提出了一种面向事务处理的容错计算机体系结构,设计了基于三模冗余的仲裁器和仲裁算法。在集群中,利用中心结点故障检测机制,采用仲裁器得到了一个逻辑上高可靠的管理态结点,通过该结点可以实现系统快速故障诊断和故障恢复。实验结果表明:节点间故障检测时间小于2s,故障节点切换时间小于5s,该系统的故障检测、故障节点切换可以满足事务处理系统对可靠性的要求。
A fault tolerant computer architecture for transaction processing was presented and an arbitration module and an arbitration algorithm were designed based on a triple modular redundancy. Using the center-node detection mechanism, a logically reliable management node was implemented by using the arbitration module in a cluster system, through which the system's rapid fault diagnosid and fault recovery can be accomplished. The experimental result showsed that the fault detection time was less than 2 seconds and the switch time of fault node was less than 5 seconds. So the performance of the system on fault detection and fault nodes switching can satisfy the reliability requirement for transaction processing.