Maintaining Organizational Multi-agent Systems:

A Reorganization-based Preventive Approach

Nawel Ghrieb

1,3

, Farid Mokhati

2,3

and Tahar Guerram

2,3

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

Department of Mathematics and Computer Science, University of Oum El Bouaghi, Algeria

RelaCS2Laboratory, University of Oum El Bouaghi, Algeria

Keywords: Preventive Maintenance, OCMAS, Software Quality, Reorganization, AOP.

Abstract: In this article, we propose a preventive maintenance approach for Organization-Centered Multi-Agent

System (OCMAS). This approach is based on the quality assessment for the maintenance of OCMAS. The

quality of OCMAS is monitored by Aspect Oriented Programming (AOP) techniques, in order to detect any

abnormal regression in the quality of the system or that of the agents composing it. This degradation in

quality is, usually, an indication of problems that may arise in the structure of the organization or its

functionalities. In the context of this work, we are interested in the functional problems that can affect the

system and we treated them by reallocating agents to roles. This reallocation is, generally, necessary when

the agent is not able to achieve its objectives, which leads to a degradation of the overall quality of the

system. Our maintenance approach must therefore anticipate these problems and react by reorganizing the

running system in order to improve its quality and allow it to resume its normal behaviour.

1 INTRODUCTION

Organizational multi-agent systems engineering has

been proposed as a way to design complex systems

that adapt to their environment. Agents interact and

may assign tasks to each other based on their

individual abilities (Ferber et al., 2004). However,

the system designer may not have foreseen all

possible environments in which the system can be

deployed. Unpredictable application environments

make multi-agent systems vulnerable to individual

failures that can dramatically reduce the system's

ability to accomplish its objectives. The agents

themselves may exhibit particular properties that

were not originally intended, and some agents may

then become unable to fulfill the roles assigned to

them. For example, dangers or even malicious

actions could sever the communication links

between agents. Alternatively, agents cannot fulfil

their roles due, for example, to: insufficient

capacities of agents to accomplish their roles or

agent overload resulting in failure to meet some of

its objectives.

As multi-agent systems grow, the configuration

and tuning of these systems can become as complex

as the problems they claim to solve. Thus, a robust

organizational multi-agent system must be maintained

in order to adapt to environments, recover from

degradation in its performance (quality) and improve

over time to avoid undesirable situations.

Therefore, the central question addressed in this

article can be formulated as follows: «How to ensure

the continuity of the MAS’ activities so that the

degradation of the performance (quality) of the

agents in playing their roles does not drastically

affect the functioning of the MAS? ». To address

these concerns, we propose, in this paper, a

preventive maintenance approach to enable

organizational MAS to safely reach their goals.

Preventive software maintenance is an important

software activity that involves including changes

and updates to prevent serious software problems in

the future (ISO/IEC/ IEEE 24765, 2010). In the

context of multi-agent systems, such activity is

completely omitted. The inherent specificities to

multi-agent systems (e.g. autonomy, pro-activity,

reactivity, adaptability, etc.) make their maintenance

difficult to achieve.

The proposed maintenance approach concerns, in

particular organizational MAS. Wooldridge et al.

(Wooldridge et al., 2000) view an organization as a

collection of roles that stand in certain relationships

384

Ghrieb, N., Mokhati, F. and Guerram, T.

Maintaining Organizational Multi-agent Systems: A Reorganization-based Preventive Approach.

DOI: 10.5220/0010314803840389

In Proceedings of the 13th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2021) - Volume 1, pages 384-389

ISBN: 978-989-758-484-8

to one another, and that take part in systematic

institutionalized patterns of interactions with other

roles. This type of MAS achieves its goals by

assigning agents to different roles according to their

individual abilities (Ferber et al., 2004). We then

admit, in the context of this work, that an

organization works properly and can achieve its

objectives if its agents perform their roles in an

efficient manner and that the decrease in their

efficiency entails a disturbance which leads to a

degradation of the efficiency of the entire system.

Therefore, a solution where multi-agent systems are

enhancing with the ability to reassign

responsibilities from defective agents to others

providing similar capabilities is needed. This

requires first collecting information on the ability or

inability of a given agent to assume its roles. This

information can then form the basis of a decision as

to whether other agents should be sought for the

accomplishment of a given objective.

