论文部分内容阅读
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.
The transaction workflow is composed of a number of plane transactions whose execution satisfies loose atomicity.Because the plane transactions that make up the transaction workflow have different completion characteristics, in order to prevent non-serializable execution, the existing scheduling algorithms usually only allow one activity Which greatly limits the degree of concurrency.This paper defines the conflict between two types of granularity based on transaction type and transaction instance and proposes a scheduling algorithm based on these two granularity conflict detection to ensure that the concurrent transaction work Streaming can be serialized and recoverable execution.This algorithm improves the concurrency degree in two aspects: on the one hand, the conflict detection between transaction instances (fine granularity) reduces the probability of conflict of workflow; on the other hand, (Coarse-grained) conflict prediction, allowing multiple workflows that do not conflict in the future to perform non-compensable transactions.