摘要
MPI(消息传递接口)是最重要的主流并行计算模式之一,它既能应用于当今的分布式环境,也可用于未来的网格环境。本文对以下三种 API 语言绑定作了综合性分析:MPI-Delphi,MPI-Java 和 MPI-Ruby,并探讨了其体系架构、实现机制及相关的技术特征。MPI-Delphi 基于 DLL(动态语言连接)模式实现 Delphi 语言到 MPI 库的绑定。在MPI-Java 绑定实现时,JVM(Java 虚拟机)、JNI(Java 本地接接口)、对象串行化范型和 Java 新 I/O 库 Java.nio 都是用作 MPI 性能优化的关键技术。MPI-Ruby 能够提供给用户最易使用和最强大的接口。本文还对一些相关的绑定机制进行了介绍和分析。
MPI(Message Passing Interface)is one of the most important mainstream parallel computing models, which not only can he employed in present distributed environment but also in future grid applications.In this article, an overview and analysis on three kinds of MPI language Bindings:MPI-Delphi,MPI-Java and MPI-Ruby including which architectures,implementation mechanisms and correlated technologies are described.The MPI-Delphi provides MPI binding for Delphi based on DLL(Dynamic Language Link)mechanism.In MPI-Java binding implementations, JVM(Java Virtual Machine),JNI(Java Native Interface),object serialization paradigm and Java New I/O library- Java.nio are the key technologies that can be employed to optimize the MPI performance.MPI-Ruby,the last one, can provide user the most apt using and most powerful interface.Several related binding implementations are intro- duced and analyzed.
出处
《计算机科学》
CSCD
北大核心
2004年第8期185-189,共5页
Computer Science
基金
华中科技大学博士后基金项目:基于 Globus/Web Service 的高性能 CFD 计算网络研究(AA183107)