Model based approach to verification of distributed and multi-agent systems

点击翻译按钮获取中文摘要

Increasing demand for Distributed software systems (DSS) has led to the development of several Agent Oriented Software Engineering (AOSE) methodologies. Unfortunately AOSE methodologies do not cover the full software life cycle and usually exclude verification, monitoring and testing. This research provides a cost effective solution to distributed system verification by detecting emergent behaviour during the requirement and design phase. A practical approach for DSS development is describing its requirements using scenarios. However, scenario-based specification and behavioural modeling based on it are prone to subtle drawbacks including incompleteness and partial description. This research is focused on the emergent behaviour that can arise when behaviour model of a system with multiple components is synthesized from its scenario-based specification. The goal is automating the process of modeling, analyzing, detecting and resolving the emergent behaviour in distributed systems. The following research questions will be addressed: (Q1) what is the exact cause of emergent behaviour in distributed systems? (Q2) How to model and analyze the system in order to find emergent behaviour? (Q3) How to detect and resolve the problem? The deliverables are: (1) a 4 step process (i.e. modeling, analysis, detection and resolution) to detect emergent behaviour in DSS; and (2) a software system that is integrated with the conventional UML-based and major AOSE methodologies to provide seamless support for design verification. The proposed research has a solid theoretical basis and an implemented prototype. Through detailed industrial case studies in various real-life domain problems such as information retrieval and distributed manufacturing systems we will show the efficiency and effectiveness of the approach. The proposed research can (1) contribute to system level safety of DSS; (2) help wide spread adoption of multi-agent systems (MAS) in software industry by removing the vexatious emergent behaviour hurdle; and (3) can potentially be used to suggest possible positive emergent behaviour scenarios during MAS simulation and data mining.