基于混合粒度冲突检测的事务工作流调度算法

事务工作流由若干个平面事务组成,其执行满足松弛原子性.由于组成事务工作流的平面事务具有不同的完成特性,为了防止不可串行化的执行,现有的调度算法通常只允许一个活动工作流执行不可补偿事务,这大大限制了并发度.定义了基于事务类型和事务实例两种粒度的冲突关系,并提出了一种基于这两种粒度冲突检测的调度算法,保证了并发事务工作流的可串行化和可恢复执行.该算法从两个方面提高了并发度:一方面通过事务实例之间(细粒度)的冲突检测减少了工作流冲突的概率;另一方面通过事务类型之间(粗粒度)的冲突预测,允许多个将来不冲突的工作流执行不可补偿事务.

A Scheduling Protocol for Transactional Workflows Based on Mix-Grained Conflict Detection

A transactional workflow is composed of traditional flat transactions, and its execution has relaxed transactional atomicity. Due to different termination characteristics of transactions, only one workflow is allowed to execute non-compensatable transactions with current scheduling protocols. In this paper, two granularities of conflict based on transaction classes and transaction instances are defined, and a scheduling protocol by using both granularities of conflict detection is proposed. Besides generating serializable and recoverable schedules, this method provides a higher degree of concurrency in following two ways. On the one hand, the fine-grained locking mechanism based on transaction instances is used to reduce conflict possibility among concurrent workflows. On the other hand, the coarse-grained conflict mechanism based on transaction classes is used to predict future conflict among workflows, multiple workflows are therefore allowed to execute non-compensatable transactions if they will not conflict in predicated future execution.