Distributed Task Allocation in Multi-Agent System Based on Decision Support Module

Description

The International Journal of Computer Science and Information Security (IJCSIS) is a well-established publication venue on novel research in computer science and information security. The year 2010 has been very eventful and encouraging for all IJCSIS authors/researchers and IJCSIS technical committee, as we see more and more interest in IJCSIS research publications. IJCSIS is now empowered by over thousands of academics, researchers, authors/reviewers/students and research organizations. Reaching this milestone would not have been possible without the support, feedback, and continuous engagement of our authors and reviewers. Field coverage includes: security infrastructures, network security: Internet security, content protection, cryptography, steganography and formal methods in information security; multimedia systems, software, information systems, intelligent systems, web services, data mining, wireless communication, networking and technologies, innovation technology and management. ( See monthly Call for Papers) We are grateful to our reviewers for providing valuable comments. IJCSIS December 2010 issue (Vol. 8, No. 9) has paper acceptance rate of nearly 35%. We wish everyone a successful scientific research year on 2011. Available at http://sites.google.com/site/ijcsis/ IJCSIS Vol. 8, No. 9, December 2010 Edition ISSN 1947-5500 � IJCSIS, USA.

Shared by: ijcsiseditor

Stats

views:: 277
posted:: 1/19/2011
language:: English
pages:: 6

Public Domain

Document Sample

(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 9, 2010

Distributed Task Allocation in Multi-Agent System
Based on Decision Support Module
Sally M. El-Ghamrawy Ali I. El-Desouky Ahmed I. Saleh
Computers and Systems Computers and Systems Computers and Systems
Department, Department, Department,
Faculty of Engineering, Faculty of Engineering, Faculty of Engineering,
Mansoura University, Egypt Mansoura University, Egypt Mansoura University, Egypt
Sally@mans.edu.eg ali_eldesouky@yahoo.com aisaleh@mans.edu.eg

