基于多核机群的事务内存并行编程模型的研究  

Research on Software Transactional Memory Parallel Programming Model Based on Multicore Cluster

作　　者：李文敬[1] 李双[1] 元昌安[1] 廖伟志[2] 

机构地区：[1]广西师范学院计算机与信息工程学院,南宁530023 [2]嘉兴学院数理与信息工程学院,浙江嘉兴314001

出　　处：《小型微型计算机系统》2014年第8期1732-1737,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61163012)资助;广西自然科学基金项目(2012GXNSFAA053218)资助;广西高校科学技术研究项目(2013YB147)资助

摘　　要：为了解决多核机群并行程序中共享数据所使用的锁机制存在的不足,提出多核机群软件事务内存(STM,Software Transactional Memory)并行编程模型.首先分析MPI+OpenMP并行编程模型的优缺点,然后根据软件事务内存的优势,对多核节点内的多线程共享存储采用乐观检测冲突及优先数优先的竞争管理策略设计事务内存替代锁机制,在多核机群平台中构建新的MPI+OpenMP+STM三级并行编程模型,并将模型应用于随机摄动人工蜂群并行算法的设计与实现,运用MPI+TBB平台对人工蜂群并行算法进行对比实验.实验结果表明,该模型是解决多核机群数据共享锁机制问题的一种有效的编程模式.In order to solve the existing problems of the locking mechanism used in the shared data of multi-core cluster parallel pro- gram, the paper puts forward multi-core cluster software transactional memory parallel programming model. First the advantages and disadvantages of ＂MPI ＋ OpenMP＂ parallel programming model are analyzed. Then according to the advantage of software transac- tional memory, the multithreaded shared memory is under optimistic collision detection and priority number priority competition man- agement strategy is used to design the transactional memory instead of locking mechanism in multi cores. The new ＂MPI ＋ OpenMP ＋ STM＂ parallel programming model is constructed in multi-core cluster platform. This model is applied to the design and imple- mentation of the random perturbation artificial bee colony parallel algorithm, which is contrasted in ＂MPI ＋ TBB＂ platform. The ex- perimental results show that the model is an effective programming mode in solving multi-core cluster data sharing.

关 键 词：多核机群 锁机制 消息传递 软事务内存 检测冲突 并行编程模型 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]