论文部分内容阅读
1 引言众所周知,结点间密集的通信操作会极大地降低并行应用的性能,因而对于并行系统来说,一个有效的通信操作库是极为重要的。常用的通信操作包括:广播,多点播送,路障同步,全对全通信等。 1994年,MPI标准的制定更加突出了并行通信问题的重要性。近年来,针对各种不同的并行系统都有大量的通信算法出现。有些算法的性能分析相当精确,但也有很多算法只是泛泛而谈,内容过于模糊,使算法的性能和应用范围不太清楚,系统设计者很难知道该通信算法是否符合其系统的具体特性和要求。
1 Introduction As we all know, intensive communication between nodes will greatly reduce the performance of parallel applications, and thus for parallel systems, an effective library of communication operations is extremely important. Commonly used communication operations include: broadcast, multicast, roadblock synchronization, full of all communications. The development of the MPI standard in 1994 underscored the importance of parallel communications. In recent years, a large number of communication algorithms have emerged for various parallel systems. Some algorithms have quite accurate performance analysis, but many algorithms are just generalizations and the content is too vague to make the performance and application scope of the algorithm unclear. It is difficult for the system designer to know whether the communication algorithm conforms to the specific characteristics and requirements of the system .