| The mobile agent is a computer program that is able to continuously migrate among hosts in a network, and use local host services to execute tasks. As compared to RPC-based distributed computing model, the mobile agent computing has the advantages of lower network load, adaptability to environmental changes and support for disconected computing. As a result, the mobile agent computing is not only a research hotspot in the academic world, but also an important trend for industry applications, for example in the areas of Network Information Retrieval, Workflow Management, Collaborative Product Commerce, Intelligent Robot, Wireless Sensor Network, Network Security Monitoring, etc.In the research of the migration workflow manegement, the mobile agent is a task proxy for the manager of the business process. The network host where mobile agent executes the task is known as workplace, provided by the workflow participant and designed as a service proxy for mobile agents. The sequence of workplaces on which mobile agent executes tasks to complete its business goal is known as its working path. In order to utilize host service resources more efficiently, maximize workflow gain and improve workflow execution efficiency, mobile agent working path planning has become one of the key issues in the research of the migration workflow manegement.Mobile gent working path planning methods can be divided into two classes, which are static planning and dynamic planning. The former refers to methods that the designer, based on known host service information, generates and makes mobile agent carry the working path before sending it for migration. As a result, working path generated through static planning lacks adaptability to dynamic environmental changes. The latter assumes that every time mobile agent moves, it would dynamically select the next workplace based on its goal and knowledge of the current environment. Information of the environment can be detected by mobile agent itself, known as service discovery method, or detected by collaborating hosts, known as service recomendation method. Since service discovery method requires mobile agent to carry enough knowledge and codes for environment detection, it results in low migration efficiency and higher possibility for migration failure. Service recomendation method, however, is able to make mobile agent light weighted, improve migration efficiency and reduce migration failure, but it requires trust relationship and signing service contracts among mobile agent and service recomender. Research results of sociology show that, social acquaintance relationship is a trust relationship built through long-term cooperation. For mobile agent working path planning, adopting acquaintance network is therefore a reasonable choice to perform service recomendation method.The work of this paper, funded by the National Natural Science Foundation of China and the Natural Science Foundation of Shandong Province, uses the migration workflow conceptual model built by Professor Guangzhou Zeng as implementation framework, and rests on the basis of previous work of our group studied on the mobile agent working path dynamic planning model and method. Main work of the paper includes four parts:1. Resaerch on the model and method of mobile agent working path dynamic planning with variable decision-making space.The methods of mobile agent working path dynamic planning with service discovery require mobile agent itself to have sufficient knowledge for service discovery and ability for environment detection, making mobile agent heavy and inconvenient for migration. The methods of mobile agent working path dynamic planning with navigation mechanism can make mobile agent light weighted, but they require the whole workplace space to be organized and divided beforehand therefore they are not able to support dynamic open environment well. The methods of mobile agent working path dynamic planning with acquaintance recommendation were proposed so far, they have no mathematical model and can not able to demonstrate the dynamic nature of the mobile agent working path dynamic planning.This paper introduces Markov Decision Process (MDP) and referral network to mobile agent working path dynamic planning, with sequential business process execution as application background, built a MDP model with variable decision-making space and provided the method for mobile agent working path dynamic builiding where, the dynamic nature of mobile agent working path planning is described by MDP, the variable decision-making space during migration is composed of all the social acquaintance relationships of the current workplace member, and the service referral is generated on the direct and indirect acquaintance chain of the current workplace member. At different migrating time, the social acquaintance network of the current workplace member is different. The variability and openness of the social acquaintance relationship provides a dynamic variable decision environment for mobile agent working path planning. Chapter 2 of this paper gives the mobile agent working path dynamic planning model and method with variable decision space.2. Research on the method of mobile agent working path dynamic planning based on bounded acquaintance referral tree.Research of sociology shows that, for any rational social member, the size of social acquaintance network composed by all direct and indirect acquaintance relationships can be very large, or even boundless. Therefore, using the whole social acquaintance network to search service for mobile agent can be inefficient, and in many cases infeasible. According to the Six Degrees of Separation principle in the small world phenomenon research, in a social network, distance between any two strangers on average is no larger than 6. As a result, mobile agent working path dynamic planning adopting service recommendation method can be conducted in a finite set of social acquaintance relationships.This paper introduces Six Degrees of Separation principle to the research of mobile agent working path dynamic planning, and with sequential business process execution as application background and on the basis of the mobile agent working path dynamic planning MDP model, built a method of mobile agent working path dynamic building based on bounded acquaintance referral tree where, referral network is described by the social acquaintance relationships of the current member, bounded acquaintance referral tree is generated through pruning according to Six Degrees of Separation principle and can be built beforehand and dynamically maintained based on the coordination progress by the member. Chapter 3 of this paper discusses the method of mobile agent working path dynamic building based on bounded acquaintance referral tree.3. Research on the model and method of mobile agent working path dynamic planning based on window strategy.Similar to a chess game, in the mobile agent working path dynamic planning for multiple continuous tasks, there are two types of migration strategies, they are "planning for one step ahead before taking one step" and "planning for multiple steps ahead before taking one step". The former refers to the strategy that every time mobile agent migrates, it only looks for the suitable workplace for the current task, but does not consider the impact of the rest of the tasks to the whole working path. The latter refers to the strategy that every time mobile agent migrates, it not only considers the