Based on these ascertainment, we propose a

conditional preventive maintenance approach by

monitoring certain quality criteria corresponding to

the running MAS in order to predict any failure that

may appear. The idea of our approach therefore

consists in keeping the system’s quality level and

that of the agents beyond certain thresholds defined

by the designer. When the system quality

deteriorates below the defined thresholds, we

reorganize the system in order to restore its quality

and allow it to resume its normal functioning.

The remainder of this paper is organized as

follows. In Section 2, we give a brief overview of

major related work. We present in Section 3 the

proposed approach. The architecture of the proposed

system is given in section 4. We discuss in section 5

the advantages and limitations of our approach.

Section 6 gives some conclusions and future work

directions

2 RELATED WORK

Very few approaches have been proposed in the

literature to deal with the problems which are related

to preventive maintenance of software (Garget et al.

1998, Vaidyanathan et al, 2002, Singh and

BinduGoel, 2007, Cheluvaraju et al, 2012, Sun and

Wang, 2012).

In order to solve the problem of preventive

maintenance of software systems for transactions, an

analytical model has been proposed by Garget and

al. This model takes into account: the availability of

the software to provide a service, the probability of

losing a transaction and the response time of a

transaction (Garget et al. 1998).

Through a Markov regeneration process with a

subordinate semi-Markov reward process,

Vaidyanathan et al. proposed an analytical model of

a software system using preventive maintenance

based on inspection (Vaidyanathan et al, 2002).

Singh and BinduGoel first attempted to analyse

the issues governing software maintenance and how

preventive maintenance can improve the lifespan

(aging) of the software product. These authors then

integrated a model for preventive maintenance into

the software lifecycle (Singh and BinduGoel, 2007).

Cheluvaraju et al. proposed a software quality

metric called "prevention metric" to measure the

avoidance of defects in software. This metric derives

from a quantitative assessment of both the efficiency

and effectiveness of individual prevention

techniques that are employed on the software prior

to its deployment. It helps provide confidence in

how defect prevention is handled prior to

deployment (Cheluvaraju et al, 2012).

A preventive software maintenance policy based

on the ant colony algorithm has been studied by Sun

and Wang. The system as a whole has been divided

into several subsystems each of which has four types

of maintenance policy with different maintenance

costs. This type of model makes it possible to obtain

an optimal preventive maintenance policy for each

sub-system, thus guaranteeing excellent reliability of

the software system with relatively lower costs (Sun

and Wang, 2012).

Ghrieb et al. (Ghrieb et al., 2020) have proposed

a conditional preventive maintenance approach for

multi-agent applications that is based on MAS

quality measurements and uses aspect-oriented

programming. This proposed approach includes

three major steps: (i) measurements of two quality

metrics (autonomy and sociability) of the running

application in a dynamic and continuous manner

using the AspectJ code and comparing them to the

minimum thresholds previously defined by the

designer, (ii) warning the maintainer in case of

detection of abnormal regression in the MAS

quality, and (iii) intervention by the maintainer to

preserve the quality of the application and thus avoid

potential damage.

As mentioned above, all of these approaches

relate to preventive software maintenance. However,

excepting the approach we proposed in (Ghrieb et

al., 2020), none of this work deals with preventive

maintenance of MAS. Our proposed approach was a

first step towards proposing a generic preventive

maintenance approach for multi-agent applications.

Maintaining Organizational Multi-agent Systems: A Reorganization-based Preventive Approach

385

In this article, we present a new and original

approach to conditional preventive maintenance for

Organizational MAS. The proposed approach uses,

on the one hand, aspect-oriented programming to

monitor the quality of the running MAS and, on the

other hand applies reorganization techniques to

allow the system to resume its efficiency in case of

problems.

3 THE PROPOSED APPROACH

The conditional preventive maintenance approach

we propose is based on MAS quality measurements

and uses aspect-oriented programming (Figure 1).

Our approach is used to continuously measure

the quality metrics of the application and the agents

composing it using the control code written in

AspectJ and compare them with thresholds

previously defined by the designer. When the

measured values are lower of the specified

thresholds, preventive intervention must be carried

out in order to restore the quality of the agents to

normal values above the previously defined

thresholds. This intervention is carried out by

reorganization techniques allowing the reallocation

of agents to roles.

