Research And Implementation Of Fault-tolerant Mechanism In The System Of Digital Organisms,

Non-centralized Distributed System which has no single failure and provides services over WAN with highly scalability is becoming the research focus in recent years. However, most of its applications are confined to file sharing which has low requirement for fault-tolerance. Consequently, the fault-tolerant mechanism in this kind of system hasn't been studied well, but it has to be done the same as in the traditional distributed system in order to support more applications.The Digital Organism System is a kind of computer system that owns the property of organism, which is based on distributed parallel system developed by 8010 Lab, UESTC for many years. The dissertation integrated with fault-tolerance, distributed principle, parallel processing as well as networking technology, proposed several novel fault-tolerant mechanisms which especially catered for the Digital Organism System's characteristics.Based on available resource search algorithms and fully analysis on resource characteristics, the dissertation proposed to divide resource into two sections, namely original data and attribute information, managing them with independent algorithms separately and integrating them smoothly to achieve resource management. In addition, combining with resource search rule, the management mechanism of attribute information could accelerate the access to resource. Abnormal processing mechanism designed elaborately could guarantee the highly reliability of getting access to resource.Based on fully studying on the available distributed mutux algorithm, the dissertation proposed a kind of mutex mechanism which has the characteristics of quick, reliable, load-balance by combining unique features of non-centralized distributed system,. By assigning different lock-authorizing node sets to different resources, load-balance could be achieved in the course of lock-authorizing at the maximum level. By backing up lock request among nodes, lock-authorizing process with fast and reliable features could be achieved.In allusion to the poor performance of current node failure detection algorithms under the complex networking structure, the dissertation proposed a kind of self-adapted node detection algorithm. It could automatically switch among stellate, orbicular and miscellaneous detection modes according to real communication state. In doing so, failed nodes could be kicked out immediately.Several aspects mentioned in the dissertation are necessarily taken into account in designing any highly fault-tolerant system. As for this, the dissertation is valuable for reference.