摘要
形式地给出了合作事务定义及事务间由于消息交换而形成的事务间的检验点依赖和恢复依赖 .基于这些定义 ,某一事务记检验点时 ,所有检验点依赖于该事务的事务也记检验点 ;当某一事务恢复时 ,所有恢复依赖于该事务的事务也恢复到其检验点的状态 .开发了一种适合于移动环境的事务恢复技术 :在事务正常的消息传送中 ,通过额外的一点带宽附加传送事务间的依赖关系 ,在事务记检验点时 ,尽可能并行地通知所有相关的事务记检验点 .实验表明 ,该事务恢复技术能减少系统记检验点的响应时间 ,提高系统的恢复速度及系统吞吐率 。
The formal definition of a cooperative transaction model was introduced. Based on message exchanges between transactions, different transaction can be checkpoint dependent and recovery dependent upon each other. To maintain the system consistency, when a transaction T took a checkpoint, all the transactions which were checkpoint dependent on T were forced to take their check points. When a transaction T rolled back to a checkpointed state, all the transactions which were recovery dependent on T were asked to roll back to their most recent checkpointed states. A recovery strategy was designed and implemented: In the normal message transportation, the checkpoint dependency and recovery dependency information were accompanied. When a transaction T initiated a checkpoint, all the transactions which were checkpoint dependent on T were informed simultaneously to take their checkpoints respectively. The experiments showed that the proposed recovery mechanism reduced the global checkpointing time latency and enhanced the recovery speed and the system throughput was increased dramatically. The requirements for the network bandwidth were not changed.
出处
《华中科技大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2003年第1期28-30,共3页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
国家自然科学基金资助项目 (6 0 2 0 30 17)
