共享内存并行编程最优同步方法的研究

共享内存并行编程是并行编程开发与研究的标准之一,为了保证程序运行的准确性和线程执行的高效率,对共享内存同步机制的研究具有重要意义。互斥锁与事件同步是Open MP标准中的混合同步机制,而斯坦福大学的Baek等人扩展Open MP提出的面向事务存储的Open TM应用编程接口,事务内存同步是其突出特点。通过研究发现混合同步存在部分互斥锁同步会严重消耗系统空间资源、线程资源利用率有待提高等问题,而事务同步忽视不可恢复行为的影响。因此,提出一种基于共享内存的最优同步方法,即优化后的混合同步和事务内存同步的结合体,实验论证此优化方法的可行性,程序运行更加高效。

RESEARCH OF THE MOST OPTIMAL SYNCHRONIZATION METHOD IN SHARED-MEMORY PARALLEL PROGRAMMING

Shared-memory parallel programming is one standard of the parallel development and research, in order to ensure the accuracy of program and high efficiency thread of execution, the research of synchronization mechanisms has important significance. Hybrid synchronization composed of mutex lock and events is the synchronization mechanisms in OpenMP standard. And transactional memory is the salient features of the transaction-oriented storage OpenTM application programming interface, expanded OpenMP interface proposed by Beak of Stanford University. There are some problems in these two synchronization mechanisms, for example some mutex synchronization consumes system storage resource serious and the utilization of thread needs to be improved in hybrid synchronization. And transaction synchronization ignore unrecoverable behavior. Therefore, this paper presents an optimal synchronization method based on shared memory, the combination method of hybrid synchronization and transaction memory mechanisms. The feasibility of this optimization method is demonstrated by experiments, and the program run more efficiently.