Research On Mobile Agent Computing Model And Its Application In Distributed Parallel Computing

An active computation model based on mobile agent technology is proposed in this thesis, which goal is to study whether it is feasible and efficient to apply mobile agent technology to large scale distributed parallel computing.Because of its many features such as autonomy, reactivity and mobility, mobile agent technology can be used suitably for distributed parallel computing in dynamic and heterogeneous network. After the introduction of the technologies and applications which are relative to distributed parallel computing based on mobile agents, the system infrastructure and basic model based on n calculus are given.If mobile agent technology is to be applied to distributed parallel computing, Both the performance and programmability must be taken into consideration overall.A lightweight high-performance-computing-oriented agent migration model is proposed and implemented in this thesis. An agent is made up of many strategies, which are physically located on strategy bases or local resident. When the agent moves to the other place, it is just needed to transport a blueprint which contains the strategy names and running states. When the blueprint arrives, it is used to regenerate the agent. Hence the activity of mobile agents is greatly increased.A multi-tier distributed tuple spaces model is proposed and implemented in this thesis. The design of distributed tuple spaces guarantees the efficiency and flexibility of coordination among agents, as well as the system scalability that will contributes to large scale distributed parallel computing.A asynchronous group communication service proposed in this thesis is based on tuple space and support migrating agents as members of the group. Not only can this service guarantee virtual synchrony to general group actions (such as process crashes, joins, and leaves), but can do to mobile actions. The group broadcast is asynchronous in that the broadcast message is written to application's tuple space after the communication is synchronized by group communication service middleware. How and when to use the message is decided by application so as to ensure the flexibility of application development.In this thesis we investigate the feasibility and efficiency of mobile agent technology for distributed parallel computing applications. Our discussion and observations are based on several typical case studies from special domain.n calculus is suitable to describe mobile computation for its simplicity and its potential to express mobility. On the basis of standardπcalculus, some concepts(such as network, place) and some pre-actions (such as mobile migration, group communication, reading writing or tuple space) is expanded. The definitions of barbed bi-simulation and bi-simulation are given, and some basic features are analysed.