Towards Preventive Control for Open MAS

An Aspect-based Approach

Mohamed Sedik Chebout

1, 2

, Farid Mokhati

, Mourad Badri

and Mohamed Chaouki Babahenini

Department of Mathematics and Computer Science, University of Ouargla, Ouargla, Algeria

Department of Mathematics and Computer Science, RELA(CS)2 Laboratory,

University of Oum el Bouaghi, Oum el Bouaghi, Algeria

Department of Mathematics and Computer Science, Software Engineering Research Laboratory,

University of Quebec, Trois-Rivières, Canada

Department of Mathematics and Computer Science, LESIA Laboratory, University of Biskra, Biskra, Algeria

Keywords: Open MAS, Preventive Control, Aspect Paradigm, Profiling.

Abstract: In Open MAS (Open Multi Agent Systems), agents can freely join and leave systems at any time. The inherent

specificities of such systems like dynamicity, non-determinism and emergency make their target states

difficult to achieve. Agents, in Open MAS, are often heterogeneous, self-interested with conflicting individual

goals and limited trust. Consequently, newly entered (external) agents are often considered as a potential

disturbance of systems. In this paper, we present a novel preventive control approach based on Aspect-

Oriented Programming (AOP) paradigm for mastering Open MAS’ behaviour. The proposed control process

is mainly accomplished in two steps: (1) Observing agents movements by intercepting all external requests

of agents wanting accessing to the system. A request analysis process will be held in terms of compliance

capabilities presented by this agent, and (2) Deciding, based on AspectJ constructors, either to allow agents

if the capabilities they have enable a possible progress in actual system state to the target state, or prevent it

otherwise. The proposed approach is illustrated using a MaDKit-based application.

1 INTRODUCTION

Open multi-agent systems (Open MAS) are societies

in which autonomous, heterogeneous and

independently designed entities can work towards

similar or different ends (Lopez, 2003). An open

multi-agent system is a MAS where agents may be

added, removed or evolve (e.g. modify their abilities).

Moreover, open MAS should support and deal with

three possible modifications of their composition

(Vercouter, 2000): Addition of an agent to the MAS

(in order to provide new capabilities), Removal of an

agent from the MAS (if its capabilities are obsolete)

or Evolution of an agent (the agent gains new

capabilities and loses some of its old capabilities).

Although heterogeneity and autonomy are the

required properties of the agents to build open and

flexible systems, they are always subject to

unanticipated interactions, and increase the

vulnerability of the system to the introduction of

faulty or malevolent agents, because it is not possible

to directly constrain agents’ behaviour (Vercouter

and Muller, 2010). To avoid that, it is necessary to

define and implement control mechanisms to reduce

as possible the gap between the observed behaviour

and the desired one, i.e.: is it possible for a given

system starting from any initial state, to get a given

target state in a predefined fixed time? For this, it is

essential to use (or to define) control mechanisms to

guide the system toward target state.

In addition, controllability plays a crucial role in

many control problems such as openness that we are

trying to tackle it. Moreover, to ensure this property

for a given system, a control process needs, at first

time, a monitoring mechanism to measure system

behaviour in a given moment (It’s about

observability). After that, we proceed through means

of actions to redirect current measured state for the

target one. Accordingly, observability and control are

dual aspects of the same problem (Ionete et al., 2006).

In order to reach our goal, we start by identifying

the main features relating to open MAS (Table 1).

Our aim behind this collection is to focus on a very

specific aspect around open MAS. In order to well

Chebout, M., Mokhati, F., Badri, M. and Babahenini, M.

Towards Preventive Control for Open MAS - An Aspect-based Approach.

DOI: 10.5220/0006005602690274

In Proceedings of the 13th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2016) - Volume 1, pages 269-274

ISBN: 978-989-758-198-4

269

position the current research work, the following

table summarizes some characteristics that we

consider interesting for our contribution:

Table 1: Some open MAS characteristics in literature.

open MAS Characteristics ref

Agents are owned by various stakeholders

with different aims and objectives.

Huynh, 2006 and

Vercouter, 2000

No agent can know everything about its

environment.

The agents are likely to be unreliable and

self-interested.

No central authority can control all the

agents.

An open system should support the

addition or the removal of some functions

after its design (and generally, during its

execution).

Vercouter,

2000

In open MAS, the internal aspects of

agents are inaccessible. The only

accessible information about agents is their

observable behavior from the exchange of

agents' messages.

Paes, 2005a

Open multi-agent systems are societies in

which autonomous, heterogeneous and

independently designed entities can work

towards similar or different ends.

