一种分布式实时事务调度算法  

A Kind of Distributed Real-time Concurrency Control Algorithm

作　　者：刘云生[1] 覃飙[1] 杨进才[1] 

机构地区：[1]华中科技大学计算机学院,湖北武汉430074

出　　处：《小型微型计算机系统》2003年第6期962-965,共4页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目 ( 60 0 73 0 45 )资助 ;国防预研项目 ( 0 0 J15 .3 .3 .JW0 5 2 9)资助

摘　　要：传统的乐观并发控制策略利用了一些不必要的事务重启来保证数据的一致性 ,事务重启能够极大的增加系统载荷以及加强资源和数据的竞争 ,在分布式环境下 ,由于系统的复杂性和较高的通讯开销加剧了该问题 .针对该问题本文提出了一种新的乐观并发控制策略 ,通过动态调整事务串行化执行顺序来避免不必要的事务重启 .当把这种新的并发控制策略在分布式实时环境中实现时 ,考虑到分布式事务的实时性要求 ,本文提出把写阶段从临界区中分离出来的方法 ,并用顺序加锁的策略来保证分布式事务执行的正确性 。The traditional optimistic approaches incur restarts unnecessary to ensure data consistency. Transaction restarts can significantly increase the system workload and intensify resource and data contention. In distributed environments, the complexity of the system and the high communication overhead exacerbate the problem. In order to mitigate this problem, this paper proposes a new optimistic concurrency control algorithm that can avoid unnecessary restarts by adjusting serialization order dynamically. Considering the property of distributed real time transaction, when we implement the new algorithm in the distributed real time environment, this paper proposes a new strategy that the write phase can be disintegrated from the validation phase by pulling it out of the critical section. And the sequential locking method is adopted to ensure the correct execution of the distributed transaction. Finally, its correctness is proved.

关 键 词：分布式实时事务 动态调整串行化顺序 临界区 顺序加锁策略 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]