支持分布式合作实时事务处理的协同检验点方法

A Coordinated Checkpointing Method for Distributed Cooperative Real-Time Transaction Processing

在实时事务执行时 ,事务故障或数据竞争会导致事务重启 ,为减少事务重启损失的工作量 ,可以采用检验点技术保证事务的时间正确性 .在一类分布式实时数据库应用中 ,不同结点的事务通过消息交换形成合作关系 ,为保证合作事务间的全局一致性 ,当某一事务记检验点时 ,相关事务也要记检验点 .传统协同检验点方法没有考虑应用的定时约束 ,不能很好地支持分布式合作实时事务处理 .该文提出了一种基于图论的协同检验点方法 ,利用在每个计算结点上为每个合作事务集维护的局部有向图 ,使用一个基于图论的计算过程标识出应记检验点的事务 ,该方法既具有最小协同检验点特性 ,又使全局检验点的时延最小 .实验表明该算法减少了全局检验点时延 ,有利于实时事务截止期的满足 .

Transactions may restart due to transaction failure and data contention in real-time database systems. Taking checkpoints periodically can reduce the amount of the lost work of real-time transactions. In some distributed real-time database applications, transactions in different sites execute cooperatively and communicate with each other by message exchanges. To maintain the consistency among these transactions, when a transaction initiates a checkpointing procedure, the related transactions are forced to take corresponding checkpoints. Traditional coordinated checkpointing methods do not take the time constraints of applications into consideration and are unsuitable for distributed cooperative real-time transaction processing. This paper proposes a coordinated checkpointing method based on graphic theory. A local directed graph is maintained at each site for each distributed cooperative real-time transaction set. Using involved local directed graphs, transactions that should take checkpoints are identified by an algorithm based on graphic theory during a global checkpointing procedure. The proposed method not only minimizes the time latency involved in a global checkpoint, it also forces a minimal number of transactions to take their checkpoints. Extensive experiments show that the proposed method results in the reduced time latency and an obvious reduction in the transaction miss rate as well.