Middleware-based Enterprise Cluster System Structure Design

With the continuous development of all kinds of network applications and enterprise information systems, data sharing and integration would be in urgent need. Data centralizing and Enterprise Application Integration (EAI) will be of great benefit to an enterprise, which need a solution with high scalable capability and facility in management, which would provide a powerful computing and storing capability.However, most current clusters are built on high performance computing, which has some deficiency of supporting enterprise information system. The data integrity and consistency couldn't completely guarantee.To solve these problems, this thesis analyzes the pivotal technology on common cluster architecture, which includes parallel computing and communication, process migration, load balancing and fault tolerant. According to the actual demand of the EAI environment, a cluster architecture based on middleware for enterprise computing (MECS) is designed.Stem from the service's separability and non-balancing, the hierarchy model is introduced by MECS to provide the flexible configuration and management characteristic besides high performance.After analyzing the cluster load-balancing algorithm, this thesis presents an algorithm based on dynamic feedback, which regards the residual task queues in CPU as the index of node's computing resource. Taking the weighted parameter into the least connection number scheduling, this algorithm can accurately predict the node's load status in the near future.Eventually, a model evaluation and simulation are used to test MECS. It mainly focuses on the data relevancy and dependency. The experiment proves that the approach is valid and practical.