Silva,

2007

In Open MAS context, MaDKit: Multi-Agent

Development KIT (Gutknecht and Ferber, 2000), is a

generic multi-agent platform that allows development

of open, dynamic and distributed applications based

on an organizational model called AGR (Agent,

Group, Role) presented in (Mansour and Ferber,

2007). Furthermore, Aspect-Oriented Programming

(AOP) was proposed firstly in (kiszales et al., 1997),

and since then, it has received considerable attention

in many research works. AOP provides a band new

modularization technique by encapsulating

crosscutting concerns, leading to the production of

software systems that are easier to understand,

maintain and reuse. Crosscutting concerns are defined

as system concerns (such as logging, exception

handling, etc.) that crosscut conventional system

modules (such as objects, components, and agents)

(Kiszales et al, 1997; Kiszales et al., 2001).

The programmer specifies in the AOP language

how and where to place the instrumentation (adding

new or modifying existing program functionalities

and also extending or enhancing the execution of the

program to be observable) (Sarrab, 2015). Although

its great success as AOP language, AspectJ (Kiczales

et al, 2001) is used with an indecent manner for

profiling (David et al, 2007).

In this paper, we present a novel preventive

control approach based on Aspect-Oriented

Programming (AOP) paradigm for mastering open

MAS’ behaviour. The proposed control process is

mainly accomplished in two steps: (1) Observing

agents movements by intercepting all external

requests of agents wanting accessing to the system. A

request analysis process will be held in terms of

compliance capabilities presented by this agent, and

(2) Deciding, based on AspectJ constructors, either to

allow agents if the capabilities they have enable a

possible progress in actual system state to the target

state, or prevent it otherwise.

The remainder of this paper is organized as

follows: In section 2 we give an overview of major

related work. In section 3 we present the

organizational model and its implementation under

MaDKit agent platform. We give an overview of

Aspect oriented programming and AspectJ language

in section 4. Section 5 is devoted to the presentation

of our approach. Finally, conclusion and some future

work directions are presented in section 6.

2 STATE OF THE ART

In this section, we present briefly some works

targeting the control issue in MAS:

François Klein et al. have proposed in (Klein et

al., 2009) an experimental dynamical approach to

enhance the control of the global behaviour of

a reactive MA using a Markov decision process

(MDP). This approach is based on modeling of global

system behaviour and the possible transitions

between these behaviours that system can adopt,

exploited by a control policy defined by machine

learning. The control process implemented has the

ability to directly change agent behaviour through

parameters. However, modeling system behaviour

represents a centralized view of system status, which

conflict with the distributed nature of MAS.

In the same context, Howard Carolina Felicissimo

et al. have proposed in (Felicissimo, 2005a;

Felicissimo, 2005b; Felicissimo, 2008) an extended

study on automatic regulation of open MAS based on

normative generic ontologies:

In (Felicíssimo, 2005a), the authors have

proposed a normative ontology-based approach to

define regulations over roles in open MAS. The

ontology is independent of domain and has five

related basic concepts: Role, Norm, Penalty, Action

and Place. This ontology’s structure provides a

ICINCO 2016 - 13th International Conference on Informatics in Control, Automation and Robotics

270

semantic support for agents to base their behaviour

according to norms and to reason about action

selection.

In (Felicíssimo, 2008) an approach called

DynaCROM (DYNAmic Contextual Regulation in

formation provision in open MAS) has been proposed

for smoothly applying and managing regulations in

open MASs as well as for enforcing precise

contextual norms. The DynaCROM’s main

contributions are: “(i) a definition of a top-down

classification for contextual norms, which facilitates

the tasks of elucidation, organization and

management of norm information; (ii) a contextual

normative ontology to explicitly represent the

semantic of classified norms in a meaningful way

(i.e., with a common understanding) for

heterogeneous agents; (iii) a definition of a norm

composition process, based on ontology-driven rules,

that makes it easy to update the system regulation by

both evolving norms in a unique resource (an

ontology) and by activating particular rules for

acquiring customized compositions of contextual

norms; and (iv) a solution for enforcing contextual

norms”. DynaCROM is still a work in progress.

Although these works have considerably

forwarded the domain by proposing novel approaches

for Open MAS control, they did not deal with

preventive control, which allows to avoid as soon as

possible falling into an unpredictable situation caused

by external agent’s heterogeneity and their suspect

behaviour.

The approach we propose in this paper focuses,

particularly, on the control in open MAS, in order to

provide control mechanisms to them. For this reason,

our goal is to predict disturbance source caused by the

entry of new external agents in the system. These

latters share some properties such as: heterogeneity,

different aims and objectives, internal architecture

that is inaccessible, etc. Therefore, we aim to benefit

from Aspect-Oriented Programming advantages to

avoid any kind of system disturbance caused by

external agent’s movement.

