分布式实时数据库并发控制

在集中式实时数据库中，事务并发控制一般采用基于优先级的事务重启或优先级继承．在分布式实时数据库中，由于一个主事务的各参与事务可能分布在多个节点上，事务重启一方面会造成系统资源的极大浪费，另一方面又会加大网络传输的负担．而在优先级继承策略中，被阻塞的高优先级事务的执行完全依赖于阻塞它的低优先级事务，在分布式环境中，事务的执行具有更多的不可确定的因素，这就很难保证被阻塞事务的截止期．本文提出了一种基于数据多映像的并发控制协议，事务因访问数据而形成不同的依赖关系，改变高优先级事务和低优先级事务的依赖关系从而更有利于高优先级事务而又不矢折或阻塞其他事务，从而大大提高事务执行的并发度，更好地满足实时事务的截止期．

Concurrency Control in Distributed Real-time Databases

In a Centralized Real time Database System(DRTDBS), transaction concurrency control typically adopts the priority based restart protocol or priority based block and priority inheritance. In a DRTDBS, the participating transactions corresponding to a coordinating transaction may locate on several sites, restarting a transaction results in both wastes of lots of system resources and augmentation of network transportation overhead. At the priority based block and priority inheritance circumstance, the blocked high prioritied transaction is absolutely dependent upon the blocking transaction and in the distributed environment, there are much more unpredictable factors in transaction execution. This may lead to the much more difficulties in satisfying the deadlines of the blocked transactions. This paper presents a novel concurrency control protocol based on multiple images of a data items. Transaction can be dependent upon each other because of the conflicting data accessing. If possible, at some circumstances we change the transaction dependencies by favoring the high prioritied transactions without transaction restart and blocking. This can enhance transaction concurrency and results in more real time transactions meeting their timing constraints.