On the Usage of Network Visualization for Multiagent System Verification

Abstract

Multiagent Systems (MAS) consists of many software agents that interact to each other to perform their actions and achieve system goals. Due to the growing demand of Distributed Software Systems (DSS) and MAS as a branch of DSS, the verification of these systems has taken a special attention. The verification of these systems is required because MAS and DSS are large scale systems, where a failure can result in a huge amount of cost or damage. One major field is verification of MAS designs to prevent cost of fixing problems after implementation and deployment. The agents and interactions among them form a network of agents. This network can be used for verification of MAS from different perspectives and by various techniques. Visualizing agents’ networks can lead to detect a special type of unexpected behavior in MAS referred to as emergent behaviors and implied scenarios. These unexpected behaviors are more probable in large scale systems because the functionality and control are distributed and there is lack of central controller in MAS and DSS. Consequently, a new scenario can be implied to the system at run time which may not be an acceptable behavior in the system. In this article, visualization techniques are applied to form three different networks extracted from the designs of MAS for this purpose. These networks are used to detect emergent behaviors and implied scenarios in the system. The methodology, system architecture, data preparation and visualization, required network definitions, and results are verified through case studies.