4 SYSTEM ARCHITECTURE

The overall architecture of the system implementing

our approach is illustrated by figure 2. It is

composed of two parts: our Maintenance System

baptized Main-OCMAS (MAINtenance of

Organization-Centered Multi-Agent Systems) and

the Organization-Centered Multi-agent System

(OCMAS), the running system which is composed

of the Organizational Master (OM) and the agents of

the application.

4.1 OCMAS System

The OCMAS system is designed according to a

centralized approach (Figure 3), which makes it

possible decoupling the reasoning part of the

organization from the real system composed of the

agents of the application. The Organization Master

(OM) is the only agent with complete knowledge on

the organization that is able to execute

reorganization algorithms. The OM uses their

knowledge on goals and current agents, makes the

appropriate assignments, and sends the assignments

to agents. The OM also receives events about the

execution of each agent's objectives and reorganizes

itself appropriately when needed. In this centralized

approach we adopt, application agents are

independent of any organizational model. They

receive assignments from OM, play the roles

assigned to them and report on their status to OM.

Communication between the MO and the agents is

done by sending messages.

4.2 Maintenance System

MAIN-OCMAS

The MAIN-OCMAS maintenance system is an

extension added to OCMAS system to provide

condition-based preventive maintenance of OCMAS

systems. This maintenance system continuously

evaluates the OCMAS system in order to remedy

any problems that may affect its organization and

prevent it from achieving its goal. This system is

composed of a Monitoring System (MS) and a

Maintenance Agent (MA).

4.2.1 Monitoring System

The monitor runs as a third party and consists of

gathering information using the Aspect Engine. The

latter intercepts the relevant events of the agent

platform: Communication monitoring, Message

monitoring, Agent action monitoring in order to

report all events requiring intervention to the

Organizational Master (OM).

As our objective concerns the preventive

maintenance of organizational multi-agent systems,

so we are interested, in this work, in particular in the

events leading to the regression of the efficiency of

the system. These events are monitored by the

monitoring system and will be used by the

maintenance agent for the reorganization of the

system in order to restore its efficiency value to the

adequate level and thus prevent the system from a

possible failure.

4.2.2 Maintenance Agent

The maintenance agent, in turn, is composed of an

Evaluation Component (EC), a Disturbance

Detection and Decision Component (DDDC) and a

Reassignment Component (RC). The Monitoring

system is responsible for retrieving the relevant

information from the running MAS, the evaluation

component assesses the effectiveness of the

organization and that of the agents, the disturbance

detection and decision component uses these

ICAART 2021 - 13th International Conference on Agents and Artiﬁcial Intelligence

386

Figure 1: The methodology of the proposed approach.

Figure 2: Overall system architecture.

Maintaining Organizational Multi-agent Systems: A Reorganization-based Preventive Approach

387

Figure 3: Centralized architecture of the OCMAS system.

assessments to identify situations in which: the

quality requirements are not being met, the agents

are not able to play their role well, and the current

organization no longer meets the needs of the

running MAS. The reassignment component

reallocates agents to replace agents that do not

perform their roles properly with more efficient

agents. Finally, the organization master uses the

reassignment component for the reorganization of

the system, in order to improve the efficiency of the

OCMAS system and allow it to reach its goal safely.

The maintenance agent (figure 4) in our MAIN-

OCMAS system is designed to ensure preventive

maintenance of the OCMAS system; it accomplishes

three main activities:

• Perception.

The maintenance agent watches for system

disruptions from the execution trace established by

the monitoring system. This component is constantly

listening and allows the agent to analyze the result of

the execution of roles by the agents and to keep

track of the encountered disruptions.

• Evaluation and Decision.

The agent assesses these disruptions and decides

when to start the reorganization process. The

decision depends on the overall rate of disturbances,

when the degree of efficiency in the system drops

below a certain threshold, a global adaptation of it

become necessary.

• Execution.

At this stage, the agent applies a reorganization

algorithm which acts on the assignments of agents to

different roles, in order to improve, at each step of

the algorithm, the assignment of agents to roles.

Figure 4: Model of the Maintenance Agent.