Abstract-. A Multi-Agent System (MAS) is a branch of that are beyond their individual capabilities. There are some
distributed artificial intelligence, composed of a number of modules must be considered in designing the Distributed
distributed and autonomous agents. In MAS, an effective Multi-Agent Intelligent System (DMAIS) that help in
coordination is essential for autonomous agents to reach their developing in the agent systems. There are many researches
goals. Any decision based on a foundation of knowledge and
considered some modules when designing multi-agent
reasoning can lead agents into successful cooperation, so to
achieve the necessary degree of flexibility in coordination, an systems (MAS) in different ways. Each autonomous agent in
agent requires making decisions about when to coordinate and DMAIS must be able to decide how to behave in various
which coordination mechanism to use. The performance of any situations, so in this paper our main concern is to propose an
MAS depends directly with the right decisions that the agents efficient decision support module that helps in the
made. Therefore the agents must have the ability of making improvement of DMAIS performance.
right decisions. In this paper, we propose a decision support Agents have attractive characteristics like: autonomy,
module in a distributed multi-agent system, which enables any reactivity, reasoning capability and social ability. These
agent to make decisions needed for Task allocation problem; we characteristics ensure that agent-based technologies are
propose an algorithm for Task Allocation Decision Maker
responsible for enhancing the decision support system
(TADM). Furthermore, a number of experiments were
performed to validate the effectiveness of the proposed capabilities beyond the capabilities of the old model. Any
algorithm (TADM)); we compare the efficiency of our active decision support can be facilitated by the autonomy,
algorithms with recent frameworks. The preliminary results reactivity and social ability of agents. Furthermore, the
demonstrate the efficiency of our algorithms artificial view of agents can contribute towards stronger
collaborative relationships between a human and a decision
Keywords: .Decision Making, Task allocation, Coordination Mechanism, support system. These enormous interests of researchers in
Multi-Agent System (MAS)
investigating the decision support in Multi-Agent Systems is
I. INTRODUCTION due to the great benefits of combining the decision support
technology and Agent-based technology with taking the
The notion of distributed intelligent systems (DIS) [1] has
advantage of the agent characteristics. In this sense, many
been a subject of interest for number of years. A Multi-Agent
researches [3-7] recognized the promise that agent -based
system (MAS) is one of the main areas in the DIS. Any technologies holds for enhancing DSS capabilities.
Multi-agent system consists of several agents capable of Decision support module is a vital module in the success of
mutual interaction, with heterogeneous capabilities, that DMAIS framework, due to the fact that any decision based
cooperate with each others to pursue some set of goals, or to on a foundation of knowledge and reasoning can lead agents
complete a specific task. MAS used to solve problems which into successful cooperation. So the performance of any multi-
are difficult or impossible for an individual agent or agent system depends directly with the right decisions that
monolithic system to solve. Agents are autonomous the agents made. Obviously, complex decision making tasks
programs which can understand an environment, take actions cannot be achieved by a single agent. Rather, it’s achieved by
depending upon the current status of the environment using efforts coordination of multiple agents possess different sets
its knowledge base and also learn so as to act in the future. In of expertise, attributes and assignment. This coordination
order to solve complex problems agents have to cooperate among agents, which provide satisfactory solutions to
and exhibit some level of autonomy. Agents cooperate with problems among agents, needs many decisions that agents are
each other to solve large and complex collaborative required to make before this coordination can take place. So
problems. Because the majority of work is completed through the Decision support module role is to allow agents to make
distributing the tasks among the cooperative agents, the the decision needed, which can help in the improvement of
decision support module is responsible on the most of the the DMAIS framework. In this paper, we extended the work
work. So the pros and cons in decision support module done in DMAIS [2] by proposing a Decision support module
in it, this module is concerned about taking the right
directly affect on the success of tasks completion and
decisions needed to allocate a specific task to a specific agent,
problem solving.
by using the Task Allocation Decision Maker sub-module
In [2], we proposed a Distributed Multi-Agent Intelligent
(TADM). So we spotted on these decisions due to the great
System (DMAIS,) a general purpose agent framework, in
importance for them in the improvement of agents’
which several interacting intelligent agents cooperate with
coordination in DMAIS framework. The rest of the paper is
some auxiliary agents to pursue some set of goals or tasks
organized as following. Section 2 demonstrates the proposed

210 http://sites.google.com/site/ijcsis/
ISSN 1947-5500
(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 9, 2010
decision support module. Section 3 discusses the related work The Decision support module is activated when a decision is
on task allocation in Multi-agent Systems. An algorithm for needed from the agent, we concerned about two main
the Task Allocation Decision Maker (TADM) is proposed in decisions that the agent may face: (1) the decision needed to
section 4, showing the scenario of our algorithm in allocating allocate a specific task to a specific agent, making an
tasks to specific agent/s. Section 5 shows the experimental effective decision for the task allocation problem is a critical
evaluation and the results obtained after implementing the job for any multi-agent system; it helps our DMAIS
proposed algorithm. Section 6 summarizes major contribution framework to complete its tasks and missions through
of the paper and proposes the topics for future research. cooperation among agents. The Task Allocation Decision
Maker sub-module (TADM) is responsible for making this
decision. (2) The decision needed to select appropriate
II. DECISION SUPPORT MODULE
coordination mechanism, when agents need to coordinate
In the DMAIS framework [2], the decision support module with another agent/s to accomplish a specific task.
takes the information collected in the coordination and The decision support module contains four main phases, as
negotiation module, as shown in fig 1, so that it can help in shown in figure 2, first is the decision knowledge
the decision support process. management that contains:
 The database: contains the data that directly related
Management Module

Reasoning Module
Coordination Module to the decision problem (i.e. the performance measures, the
values of nature states).
Planning Module
Negotiation Module  The knowledge base: contains the descriptions for
Scheduling Module roles and structures of document and some knowledge of
Decision Support the problem itself, (i.e. Guide for how to select decision
Module Execution Module alternatives or how to interpret outputs).
 The knowledge modeling: is a repository contains
the decision problem formal models and the algorithms and
Communication Module methodologies for developing outcomes from the formal
Figure 1. The Decision Support Module in DMAIS Framework models. Also contains different process models, each
model is represented as a set of process and event objects.

