised

scheduling is generally more effective than decen-

tralised, because fairness and predictability can be

incorporated into scheduling strategies more easily.

However, centralised scheduling does not scale well

and is therefore hardly applicable in an open, un-

reliable, peer-to-peer environment - in contrast to

the size-constrained and trustworthy environment a

Trusted Community represents. Additional cooper-

ation opportunities for TC members in the TDG are

a shared trust and reputation management (less mes-

sage overhead, more information) and collusion de-

tection by cooperative observation of untrustworthy

agents.

4 EVALUATION

We will evaluate the Trusted Community concept

in the Trusted Desktop Grid applying the scenario-

dependent utility function, formation criteria and or-

ganisation beneﬁt algorithms presented in the previ-

ous section. We will especially focus on showing that

in an open Desktop Grid, system states arise in which

agents proﬁt from forming a TC. The reward r

utility U

of the agents will be composed of standard

Desktop Grid metrics like turnaroundtime. We will

continue to show that this organisation can be made

stable (low member ﬂuctuation) and its composition

optimised, by introducing a TC Manager. Especially

its overhead balancing and short-time relief of man-

agement tasks capabilities are seen as promising can-

didates for keeping members associated. Our ﬁndings

will be compared with the application of related or-

ganisation paradigms in the same system, providing

additional evaluation data as contribution.

ICAART2013-InternationalConferenceonAgentsandArtificialIntelligence

316

5 RELATED WORK

In the survey on multi-agent organisation paradigms

by (Horling and Lesser, 2005), the MAS organisa-

tion clan proposed by (Grifﬁths, 2005) has been de-

scribed as closely paralleling the concept of congre-

gations by (Brooks and Durfee, 2003) but addition-

ally incorporating trust as a key aspect. Thus we

refer to clans as “congregations with trust” and in

the same line of argumentation we refer to Trusted

Communities as “congregations with trust and hierar-

chy” in regard of the role of the Trusted Community

Manager. Further theoretic work on MAS organisa-

tion mechanics has also been conducted by (Math-

ieu et al., 2002). As for our application scenario,

research on open, distributed, multi-agent-based and

trust-enhanced Desktop Grid systems can for instance

be found in (Domingues et al., 2007), (Shudo et al.,

2005) and (Dyson et al., 2004). Additional coverage

of MAS organisations in similar scenarios has also

been considered in (Thabet et al., 2011), (Abdallah

et al., 2004) and (Wang and Vassileva, 2004).

6 CONCLUSIONS

In this paper we have proposed a novel multi-agent

organisation called Trusted Community for the use in

open, distributed systems where trust considerations

are important. We have deﬁned the system model, de-

scribing the lifecycle and management mechanics of

a TC. In particular, we have emphasized the necessity

to balance organisation beneﬁt algorithms and over-

head introduced by membership, as agents form and

join TCs out of self-interest. We have further sketched

how TCs can be deployed in a Trusted Desktop Grid

scenario and how we will evaluate the applicability in

future work. Finally, we have stated how our research

is related to similar work in the area of multi-agent

organisations and agent-based Desktop Grid systems.

ACKNOWLEDGEMENTS

This research is funded by the research unit “OC-

Trust” (FOR 1085) of the German research founda-

tion (DFG).

REFERENCES

Abdallah, S., Zhang, H., and Lesser, V. (2004). The role

of an agent organization in a grid computing envi-

ronment. In Proc of the 14th Int Conf on Automated

Planning and Scheduling, Workshop on Planning and

Scheduling for Web and Grid Services.

Bernard, Y., Klejnowski, L., Cakar, E., H

ahner, J., and

uller-Schloer, C. (2011). Efﬁciency and Robust-

ness Using Trusted Communities in a Trusted Desk-

top Grid. In 5th IEEE Conference on Self-Adaptive

and Self-Organizing Systems Workshops. IEEE.

Brooks, C. and Durfee, E. (2003). Congregation formation

in multiagent systems. Autonomous Agents and Multi-

Agent Systems, 7(1).

Choi, S., Buyya, R., Kim, H., and Byun, E. (2008). A Tax-

onomy of Desktop Grids and its Mapping to State of

the Art Systems. Technical report, Grid Computing

and Distributed Systems Laboratory, The University

of Melbourne.

Domingues, P., Sousa, B., and Moura Silva, L. (2007).

Sabotage-tolerance and trustmanagement in desktop

grid computing. In Fut. Gener. Comput. Syst. 23, 7.

Dyson, J., Grifﬁths, N., Lim, H., Jarvis, S., and Nudd, G.

(2004). Trusting agents for grid computing. 2004

IEEE Int Conf on Systems, Man and Cybernetics

(IEEE Cat. No.04CH37583).

Fedak, G., Germain, C., Neri, V., and Cappello, F. (2001).

XtremWeb: a generic global computing system. In

Proc First IEEE/ACM Int Symp on Cluster Computing

and the Grid. IEEE Comput. Soc.

Grifﬁths, N. (2005). Cooperative clans. Kybernetes,

34(9/10).

Horling, B. and Lesser, V. (2005). A Survey of Multi-Agent

Organizational Paradigms. The Knowledge Engineer-

ing Review, 19(4).

Mathieu, P., Routier, J.-C., and Secq, Y. (2002). Princi-

ples for dynamic multi-agent organizations. In Intel-

ligent Agents and Multi-Agent Systems, volume 2413

of LNCS. Springer Berlin / Heidelberg.

uller-Schloer, C. and Schmeck, H. (2011). Organic Com-

puting - Quo Vadis? In Organic Computing - A

Paradigm Shift for Complex Systems, chapter 6.2.

Birkh

auser Verlag.

Shudo, K., Tanaka, Y., and Sekiguchi, S. (2005). P3: P2p-

based middleware enabling transfer and aggregation

of computational resources. In Proc. IEEE Int Symp

on Cluster Computing and the Grid CCGrid ’05, vol-

ume 1.

Thabet, I., Bouslimi, I., Hanachi, C., and Gh

edira, K.

(2011). A multi-agent organizational model for grid

scheduling. In Agent and Multi-Agent Systems: Tech-

nologies and Applications, LNCS. Springer Berlin /

Heidelberg.

Wang, Y. and Vassileva, J. (2004). Trust-based commu-

nity formation in peer-to-peer ﬁle sharing networks.

In Proc of the 2004 IEEE/WIC/ACM Int Conf on Web

Intelligence, WI ’04, Washington, DC, USA. IEEE

Computer Soc.

Zhou, D. and Lo, V. (2006). WaveGrid: a scalable fast-

turnaround heterogeneous peer-based desktop grid

system. In Proc 20th IEEE Int Parallel & Distributed

Processing Symp. IEEE.

TrustedCommunity-ATrust-basedMulti-AgentOrganisationforOpenSystems

317