3 MADKIT AGENT PLATFORM

MaDKit: Multi-Agent Development KIT (Gutknecht

and Ferber, 2000), is a generic platform for design

and implementation of multi-agent systems. This

platform is based on an organizational model (Figure

1), named AGR (Agent, Group, Role) (Ferber et al.,

2003), rather than an agent architecture or specific

interaction model. MaDKit, also, uses an execution

engine wherein each agent is built starting from a

microkernel.

Figure 1: Simplified UML model for AGR (Mansour and

Ferber, 2007).

Considering the role definition that represents the

primordial concept in the organizational model,

(Mansour and Ferber, 2007) have defined role as a

sixfold : <Gr, Nm, Cd, Sv, At, Rl> where:

• Gr : the group identifier in which the role

belongs,

• Nm : Role name,

• Cd : Role conditions, there are two types :

 Cdj: conditions that must meet an agent

wishing play a role.

 CdU: conditions that must meet an

agent playing a role.

• Sv : a set of services defined in the role,

• At : Internal attributes of the role,

• Rl: a set of rules that govern role.

Figure 2: Role’s Architecture (Mansour and Ferber, 2007).

We use in the rest of this paper, the term capability

rather than service. According to (Mansour and

Ferber, 2007), a service is a set of functionality that

agent should suggest or ask even basic functions:

send/receive message, creation of groupe/role,

exploration of acquaintance agent list, etc. Moreover,

there are two aspects of reasoning about the

capabilities of an agent to achieve its goals,

according to (He and Ioerger, 2003) :

• Logical reasoning, which is based on whether

the agent can determine a plan that can be used

to achieve a goal (i.e. “know-how”).

Towards Preventive Control for Open MAS - An Aspect-based Approach

271

• Quantitative reasoning about levels of skill and

whether a set of tasks can be achieved within a

specified time and with quality constraints.

Capabilities in this sense are determined by

the limits of internal processing capacity”

Therefore, capability is an intrinsic functionality

attribute. Furthermore, Agent cannot realize either

functionality or achieves its goals if it does not have

the necessary capabilities. As well, capability is an

attribute related much more to performance or how

much time and effort required for doing such task,

against, service is a concept purely related to design.

4 ASPECTJ FOR PROFILING

AspectJ (Kiczales et al., 2001) is a practical aspect-

oriented extension to the Java programming

language. AspectJ is the most widely used aspect-

oriented programming language; it supports the

definition of aspects, advices, join points, and

pointcuts. Join points (Kiczales, 2001) are well-

defined points in the execution of the program;

pointcuts are collections of join points; advices are

special method-like constructs that can be attached to

pointcuts and aspects which are modular units of

crosscutting implementation, comprising pointcuts,

advice, and ordinary Java member declarations.

For more information and details about AspectJ

constructs, the reader is invited to consult (Laddad,

2009), (Apel and Batory, 2010).

Our proposal for observing open MAS behaviour

is to use AspectJ for profiling MAS specific attributes

(i.e. dynamic analysis). This choice is motivated by

the fact that traditional profilers like (JVMPI: Java

Virtual Machine Profiler Interface) (Liang and

Viswanathan, 1999) and its successor (JVMTI: Java

Virtual Machine Tool Interface) (O’Hair, 2006)

present some drawbacks: (1) it is a fixed interface

and, as such, can only enable predefined types of

profiling; (2) enabling the JVMPI often dramatically

reduces performance. Thus, in AspectJ, the

programmer specifies how and where to place the

instrumentation. AspectJ also allows for both compile

time and load time instrumentation and makes use of

BCI (ByteCode Instrumentation) libraries to

instrument the application.

In (David et al., 2007), the authors have

demonstrated that AspectJ, after investigating four

common profiling problems (heap usage, object life

time, wasted time and time-spent), was sufficiently

flexible to support the four profiling examples. In

addition, it was reasonably efficient in most cases, of

course with some uncovered limitations according to

(David et al., 2007), like: load-time weaving standard

libraries, state association, array allocation join point

and synchronization.

5 PROPOSED APPROACH

Our proposal (Figure 3) is to control a priori open

MAS by intercepting all external requests from agents

that want to enter in the system. As it is quoted above,

open MAS share properties according to agent’s

heterogeneity, different agent’s aims and objectives.

This explains that the integration of a new agent in the

system might be a potential source of current state

disruption, and therefore emergent behaviour will

take place.