Decision knowledge Management Knowledge Knowledge
Database
Base Modeling

Agent State Evaluator Task State Evaluator Organizer
Attributes estimator Demands estimator
Data
Capabilities estimator Urgency degree estimator
Behaviors estimator Importance degree estimator
Dependency estimator Relevant
Relevant
data from
data from
Tasks
agents

Data Interpreter
Allocation
Decision

Decision making method identifier
Maker

Task

Building the representation model

Calculating the expected utilities of all possible alternative solutions

Proper Action Selection
C . M Selection
Decision Maker

C.M list
Bonus Estimator

Cost Estimator
Probability of success estimator
Calculate Maximum Switch procedure
values

Action Taker Module
Result Evaluator
Evaluator

Figure 2. Architecture of the Decision Support Module

211 http://sites.google.com/site/ijcsis/
ISSN 1947-5500
(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 9, 2010
Second phase is the data organizer, which organizes Tasks allocation is defined, in [8], as the ability of agents to
agent’s attributes using the agent state evaluator and self-organize in groups of agents in order to perform one or
organizes task’s preferences by using task state evaluator The more tasks which are impossible to perform individually. In
agent state evaluator estimates the attributes and capabilities our context, it is a problem of assigning responsibility and
of agents, in our work we concerned on specific attributes problem solving resources to an agent. There are two main
and characteristics of the agent that will help in the decision benefits for Minimizing task interdependencies in the
making procedure, as shown in fig 3, where: coordination process between the agents: First, improving the
problem solving efficiency by decreasing communication
overhead among the agents. Second, improving the chances
for solution consistency by minimizing potential conflicts.
The issue of task allocation was one of the earliest problems
to be worked on in Distributed Artificial Intelligence (DAI)
research. In this sense, several authors studied the problem
related to Task Allocation especially in MAS. The researches
in task allocation can be classified in to two main parts,
centralized and distributed, based on utility/cost functions.
The researches that investigate the task allocation
problem in a centralized manner: Zheng and Koenig [9]
Figure 3. Attributes, Capabilities and behaviors of Agents presented reaction functions for task allocation to cooperative
agents. The objective is to find a solution with a small team
Agent attributes is defined as a tuple cost and each target to be assigned to the exact number of
<AgentID(A), Addrs(A)> (1) different agents. This work assumed that there is a central
Where AgentID(A) is the identity of the agent, Addrs(A) planner to allocate tasks to agents. Kraus et al. [10] proposed
records the IP address of agent A. Agent Capability is defined an auction based protocol which enables agents to form
as a tuple coalitions with time constrains. This protocol assumed each
< Availability(A), Ability(A), Intensity(A)> (2) agent knows the capabilities of all others, and one manager is
Where Availability(A) is the ratio of total number of responsible for allocating tasks to all coalitions. Pinedo [11]
successful agents, capable of accomplishing the task, to the proposed a job shop scheduling treats the task allocation
total number of agents in the system, mostly in a centralized manner, and also ignores the
AVLB(T)= SUC(A) / TOT (A) (3) communication cost. There are many drawbacks in the
centralized task allocation like single point failure and bad
Ability(A) indicates number of agents capable of fulfilling
scalability. To conquer these disadvantages, Task allocation
the task. Intensity(A) refers to the number of tasks that the
agents can accomplish per time unit. in distributed environments has also been investigated. Davis
Agent Behaviors is defined as a tuple and Smith [12] was the first in investigating a classic
< Task History(A), Active degree(A), Dependency(A)> (4) distributed task allocation in the multi–agent system using
Contract Net Protocol (CNP), in which agents negotiate to
Where task history(A) stores the number of accomplished
assign tasks among themselves. Most of the subsequent
tasks that the agent(A) performed. Active degree(A) indicates
literature on distributed task allocation is based on either
the activity degree of a specific agent, this degree varies from
contract net protocol or auctions [13]. The authors in [14]
agent to another according to many parameters (e.g. number
and [15] developed distributed algorithms with low
of finished tasks, number of cooperation process). Agent
communication complexity for forming coalitions in large-
Dependency(A) is concerned with the Cooperation Degree of
scale multi-agent systems. Abdallah and Lesser [16] provided
an Agent(A) with respect to other agents in the system.
a decision theoretic model in order to limit the interactions
Third phase is divided into two main Sub-modules: (1)The
between agents and mediators. Mediators in this research
Task Allocation Decision Maker (TADM) ,this sub-module
mean the agents which receive the task and have connections
will be discussed in details in the next sections, and (2) The to other agents. Mediators have to decompose the task into
Coordination Mechanism Selection Decision Maker subtasks and negotiate with other agents to obtain
(CMSDM), that takes the decision needed to select commitments to execute these subtasks. However, their work
appropriate coordination mechanism, when agents need to concentrated on modelling the decision process of a single
coordinate with another agent/s to accomplish a specific task, mediator. Sander et al. [17] presented a scalable and
it will be discussed in our future work. Both of them are distributed task allocation protocol. The algorithm adopted in
responsible on making the proper decisions according to the this protocol is based on computation geometry techniques
input obtained from the Data organizer phase. The module but the prerequisite of this approach is that agents’ and tasks’
evaluator phase is the fourth phase, it has two main goals: geographical positions are known. Weerdt et al. [18]
first, take the action that the third phase decided to be taken. proposed a distributed task allocation protocol in social
Second, evaluate the results of taking this action. networks. This protocol only allows neighbouring agents to
help with a task which might result in high probability of
III. RELATED WORK abandon of tasks when neighbours cannot offer sufficient
resources. Dayong et al. [19] proposed an Efficient Task
The task allocation in Multi-Agent Systems represent a Allocation Protocol (ETAP) protocol based on the Contract
problem which occupied to a large extends the researchers of Net approach, but more suitable for dealing with task
decision support and artificial intelligence until our days. allocation problems in P2P multi-agent systems with a

