【摘要】 提出了一种适合分布式实时内存数据库的、基于日志的故障恢复模式.在该恢复模式中,日志记录被存储在非易失高速存储设备中,非易失日志存储区依据事务类型被分成四个不同的日志分区,不同类型事务的日志被存储在不同的日志分区;采用了执行频率动态可调的模糊检验点模式,在检验点执行过程中,一些无用的日志记录被删除;故障发生后,系统重启、恢复时,分区重装策略被采用;恢复算法结合相关恢复协议,确保了全局事务故障的原子性.实验显示,该故障恢复模式明显地改进了系统性能,提高了实时事务满足截止期的比率.更多还原

【Abstract】 This paper presented a logging-based recovery scheme for distributed real-time main memory database. In the scheme, log records were stored in non-volatile high-speed store. It was divided into four different partitions in accordance with transaction types. The log records belonging to the different transaction types were stored by the different logging partitions. Fuzzy checkpoint, whose executing frequency can be adjusted dynamically, was adopted. In checkpointing procedure, some useless log records were deleted. During restarting recovery after a site crash, partition-reloading policy was adopted. In combination with the corresponding recovery protocols, recovery algorithm guaranteed the atomicity of global transaction failure. Experiments showed the recovery scheme improved the system performance and raise the rate of meeting the deadlines of real-time transactions.更多还原