Figure 3: The methodologies of our approach.

In order to avoid falling into such situation, we

propose to implement an observer aspect, which is

responsible for the permanent monitoring of the entry

of new agents, once this last wants to get in system.

On the one hand, observer aspect represents,

actually, a part of control process, which is also

composed of a module that deals with evaluation of

agent capabilities; this last, is the core of the control

process. On the other hand, agent’s role request is

composed, as it is mentioned in section 3, by a set of

fields. We focus in our context on Cd (Role

Condition) especially CdU field that denote

ICINCO 2016 - 13th International Conference on Informatics in Control, Automation and Robotics

272

conditions that must meet an agent playing a role. In

this case, Agent capabilities evaluation process

analyses the compliance of role in terms of agent

capabilities. We suppose that provided role by agent

includes several capabilities, every capability is

associated with a control degree. We assume, also,

that we cannot judge the reliability of the capability

offered by the agent (Table1, characteristic n°3).

After that, if the rate of agent capability reaches

certain level (we choose initially a considerable rate:

up then 70%), the agent is allowed to enter, otherwise,

it will be prevented using a specific aspect. As well,

capabilities evaluation process must analyse another

agent’s request and so on.

The control process proceeds until the target

system state is reached. According to (Klein et al.,

2009), current system state measurement, depend

strongly of the studied system. Therefore, each

system has a different configuration compared to

other systems. Thus, we adopt a performance measure

related to control time, i.e.: the control process

elapsed time to master the behaviour of the studied

system.

Figure 4: Control process time.

For that, we propose to implement an aspect

named profiler Aspect. This latter enables

intercepting the first call to the first explicit agent

requesting role. Then, we will capture current state

each sample time (Figure 4), to compare it with the

target state (i.e expected system result). We check

also, if the total control time is reached without

getting desired result, the preventive control process

present a failure, and in this case the system no longer

has the controllability property.

6 CONCLUSION AND FUTURE

WORK

In this paper, we have studied the controllability of

Open multi-agent system, using Aspect oriented

programming paradigm under MaDKit platform.

Our approach consists, initially, to predict any

potential source of system disturbance caused by

openness property, by analyzing system dynamics in

terms of external agents that get in the system,

through profiling mechanism using AspectJ

language. Thus, we exhibit that AspectJ has

demonstrated its efficiency in such situation against

existing profilers that limit profiling types in a very

reduce number and its negative impact on analysed

system performance. Furthermore, the controller uses

the power of AspectJ constructs to act in a way to

enable or block all requests coming from entering

agents, having a certain degree of impeding of current

system state.

In fact, this work is in progress, and the approach

explained in this paper represents a first step of a

novel open MAS control process. As future work, we

plan to extend the proposed approach in order to

support more features relating to open MAS control,

especially, posteriori control, using reinforcement

learning, or simply, how to endow agents with

learning mechanisms for achieving their goals?

REFERENCES

Apel, S., Batory, D. 2010. ‘How aspectj is used: an analysis

of eleven aspectj programs’. Journal of Object

Technology (JOT), 117-142.

David, J, P., Matthew, W., Robert, B ., Paul, H, J, K., 2007

. ‘Profiling with AspectJ’, Software—Practice &

Experience, 37(7), 747-777.

Felicíssimo, C., Lucena, D., Carvalho, G., Paes, R., 2005.

‘Normative ontologies to define regulations over roles

in open multi-agent systems’ , In Proceedings AAAI

Fall Symp, 171 -176.

Felicíssimo, C., Lucena, D., 2005. ‘An approach to regulate

open multi-agent systems based on a generic normative

ontology’ In Proceedings of the 1st Workshop on

Software Engineering for Agent-oriented Systems

(SEAS).

Felicíssimo, C., Chopinaud, C., Briot, J, P., Seghrouchni,

A., Lucena, C., 2008. ‘Contextualizing normative open

multi-agent systems’. In Proceedings of 23rd Annual

ACM Symposium on Applied Computing (SAC 2008), 1,

52-59.

Ferber, J., Gutknecht, O., Michel, F., 2004. ‘From Agents

to Organizations, an Organizational View of Multi-

Agent Systems’. Agent-Oriented Software Engineering

Towards Preventive Control for Open MAS - An Aspect-based Approach

273

(AOSE) IV, P. Giorgini, Jörg Müller, James Odell, eds,

Melbourne, July 2003, LNCS 2935, 214-230.

Gradecki, J, D., Lesiecki, N., 2003. ‘Mastering AspectJ:

Aspect- Oriented Programming in Java’. Wiley

Publishing, Inc., Indianapolis, Indiana.

