摘要
面对编程语言以及并行计算实现方式的多样性,如何为不同语言选择合适的并行方式,解决不同硬件环境和语言对并行技术的约束。首先介绍了基于线程的Open MP和Java的并行编程模型,适合单机多核的并行计算;其次详细阐述了基于进程的MPI技术,适合单机多核和集群系统,主要从MPI编程过程和PBS作业调度体系做了研究,并以两种队列方案的对比说明PBS作业调度的实现过程。通过对并行技术的研究,既可实现流行语言Java的并行计算,又可实现C、C++、Fortran语言的并行;既可实现单机多核的并行,又可实现集群系统的并行;既可实现指定节点的并行,又可由调度系统分配节点并行。
Faced with the diversity of programming languages and parallel computing implementations,how to select the appropriate paral-lel way for different languages,solve constraints of parallel technology for different hardware environment and language. Parallel program-ming model for OpenMP and Java based on thread is first introduced briefly,which is suitable for parallel computing on multi-core com-puter. Then,MPI technology based on process is described in detail,which is suitable for multi-core computer and cluster system. It is mainly researched from the programming process of MPI and PBS job scheduling system,and the realization of PBS job scheduling is il-lustrated from comparison between two queues. Through the study of parallel technology,it can achieve parallel computing for Java that is popular language,but also for C,C++,Fortran,implement parallel computing for single multicore,but also for cluster system,realize par-allel computing by the specified nodes,but also by distributed nodes by the scheduling system.
出处
《计算机技术与发展》
2015年第9期174-177,181,共5页
Computer Technology and Development
基金
广东省科技计划项目(粤科财字[2013]82号
粤财工[2012]406号)
关键词
并行计算
线程
进程
OpenMP
Java
MPI
PBS
parallel computing
thread
OpenMP
Java
process
MPI
PBS