Strategic Structural Reorganization in Multi-agent Systems Inspired by Social Organization Theory

Abstract

Autonomic systems, capable of adaptive behavior, are envisioned as a solution for maintaining large, complex, real-time computing systems that are situated in dynamic and open environments. These systems are subject to uncertainties in their perceptual, computational, and communication loads. As a result, the individual system components find the need to cooperate with each other to acquire more information and accomplish complex tasks. Critical to the effective performance of these systems, is the effectiveness of communication and coordination methods. In many practical applications of distributed and multi-agent systems, the problem of communication and coordination becomes even more complicated because of the geographic disparity of tasks and/or agents that are performing the tasks. Experience with even small systems has shown that lack of an effective communication and coordination strategy leads the system to no-answer, or sub-optimal answer situations. To address this problem, many large-scale systems employ an additional layer of structuring, known as organizational structure, which governs assignment of roles to individual agents, existence of relations between the agents , and any authority structures in between. Applying different organizational structures to the same problem will lead to different performance characteristics. As the system and environment conditions change, it becomes important to reorganize to a more effective organization. Due to the costs associated with reorganization, finding a balance in how often or when a reorganization is performed becomes necessary. In multi-agent systems community, not a lot of attention has been paid to reorganizing a system to a different organizational structure. Most systems reorganize within the same structure, for example reorganizing in a hierarchy by changing the width or depth of the hierarchy. To approach this problem, we looked into adaptation of concepts and theories from social organization theory. In particular, we got insights from Schwaninger's model of Intelligent Human Organizations. We introduced a strategic reorganization model which enables the system to reorganize to a different type of organizational structure at run time. The proposed model employs different levels of organizational control for making organizational change decisions. We study the performance trade-offs and the efficacy of the proposed approach by running experiments using two instances of cooperative distributed problem solving applications. The results indicate that the proposed reorganization model results in performance improvements when task complexity increases.