软件事务存储自适应竞争管理策略研究

Adaptive Contention Management Strategy Research of Software Transactional Memory

在并行程序设计中,软件事务存储(STM)是对传统基于锁的同步机制的一种替代方式。其中竞争管理器用于消解事务冲突,决定哪个事务继续执行,哪个事务放弃,对于STM系统的性能有非常重要的影响。但是当前已有的竞争管理策略因受限于环境负载和STM系统的参数配置,性能表现往往是不稳定的。因此,提出了一种自适应的竞争管理策略(ACM),该策略通过构建竞争管理器池(Contention Manger Pool)和自适应竞争管理算法动态地为STM系统选择当前最优的竞争管理策略。实验结果表明ACM不仅可以通用于不同环境而且其平均性能也优于现有的静态竞争管理策略。

In the design of parallel programs,software transactional memory （STM）is an alternative to traditional lock synchronization. The competition manager for digestion transaction conflict,determine which transaction continues to execute,which transaction give up,have a very important influence on the performance of STM system. However,the performance of existing contention manager is sensitive to transaction workloads and STM system configuration always tend to be unstable. This paper proposes an adaptive strategy for competition management （ACM）,the strategy adopted by the construction of competitive manager pool and the adaptive contention management algorithm dynamically select a suitable contention manager for the current transaction workload.The experimental results show that ACM not only can be used in different environments and its average performance is better than that of the existing static competition management strategy.