Migration Strategy Of Mobile Agent

Research regarding Mobile Agent (MA) originated from Artificial Intelligence (AI) area and later was separated as an isolated mainstream attracting lots of research efforts. MA technology is enjoying more and more popularity as a newly developed methodology for building distributed applications. Generally speaking, a MA is a type of computer program that can migrate from one host to another in a heterogeneous network, at times and to places of its own choice. Featured with autonomy, mobility and intelligence, MA has a broad future to be employed in distributed and mobile network environment, such as software design, network load balancing and distributed computation, etc.As the most distinct characteristic of mobile agents, migration is crucial to the effective performance of an agent program. Unlike conventional computing paradigms, the migration of a mobile agent is a complex process characterized by intelligence and autonomy, which is very different from the work manner of process migration.Among the many appealing hot research spots in the MA field, in this thesis we lay emphasis on the migration strategy issue in particular. Specifically, we decomposed the migration process into several stages: locating services, determining destination, choosing appropriate time to start migrating, and selecting path to get to the destination. In other words, a well employed strategy should be able to tackle four problems -"Why", "Where", "When" and "How", which are faced by agents at different migration stages. We present our own solutions to solve these problems, the merits of our approaches are evaluated either by detailed analysis or by experiments.In Chapter 2 we first present a model for agents to locate services and to detect candidate destination hosts in the network. Based on this model, algorithms for service management and service locating are also sketched. We make a comparison between our algorithm and two existing popular approaches. In Chapter 3, a neural network based evaluation approach will be presented. The goal of this approach is to evaluate the usability of different candidate hosts and then let the agent find out which one is the most appropriate destination. Experiments and discussion are given at the end of this chapter. Chapter 2 and Chapter 3 tackle the problem that where should agents migrate to. Regarding the time choosing issue, in Chapter 4 we propose a solution that aims at solving the problem that when should agents start migrating. This solution is based on a probabilistic model of host condition. By probabilistic analysis, the varying trend of host condition can be estimated so that the appropriate time for agents to start migration can be determined. In Chapter 5 we also present a reinforcement learningscheme for routing agents in a dynamically changing network environment. Our algorithm is based on the classical Q-Routing, with some modifications on Q-function updating rule. The efficiency of this algorithm is verified by simulated experiments. This tackles the problem how agents migrate. Last but not least, we put forward a series of theorems and lemmas to formalize agents' migration process. These formalizations cover a variety of migration-related issues thus can be the infrastructures of relevant mechanisms whose performance is closely related to and even mainly determined by migration strategies.