212 http://sites.google.com/site/ijcsis/
ISSN 1947-5500
(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 9, 2010
decentralized manner. It enables agents to allocate tasks not used the rough set theory for classifying these metrics to
only to their neighbours but also to commit unfinished tasks obtain the decision of allocating tasks to agent/s. Then this
to their neighbours for reallocation. In this way, the agents decision is tested, it may be one of three decisions: first, if the
can have more opportunities to achieve solution of their tasks allocation cannot be done, in this situation the capabilities of
Brahmi et al. [20] developed a decentralized and scalable agents and demands of tasks must be defined all over again, it
method for complex task allocation for Massive Multi-Agent may encounter any kind of error. second, if the decision is to
System, distributing the process of computing the optimal allocate the task to only one agent, then task priority must be
allocation among all agents based on the hypothesis: non reviewed to check the task queue for each agent, and depend
conflict will be generated in the task allocation processes. on this queue an action must be taken whether to delay, reject
Indeed, while being based on its Galois Sub-Hierarchy (GSH) or execute this task. Third, if the decision is to allocate the
and cooperation with other agents, each agent chooses the task to more than one agent, then the coordinated agents must
appropriate sub-task that ensures the global allocation registered, and the sub-tasks must be distributed among those
optimality. Cheng and Wellman [21] used a market based coordinated agents according to the algorithm in figure 5.
protocol for distributed task allocation. TADM algorithm
1. Agent Ai randomly select task T
2. AinCharge= Ai
IV. TASK ALLOCATION DECISION MAKER (TADM) 3. For each a(i) in G
4. Ai=Send request ( )
The task allocation Decision Maker (TADM) main goal is 5. Ai=Wait response( )
to take the proper decisions of allocating tasks to the right 6. If time waited > expired time then
7. Exit for
agents. The first step in allocating tasks to specific agents is 8. End if
to take the decision whether the agents are capable of 9. Ai=Receive response( )
10. Ai=Process response( )
executing part or the entire task. The allocation decisions are 11. Ai=Store response()
made independently by each agent 12. Max-value=0
13. If helpfulness-value> Max-value then
The decisions needed for the Task allocation problem are 14. Max-value = helpfulness-value
mainly concerned about allocating the tasks to number of 15. Nominated-agent= a(i)
16. End if
agents, whether these agents can complete its tasks by 17. Next
themselves or not. If agents can’t achieve the task by 18. Xx:
19. Ai=Check (Nominated-agent( ))
themselves, they attempt to give a decision to specify other 20. If helpfulness-value(Ai)< Nominated-agent then
agents which have the appropriate capabilities and assign the 21. Ai= send Response(Nominated-agent)
22. If reply (Nominated-agent) =1 then
task, or part of the task, to those agents. Fig 4 depicts the 23. AinCharge= Nominated-agent
scenario of allocating tasks to specific agent/s in the TADM. 24. Else
25. Ai= find_scnd_highst( )
26. Go to xx
27. End if
28. Else
29. AinCharge= Ai
30. End if
Figure 5: The Pseudo code for TADM algorithm