gain from the current task, but also considers the contribution to the overall goal from the following one or more tasks.This paper introduces "planning for multiple steps ahead before taking one step" strategy to the research of mobile agent working path dynamic building method, and with sequential business process execution as application background and on the basis of the improved mobile agent working path dynamic planning MDP model, built a method of mobile agent working path dynamic building based on window strategy where, the multiple continuous tasks that are examined during migration is known as planning window, the sequence of workplace that are one-to-one mapped to each task in the planning window is known as window path, and the first workplace on the window path is the next target workplace of the mobile agent. Chapter 4 of this paper discusses the method of mobile agent working path dynamic building based on window strategy.4. Research on the model and method of mobile agent working path dynamic planning for structured business process collaboration.Research of computer supported cooperative work (CSCW) shows that, task and result sharing is a basic mode for multiple agents to collaborate and solve problems. Therefore, the execution of a complex business process with "add/or" structure can be processed by decomposing the business process to transform a complex business process into a set of sequential task branches, having each mobile agent to execute one sequential task branch, and drawing the collaboration view between multiple mobile agents with connection patterns such as sequence, split and merge.This paper introduces principle of task and result sharing and Partially Observable Markov Decision Process (POMDP) model to the research of mobile agent working path dynamic building method, with execution of complex business processes as application background, built an agent path dynamic planning POMDP model with variable observation space and provided the method of mobile agent collaboration-oriented working path dynamic building consisting of four basic steps: decomposing structured business process, defining collaboration view and rules, assigning sequential task branches, and path planning for mobile agent collaborative work. Chapter 5 of this paper discusses the model and method of mobile agent collaboration-oriented working path dynamic planning for structured business process.The innovative points of this paper are reflected in the followings:1. Provided an MDP model with variable decision-making space and a method based on bounded acquaintance referral tree for mobile agent working path dynamic planning.In similar researches on the method of mobile agent path dynamic planning adopting service recommendation, the navigation method did not incorporate referral network and mathematical model, the acquaintance recommendation method, although incorporated referral network, did not build mathematical model, and fixed the working space of the mobile agent. These methods are not able to demonstrate the dynamic nature of the mobile agent working path planning, and lack of adaptability to environmental changes.The MDP model for mobile agent working path planning built in this paper, however, has the set of acquaintance relationships of social members as decision-making space. At different migrating time, the service members are different, and the decision-making space is also different. As a result, this model is able to well demonstrate the dynamic nature of mobile agent working path planning. The bounded acquaintance recommendation tree method can effectively reduce time for searching service referral and improve efficiency for working path planning.2. Provided an MDP model suited for window strategy and a method with variable window width for mobile agent working path dynamic planning.In similar researches on the method of mobile agent working path dynamic planning, both service discovery and service recommendation, all used "planning for one step ahead before taking one step" strategy, a strategy that only looks for the suitable workplace for the current task, but does not consider the impact of the rest of the tasks to the whole working path.The MDP model suited for window strategy and the method of mobile agent working path dynamic planning with variable window width built in this paper, however, not only examines the gain from the current task, but also considers the contribution to the overall goal from the following one or more tasks, therefore it works better for the optimization of the overall working path. The MDP model with referral network as variable decision space is able to well demonstrate the dynamic nature of the mobile agent working path planning.3. Provided a POMDP model with variable observation space and a method of mobile agent collaboration-oriented path dynamic planning for structured business process.In researches on solving the multiple mobile agent collaboration problems, the partial-global planning is the most common used method; however it lacks of mathematical model description. Since all the mobile agents have to participate in the global planning negotiation and continuously notify other mobile agents of any changes in execution of partial planning, the more mobile agents there are, the longer time the negotiation takes, the lower planning efficiency it leads to.The POMDP model with variable observation space and the method of mobile agent collaboration-oriented path planning built in this paper, however, on the basis of structured business process decomposition and task branch assignment, has the execution relations for AND-split, OR-split, AND-merge and OR-merge tasks as the collaboration knowledge of the mobile agent, therefore, unlike partial-global planning, does not have to conduct global negotiation at each step. The POMDP model with variable observation space is able to well demonstrate the dynamic nature of the mobile agent path planning.Mobile agent computing is a developing research area. There is space for discovery and improvement in the research of the mobile agent path dynamic planning model and method. The next step of research following this paper includes:1. Research on the method of mobile agent path dynamic building with time constraints. Methods in this paper did not involve the time factor in mobile agent path planning. Under many circumstances, there are time constraints for the execution of the business process. Therefore, further research on the method of mobile agent path dynamic building with time constraints is necessary.2. Research on the method of mobile agent path dynamic planning with creditability strategy. The method in this paper is based on the assumption that all referred workplaces have equal creditability. Difference in creditability was not taken into consideration. Introducing credit rating for referral workplaces is expected to improve the reliability of task execution. Therefore, research on creditability strategy and the method of mobile agent path dynamic building with creditability strategy is one of the next steps following this paper.3. Research on the method of mobile agent collaboration-oriented path dynamic planning for weak-structured business process. This paper discussed structured business process, but did not consider the case of weak-structured business process. Therefore, research in this area is needed to meet the management requierment for weak-structured business processes.
|
PDF Full Text Request