5 DISCUSSION

As noted above, some research works have been

done on preventive software maintenance. These

works have provided interesting solutions to

different problems in different contexts. However,

none of them deal with preventive maintenance of

MAS. The approach we have already proposed in

(Ghrieb et al., 2020), was a first step towards

proposing a generic preventive maintenance

approach for multi-agent applications. However, the

limitation of this approach is its semi-automatic

character. The approach presented in this article

addresses this problem and allows automatic tuning

of certain problems that may arise in organizational

MAS. This approach is a new proposal that takes

into account other quality attributes of multi-agent

systems and allows preventive maintenance for

organizational multi-agent applications. The

maintenance system we have proposed in this article

adopts a reorganization strategy that takes into

account the drop in agent capacities and performs

agent reassignment by assigning roles to the best

ICAART 2021 - 13th International Conference on Agents and Artiﬁcial Intelligence

388

agents. This makes it possible, on the one hand, to

improve the efficiency of the system in case of

degradation of agent capacities and, on the other

hand, to protect the system from harmful

consequences in case of intrusion of non-competent

agents into the running OCMAS. Furthermore, our

monitoring system monitors the efficiency of the

OCMAS system, in order to trigger a preventive

intervention for preserving the system from an

excessive drop in performance that could lead to its

failure.

6 CONCLUSION AND FUTURE

WORK

Multi-agent systems can become quite complex and

can be deployed in open and unpredictable

environments. The system designer may not have

the resources to manually tune the system for a

particular deployment. With these issues in mind, we

propose a conditional preventive maintenance

approach to create a self-tuning multi-agent system.

The proposed approach is based on evaluating the

assurance that OCMAS system agents are

performing well to fulfill their corresponding roles.

These performance metrics are used to trigger the

automatic reassignment of roles to agents that is

more effective, for example, in case of a danger

which can lead to agent isolation or inability to

achieve a goal. The reorganization currently applied

by our MAIN-OCMAS maintenance system is

achieved by a strategy of reassigning agents to roles

in order to remedy the problems of the current

organization. However, the architecture of our

maintenance system is extensible; it allows adding

other reorganization strategies in the maintenance

agent in order to deal with other organizational

problems. As future work, we plan to develop a

software tool supporting our approach and allowing

users to specializing it by offering them the

possibility of choosing their organizational model.

REFERENCES

Ferber, J., Gutknecht, O., & Michel, F., (2004). From

agents to organizations: an organizational view of

multi-agent systems. In Agent-Oriented Software

Engineering IV, Springer Berlin Heidelberg, pages

214–230.

ISO/IEC/ IEEE 24765, (2010), Systems and software

engineering – Vocabulary.

Wooldridge, M., Jennings, N. R., David, K., (2000). The

Gaia methodology for agent-oriented analysis and

design. Journal of Autonomous Agents and Multi-

Agent Systems 3: 285-312.

Garg, S., Puliafito, A., Telek, M. and Trivedi, K. S.,

(1998). Analysis of preventive maintenance in

transactions based software systems. IEEE Trans.

Computers 47(1): 96–107.

Vaidyanathan, K., Dharmaraja, S. and Trivedi,K.S.,

(2002). Analysis of Inspection-Based Preventive

Maintenance in Operational Software Systems. In

Proceedings of 21st IEEE Symposium on Reliable

Distributed Systems, pages 286–295.

Singh, Y. and Goel, B., (2007). A step towards software

preventive maintenance. In Proceedings of ACM

SIGSOFT Software Engineering Notes 32(4):

Cheluvaraju, B., Pasala, A., Padmanabhuni,S., and

Chevireddy,S.,(2012). A quantitative measure for

preventive maintenance in software. In ACM

SIGSOFT Software Engineering, Notes 37(4), pages

1–5.

Sun, P. and Wang, X., (2012). Application of ant colony

optimization in preventive software maintenance

policy. In Proceedings of IEEE International

Conference on Information Science and Technology,

pages 141–144.

Ghrieb, N., Mokhati, F., Ghorab, M. A., and Guerram, T.,

(2020). Towards a preventive maintenance approach

for multi-agent applications. In Multi-agent and Grid

Systems – An International Journal 16, pages 83–99.

Maintaining Organizational Multi-agent Systems: A Reorganization-based Preventive Approach

389