期刊文献+

基于RapidIO的MPI设备层的设计与实现 被引量:2

Middleware for MPI based on RapidIO
下载PDF
导出
摘要 针对RapidIO网络的特点,分析MPICH2的层次设计以及建立在TCP,SCTP网络通信协议上的MPI通信方法,通过重新定义ADI3下的CH3层,设计并实现了一种基于RapidIO的MPI设备层,建立了从MPI到RapidIO的通信通道并实现了多流通信的思想。通过在装有RapidIO网卡机器上的实验表明,在带宽和延迟通信性能上,这种专用的MPI设备层要比以太网模拟器表现出色,而且对于大数据量的通信,性能表现更好。 The structure and feature of MPICH2 is analyzed and its communication based on TCP and SCTP protocol for transferring data is illustrated. By redesigning the CH3 layer which is the lower layer of ADI3, the special middleware for RapidIO and the idea of multi-stream communication so that it can build a way of transferring data from MPI to RapidlO are implemented. The experimental result shown that this special middleware have considerable advantages comparing with the Ethernet simulator, especially for those big block data communication.
出处 《计算机工程与设计》 CSCD 北大核心 2008年第21期5464-5467,共4页 Computer Engineering and Design
基金 国家863高技术研究发展计划基金项目(2005AA104031) 上海高校网格技术E-研究院基金项目(20030103)
关键词 消息传递接口 MPICH2 RAPIDIO 设备层的设计与实现 集群 高性能计算 message passing interface MPICH2 RapidIO implementation ofmiddleware cluster high performance computing
  • 相关文献

参考文献8

  • 1Message Passing Interface Forum. MPI: A message-passing interface standard[EB/OL], http://www.mpiforum.org/docs, 1994.
  • 2Message Passing Interface Forum. MPI: A message passing interface standard[EB/OL]. 1997.
  • 3William Gropp,Brian Toonen. The CH3 design for a simple implementation of ADI3 for MPICH2 with a TCP-based implementation[Z]. 2002.
  • 4Kamal H, Penoff B, Wagner A. SCTP versus TCP for MPI [C]. Washington, DC, USA: Proceedings of the ACM/IEEE Conference on Supercomputing,2005.
  • 5PenoffB,Wagner A.Towards MPI progression layer elimination with TCP and SCTP[C]. 11th International Workshop on High- Level Programming Models and Supportive Environments. IEEE Computer Society,2006:32-46.
  • 6Aulwes R T, Daniel D J,Desai N N, et al. Architecture of LAMPI, a network fault-tolerant MPI[C].Sante Fe,New Mexico: 18th International Parallel and Distributed Processing Symposium,2004, http://www.mpiforum.org/docs/mpi-20.ps.
  • 7Argonne National Laboratory[EB/OL]. http://www-unix.mcs.anl. gov/mpi/mpich.
  • 8Network-based computing laboratory of the Ohio state university[EB/OL], http://mvapich.cse.ohio-state.edu/benchmarks/.

同被引文献11

  • 1辛永利.高性能嵌入式信号处理应用中的互连结构[J].航空计算技术,2004,34(2):71-74. 被引量:3
  • 2Orbet J,Rubini A,Kroah-Hartman G.Linux设备驱动程序[M].魏永明,耿岳,钟毅,译.3版.北京:中国电力出版社,2007:258-285.
  • 3Benvenuti C.Understanding Linux network intemals[M].Ameriean: O' Reilly, 2005.
  • 4RAPIDIO TRADE ASSOCIATION.RapidIO interconnect specifi- cation[EB/OL]. (2005-06).http://www.rapidio.org.
  • 5Romanow A, Bailey S.An overview of RDMA over IP[C]//1st International Workshop on Protocols for Fast Long-Distance Networks, 2003.
  • 6Feng B.串行RapidlO:高性能嵌入式互连技术[EB/OL].(2007).http://focus.ti.com.cn/crggeneral/docs/gencontent.tsp?contentId=50741.
  • 7Maloy J.TIPC: Transparent inter process communication protocol (Version2.0) [EB/OL]. (2006-05) .http://tipc.sourceforge.net/doe/draft- spec-tipe-02.html.
  • 8冀映辉,蔡炜,蔡惠智.TIPC透明进程间通信协议研究和应用[J].计算机系统应用,2010,19(3):76-79. 被引量:5
  • 9卢云娥,黄宗宇,李超阳,郭祥斌,殷慧明.基于微机集群系统的MPI并行计算[J].电子设计工程,2011,19(5):78-81. 被引量:12
  • 10孙忠义,金同标,殷进勇.通信协议性能测量与分析[J].计算机系统应用,2012,21(9):224-227. 被引量:3

引证文献2

二级引证文献5

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部