Gutknecht, O., Ferber, J., 2000. ‘MaDKit : une architecture

de plate-forme multi-agent générique’. Rapport de

recherche, Laboratoire d’Informatique, de Robotique

et de Microélectronique de Montpellier (LIRM),

Université de Montpellier II.

He, L., and Ioerger, T, R., 2003. ‘A quantitative model of

capabilities in multi-agent systems’. In Proceedings of

International Conference on Artificial Intelligence,

730-736.

Huynh, T. D., Jennings, N. R., Shadbolt, N. R. 2004b.

‘FIRE: An integrated trust and reputation model for

open multi-agent systems’. In Proceedings of the 16th

European conference on artificial intelligence (ECAI),

18–22.

Ionete, C., Cela, A., Gaid, M, B., 2006. ‘Controllability and

observability of input/output delayed discrete

systems’, In Proceedings of IEEE Amer. Control

Confeerence, 3513-3518.

Kiczales, G.,Hilsdale, E., Hugunin, J., Kersten, M., Palm, J.

& Griswold, W.G., 2001. ‘An Overview of aspectJ’. In

proceedings of the European Conference on Object-

Oriented Programming.

Kiczales, G., Lamping, J., Menhdekar, A., Maeda, C.,

Lopes, C., Loingtier, J-M., Irwin, J., 1997. ‘Aspect-

oriented programming’. In Mehmet Aksit and

Satoshi Matsuoka, editors, European Conference on

Object-oriented Programming, Vol 1241 of Lecture

Notes in Computer Science, 220–242. Springer.

Klein, F., Bourjot, C, Chevrier, V., 2009. ‘Contribution to

the Control of a MAS's Global Behaviour:

Reinforcement Learning Tools’. Alexander Artikis and

Gauthier Picard and Laurent Vercouter. Engineering

Societies in the Agents World IX - 9th International

Workshop, ESAW 2008, Saint-Etienne, France,

September 24-26, 2008, Revised Selected Papers, 5485,

Springer-Verlag Berlin Heidelberg, P.173-190, Lecture

Notes in Computer Science, 978-3-642-02561-7.

<10.1007/978-3-642-02562-410>. <inria-00400348>

Laddad, R., 2009. ‘AspectJ in Action’ . Enterprise AOP

with Spring Applications. Manning, 2nd edition.

Liang, S., Viswanathan, D., 1999. ‘Comprehensive

profiling support in the Java Virtual Machine’.

Proceedings of the USENIX Conference On Object

Oriented Technologies and Systems. USENIX

Association: Berkeley, CA, 229–240.

López, F., 2003 ‘Social Power and Norms : Impact on agent

behaviour’. PhD thesis, Univ. of Southampton.

Mansour, S., Ferber, J., 2007. ‘Agent Groupe Role‚ et

Service: Un modèle organisationnel pour les systèmes

multi-agents ouverts’. JFSMA07.

O’Hair, K., 2006. ‘The JVMPI transition to JVMTI’.

http://java. sun.com/developer /technicalArticles/

Programming/jv mpitransition/.

Paes, R., Carvalho, G.R., Lucena, C.J.P., Alencar, P.S.C.,

Almeida, H.O., Silva, V.T. 2005a. ‘Specifying Laws in

Open Multi-Agent Systems’. In Agents, Norms and

Institutions for Regulated Multi-agent Systems

(ANIREM), AAMAS.

Sarrab, M. 2015. ‘Bytecode instrumentation mechanism for

monitoring mobile application information flow’. In

International Journal on Security and Networks, 10(3).

Silva, V., Duran, F., Guedes, J., & Lucena, C. 2007.

‘Governing multi-agent systems’. Journal of Brazilian

Computer Society. Special Issue in Software

engineering for multi-agent systems, 13(2).

Vercouter L., 2000. ‘A Distributed Approach to Design

Open Multi-agent Systems’. In 2nd International

Workshop Engineering Societies in the Agents' World

(ESAW).

Vercouter, L., Muller, G., 2010. ‘L.I.A.R.: Achieving social

control in open and decentralized multiagent systems’.

Applied Artificial Intelligence: An International

Journal, 24(8), 723-768, DOI: 10.1080/

08839514.2010.499502.

Wang, Q., Gao, H., Alsaadi, F and Hayat, T. 2014. ‘An

overview of consensus problems in constrained multi-

agent coordination’. Systems Science and Control

Engineering, 2(1), 275.

ICINCO 2016 - 13th International Conference on Informatics in Control, Automation and Robotics

274