The Research And Implementation Of MapReduce-based Distributed Rule Matching System

With the wide use of the rule-based system, the efficiency of rule matching arouses more and more attention. Since the seventies of last century, people have been trying to enhance the efficiency of rule matching. After studying and summing-up the outstanding rule inference engines, while studying the distributed computing framework, we propose a MapReduce-based architecture for distributed rule matching. Using the concept of MapReduce, the architecture will benefit production system in efficient and performance, especially for large scale of rules and facts for special. And the system has better scalability and flexibility.Firstly, the thesis analyses the system requirement from three aspects:the overall requirement, the external interfaces and the performance and reliability requirement. Then, we propose the overall system architecture. In accordance with the MapReduce, the system uses a Master/Worker model. The Master Server is responsible for the decomposition and allocation of tasks, and manages all the Workers, while the Map Worker and the Reduce Worker, respectively, are responsible for the distributed rule matching and the result merging.Secondly, we give an introduction to the task allocation strategy of the system. Based on the dividing of rule into the sub-rules, we propose a sub-rule allocation strategy. According to the server load information and the assigned sub-rules information, new sub-rules and the corresponding matching tasks are evenly allocated to the distributed environment. The allocation of the facts is base on the distribution of the sub-rule. The facts are assigned to the Map Worker exists the corresponding sub-rules. And we give the specific Map function to deal with the matching.Thirdly, the way of firing and execution of the distributed rule are introduced in detail. We give the specific Reduce function to merge intermediate results. In processing the firing of rules, we propose the conditions and the methods to firing a rule, and give the way to control the firing of rule. In dealing with conflict, we give some common Conflict Resolution Strategies.Finally, we present the class diagram of prototype system which contains Master Server, Map Worker, Reduce Worker and the IO stream. The system performance was tested in the end.