V. EXPERIMENTS EVALUATION
To evaluate the performance of TADM, we compare it
with Efficient Task Allocation Protocol ETAP [18] and with
the Greedy Distributed Allocation Protocol (GDAP) [19]. In
order to validate the effectiveness of TADM algorithm and
compare it with ETAP and GDAP two experiments are
performed, as shown in figure 5; each experiment has its own
goal and settings. In each experiment to evaluate the
experiment results two metrics are evaluated the Efficiency
Ratio and Run Time, which can be defined as follows:

Figure 4: The Proposed algorithm used in TADM

First the capabilities and behaviors of agents must be defined,
and also the demands of tasks, it’s taken from the data
organizer as an input to the TADM The metrics that facilitate
the decision making process must be evaluated, and then we
Figure 5: The experiments of Evaluating TADM algorithm

213 http://sites.google.com/site/ijcsis/
ISSN 1947-5500
(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 9, 2010
The Efficiency Ratio is the ratio between summation From figure 7, it is shown that when the number of agents
efficiency of finished tasks and the total efficiency expected is increasing, the Efficiency Ratio of TADM is much higher
of tasks. The efficiency of a task can be calculated as follows: and more stable than of ETAP and GDAP that is continually
descending with the increasing of agents.
Figure 8 shows the Run Time of TADM, GDAP and ETAP
Where: when the number of agents in the network change. It can be
is the rewards gained from successfully finishing the noticed that ETAP spends more time when there are more
agents in the network, this is because ETAP make many
task. is the resources required for accomplishing the reallocation steps which results in time and communication
task. overhead rising. On the other hand, the time consumption of
Run Time is the time of performing TADM algorithm in the GDAP is steady during the entire test process and keeps a
network under pre-defined settings. The unit of Run Time is lower level than ETAP, this because GDAP relies on
millisecond. To investigate the effects of TADM, and neighbouring agents only.
compare it with ETAP and GDAP, a multi-agent system has
been implemented to provide a testing platform. The whole
system is implemented on a 6 Pc’s with an Intel Pentium 4
processor at 300GHz, with 3GB of Ram, connected with
network Ethernet 512Mbps. A network of cooperative agents
is designed, in which most agent team are connected to each
other, the generation of this network can follow the approach
proposed in [22]. For each experiment, there are unified
settings have to be specified.
Figure 6 shows the TADM when it’s activated, and begins to
perform its roles.

