Research Of Failure Detection Model Based On Grey Predication

Recently, with the distributed system becoming more and more widely used, the demand of its availability is gradually enhanced. As a key technology of realizing high availability, the realization of failure detection in the distributed system becomes a widely discussed question. It has been widely used in application systems, such as communication protocols, distributed algorithms, grid computing, group membership management and web server and it is always used to solve fundamental issues in distributed systems, such as coherence and atomic multicast issues. At present, the study of failure detection is mainly about its algorithm, modeling and level.Many traditional failure detection algorithms which give the next arrival time of heartbeat message are based on model of probability and statistics. This needs large amounts of data and it is supposed to accord with some specific distributions, this determines the limitation of its application. Due to the characteristic of grey theory which is modeling based on small amounts of data, a failure detection model based on grey prediction is proposed and a PGC-AFD adaptive failure detection algorithm is given on the base of this model. To improve the detection accuracy, checking detection is proposed in this paper. It uses grey prediction algorithm to give a time-out value which is used to judge whether the next message is failed or not by calculating the historical records of message delayPGC-AFD failure detection algorithm in this paper satisfies a high detection level. Experimental results show that PGC-AFD algorithm improves the performance of failure detection to some extent and reduces the influence of network delay on failure detection.