Fault Tolerant Technology Based On Autoecious Fault Detection In Distributed Paralled Server System

With the global development of economics, especially the birth of e-business, industrial economic is changing into knowledge economics. Enterprises depend greatly on information. Current server systems can't satisfy the demand in reliability. It is significative to develop a highly fault tolerant server system. The dissertation puts forward a system architecture of distributed parallel server system which integrates technologies such as fault tolerant technology, distributed technology, parallel processing technology and network technology, etc. And on the basis of this architecture, this dissertation researches the mechanisms of multi-model replica control and fault tolerant based on autoecious fault detection, then realizes the "perpetual motion" of server system.Based on the research of Web server systems and distributed parallel database system, the dissertation introduces the definition of distributed parallel server system by combining the related concept of distributed system and parallel processing system and researches system architecture and modules according to the demand of fault tolerant processing, composed with the related concept of distributed system and parallel processing system. Distributed parallel server system is composed of four parts: interface system, execution system, communication system and management system. Interface system fulfills the interaction between servers and clients. Execution system is in charge of completing the task requests of users. Communication system completes the exchange of information. And management system is responsible for the management and maintenance of the whole system.In a distributed parallel server system, under the premise of user information reliability, the mechanism of replica control must try its best to increase the system utilization efficiency. On the basis of previous research achievement, the dissertation put forward a prediction replica control algorithm, which predicts the arriving tasks of every node by selecting proper mathematics modules and then takes multi-goal...