Figure 8: The Run Time on different number of agents

While TADM avoid these two drawbacks from the recent
systems, and this experiment shows that the TADM
performance is faster than that of GDAP and ETAP.
In Experiment 2: The main goal of this experiment is to test
the influence of team grouping on TADM (i.e. show how
different average number of agent’s team influences the
performance of TADM) and compare it with ETAP and
GDAP, as shown in figure 9 and figure 10, using same
environment and settings, as follows:
 The average number of agent’s team is fixed at 8.
 The number of agents and tasks are 50 and 30 separately.
Figure 6: Screen shot of the TADM GUI
 The average number of resources for each type is 30 and
the average number of resources required by each task is
In Experiment 1: The main goal of this experiment is to
also 30.
demonstrate the scalability of TADM algorithm and compare
 The number of resources types is 5.
it with ETAP and GDAP, as shown in figure 7 and figure 8,
using same environment and settings, as follows:
 The average number of agent’s team is fixed at 8.
 The number of agents range from 100 to 600, depending
on the specific test.
 The number of tasks is range from 60 to 360.
 The number of resources types is range from 10 to 60.

Figure 9: The Efficient Ratio on different number of agent’s team

Figure 9 demonstrated that The Efficiency Ratio of TADM is
much higher and more stable than that of GDAP and ETAP.
GDAP performance is very low this is because task allocation
in GDAP is only depending on neighbours of the agent. On
the other hand, ETAP has better performance because it relies
Figure 7: The Efficient Ratio on different number of agents

