An Agent-Based Self-Adaptation Framework For Open Systems

Open systems which are compose of dynamic agents have been widely used as a software architecture, that makes it quite important to maintain the runtime qualities of the systems. By the very nature of open system, its complexity, heterogeneity of its agents and the open dynamic runtime environment make it not easy to satisfy the requirements of the open systems. So, open systems need a self-adaptation mechanism to help the agents of the systems to be self-awareness and self- optimizing, which could keep the requirements of the agents satisfied in the runtime environment.Traditional runtime self-adaptation methods usually use a control loop called Monitoring-Analyzing-Planning-Executing realize the self-computing of the software. Since this method only model the software itself, it could not describe the interactions between the agents of open systems. Tropos and i* modeling framework which extend the requirement-based modeling technologies, provide a way to model the interactions between agents of the open systems. Based on these modeling frameworks, several researches have been conducted to provide self-adaptation mechanism for agents in the open systems. However, these works have a hypothesis that the interactions between agents are static. That makes them not very useful in the runtime environment for the dynamic interactions between agents.This paper proposes an agent-based framework for the self-adaptation of the open system. Tropos modeling framework is used to describe the requirements of the agents and the interactions between agents. Considering the evolving commitments of the interaction, this paper extends the self-adaptation strategies for the agents. From the view of cost-benefit, we propose a way to estimate the value of the candidate adaptation strategies considering the risk of failure caused by the environment. And, negotiation algorithms are presented to help the creditor and debtor to reach an new agreement. Based on this self-adaptation way, we provide a implementation of our framework based on a multi-agent framework JADE. This implementation could help distributed open systems to be self-adaptive in a non-intrusive manner while providing customized self-adaption actions.To evaluate the effectiveness of our self-adaptation method and the usability of our agent-based self-adaptation framework, we conducted two experiments under different scenarios and systems to verify it. Results of the experiments confirm that our method and framework implementation could be effective in realizing the self-adaptation of open systems and could be used in a wild range of open systems...