Improved Performance of Agent Based Placement Cell System - A Performance Efficient Role Clustering Technique SOUMYA SURAVITA 1 , PRABHAT RANJAN 1 , R.K. SINGH 2 & A. K. MISRA 1 1 Department of Computer Science & Engineering, 2 Electrical Engineering Departments, Motilal Nehru National Institute of Technology, Allahabad -211004, Uttar Pradesh, INDIA Abstract: - This paper provides three criterias of clustering within a placement cell multi-agent system. Our goal is to group roles with similar objectives. Optimize the system performance by minimizing the overall interaction, data transmission and competition of shared resource between roles/agents. This paper presents a novel systematic approach to optimize the system performance by exploiting the relationships and dependencies among roles as well as clustering of roles and mapping criteria between roles to agents. The proposed clustering algorithm partition the overall system roles/agents into several clusters. Optimal cluster size can be obtained by user-defined performance parameter (η). The performance of the agent-based system, enhanced with our algorithm are investigated via implementation of placement cell case study. The results indicate that our proposed algorithm enhances the system performance if intersections of the cluster are minimal. Keywords: - Role, Agents, System Performance, Role Dependencies, Lifespan, Clustering and Mapping of Role. 1 Introduction The roles define expected behaviors of the agents and are an important concept used for different purposes like modeling of structure of multi-agent system, modeling of protocols and components of agent design [11,12]. The concept of role has been advocated to model application domain agents, which evolve dynamically during their lifespan. Understanding the relationship among roles can help the system analyzer to refine and optimize the role model [2]. Moreover the implicit conflict among roles can also be identified. Agent performs operations independently but communication between the agent and their access to shared data establishes causal dependencies in the operation [11,12]. Dependency of roles from their base agent or role is the main characteristics of roles in the Cassiopeia methodology [1]. Identifying roles and mapping the role to an agent are essential phases in many proposed AOSE methodologies like GAIA [8], PROMETHEUS [6], ROADMAP [13] and TROPOS [4]. A role can be defined as an entity consisting of a set of required permissions, a set of granted permissions, a directed graph of service invocations, and a state visible to the runtime environment but not to other agents [7]. A role can also be defined as an abstraction of the behavior of an object, which consists of a subset of the interactions of that object together with a set of constraints on which they may occur [5]. The agent based system communication can be divided into two broad categories: Tree based and Cluster based. From Figure 1, it is clear that cluster based communication is more efficient than the tree based communication. Clustering supports all communication protocols like one-to-one, one to- all, one-to-any, and many-to-one. Cluster Based Tree Based Figure 1. Type of System Communication Proceedings of the 6th WSEAS International Conference on Applied Informatics and Communications, Elounda, Greece, August 18-20, 2006 (pp492-501)