214 http://sites.google.com/site/ijcsis/
ISSN 1947-5500
(IJCSIS) International Journal of Computer Science and Information Security,
Vol. 8, No. 9, 2010
not only on neighbours of the agent, but also other agents if [5] Song Qiang, Lingxia Liu, "The Implement of Blackboard-Based Multi-
agent Intelligent Decision Support System," iccea, vol. 1, pp.572-575,
needed.
2010 Second International Conference on Computer Engineering and
The TADM has the higher performance and more stable than Applications, 2010
GDAP and ETAP, these results ensures and validates our [6] Rustam and Bijan, “Pluralistic multi-agent decision support system: a
algorithm. framework and an empirical test, ” Information & Management, vol.
41, Sept., 2004, pp. 883-898.
[7] Wang Fuzhong, "A Decision Support System for Logistics Distribution
Network Planning Based on Multi-agent Systems," dcabes, pp.214-218,
2010 Ninth International Symposium on Distributed Computing and
Applications to Business, Engineering and Science, 2010
[8] Tosic, P.T., Agha, G.A.: Maximal Clique Based Distributed Group
Formation For Task Allocation in Large-Scale Multi-Agent Systems.
In: Ishida, T., Gasser, L., Nakashima, H. (eds.) MMAS 2005. LNCS
(LNAI), vol. 3446, pp. 104–120. Springer, Heidelberg (2005).
[9] X. Zheng and S. Koenig. Reaction functions for task allocation to
cooperative agents. In Proceedings of 7th International Conference on
Autonomous Agents and Multiagent Systems (AAMAS 2008), pages
559–566, Estoril, Portugal, May 2008
[10] S. Kraus, O. Shehory, and G. Taase. Coalition formation with uncertain
Figure 10: The Run Time on different number of agents’ team heterogeneous information. In Proceedings of 2th International
Conference on Autonomous Agents and Multiagent Systems (AAMAS
2003), pages 1–8, Melbourne, Australia, July 2003.
The Run Time of GDAP, ETAP, TADM in different number [11] Pinedo, M., “Scheduling: Theory, Algorithms and Systems”, Prentice
of agents’ team is depicts figure 10. It’s obviously that the Hall, 1995.
Run Time of ETAP is higher than that of GDAP. As ETAP [12] Davis, R., and Smith, R., “Neogitiation as a metaphor for distributed
has to reallocate tasks when resources from neighbours are problem solving,” Artificial Intelligence, 20, 63–100, 1983.
[13] R. McAfee, and J. McMillan, “Auctions and bidding,” Journal of
insufficient this lead to increase the reallocation steps and Economic Literature, 25, 699–738, 1987.
more time spending. While GDAP is steady due to its [14] K. Lerman and O. Shehory. Coalition formation for largescale
considering for only neighbours which could decrease the electronic markets. In Proceedings of 4th International Conference on
time and communication cost during task allocation process. Multi-Agent Systems, pages 167–174, Boston, Massachusetts, USA,
July 2000. IEEE Computer Society.
But TADM takes the benefits of agent’s team and also uses [15] O. Shehory and S. Kraus. Methods for task allocation via agent
any other agent if needed, and in the same time reduces the coalition formation. Artificial Intelligence, 101(1-2):165–200, May
steps taken to allocate the task which lead to decrease the 1998
time and communication cost during task allocation process. [16] S. Abdallah and V. Lesser. Modeling task allocation using a decision
theoretic model. In Proceedings of 4th Automous Agents and
So this is why TADM has better Run Time that GDAP and Multiagent Systems (AAMAS 2005), pages 719–726, Utrecht,
ETAP. Netherlands, July 2005.
[17] P. V. Sander, D. Peleshchuk, and B. J. Grosz. A scalable, distributed
VI. CONCLUSION AND FUTURE WORK algorithm for efficient task allocation. In Proceedings of 1st
International Conference on Autonomous Agents and Multiagent
In this paper, a decision support module is proposed which Systems (AAMAS 2002), pages 1191–1198, Bologna, Italy, July 2002.
enables any agent to make decisions needed, focused on two [18] M. D. Weerdt, Y. Zhang, and T. Klos. Distributed task allocation in
main decisions: first, is the decision needed to allocate a social networks. In Proceedings of 6th Automous Agents and
specific task to a specific agent. Second, the decision needed Multiagent Systems (AAMAS 2007), pages 500–507, Honolulu,
Hawaii, USA, May 2007.
to select appropriate coordination mechanism. And a survey [19] Dayong Ye, Quan Bai, Minjie Zhang, Khin Than Win, Zhiqi Shen, "An
of recent algorithms in task allocation in Multi-agent Systems Efficient Task Allocation Protocol for P2P Multi-agent Systems," ispa,
is discussed. In addition an algorithm for the Task Allocation pp.11-18, 2009 IEEE International Symposium on Parallel and
Decision Maker (TADM) is proposed showing the scenario Distributed Processing with Applications, 2009.
[20] Zaki Zaki Brahmi, Mohamed Mohsen Gammoudi, Malek Ghenima:
of allocating tasks to specific agent/s. Finally, a preliminary Cooperative Agents Based-Decentralized and Scalable Complex Task
experiment is then conducted, indicating that TADM has the Allocation Approach Pro Massive Multi-Agents System. ACIIDS (2)
scalability advantage comparing to most recent systems. We 2010: 420-430
plan to propose new algorithm for the coordination [21] 4.Cheng, J., andWellman, M. , “TheWALRAS algorithm: A convergent
distributed implementation of general equilibrium outcomes,”
mechanism selection in the decision support module. Also, Computational Economics, 12, 1–24 ,1998.
we intend to propose a coordination module in our DMAIS [22] D. J. Watts and S. H. Strogatz. Collective dynamics of small world’
framework. networks. Nature, 393:440–442, 1998

REFERENCES
[1] Gruver, W., “Technologies and Applications of Distributed Intelligent
Systems”, IEEE MTT-Chapter Presentation, Waterloo, Canada, 2004.
[2] A. El-Desouky and S. El-Ghamrawy. A Framework for Distributed
Decision Support Multi-Agent Intelligent System. The 2008
International Conference on Artificial Intelligence ICAI08,
WORLDCOMP’08 July14-17, 2008, Las Vegas, Nevada, USA.
[3] T. Bui and J. Lee, “An agent-based framework for building decision
support systems , ” Decision Support Systems , vol. 25 , 1999 , pp.
225–237.
[4] S.D. Pinson, J.A. Louca and P. Moraitis, “A distributed decision
support system for strategic planning,” Decision Support Systems , vol.
20 1, 1997, pp. 35–51.

215 http://sites.google.com/site/ijcsis/
ISSN 1947-5500