期刊文献+

基于共享存储的高可伸缩嵌入式集群模型 被引量:1

Shared Memory Based Embedded Cluster Model with High Scalability
下载PDF
导出
摘要 利用对称多处理机(SMP)作结点可为嵌入式集群带来更高的计算性价比,但多个并行和存储层次也会带来存储一致性、可伸缩性、性能差异等问题.提出一种基于共享存储的嵌入式集群模型LESC.该模型通过高度综合实现"计算单元-互连一致性模块-系统"三级高可伸缩结构,获得功耗成本有效性.LESC完成分布式共享存储的基本功能,其目录缓存一致性和扩展的共享存储机制改善了传统存储层次,并利用"共享存储虚拟网络"提供模块级的高效通信,避免了网络硬件开销,同时支持MPI编程.经该模型的真实系统平台测试,模块内MPI通信性能是传统嵌入式集群的3倍以上,单元间通信性能可达单元内性能的86%以上,Linpack测试其扩展性能在最差情况下接近理想值的70%. 利用对称多处理机(SMP)作结点可为嵌入式集群带来更高的计算性价比,但多个并行和存储层次也会带来存储一致性、可伸缩性、性能差异等问题.提出一种基于共享存储的嵌入式集群模型LESC.该模型通过高度综合实现"计算单元-互连一致性模块-系统"三级高可伸缩结构,获得功耗成本有效性.LESC完成分布式共享存储的基本功能,其目录缓存一致性和扩展的共享存储机制改善了传统存储层次,并利用"共享存储虚拟网络"提供模块级的高效通信,避免了网络硬件开销,同时支持MPI编程.经该模型的真实系统平台测试,模块内MPI通信性能是传统嵌入式集群的3倍以上,单元间通信性能可达单元内性能的86%以上,Linpack测试其扩展性能在最差情况下接近理想值的70%.
出处 《计算机研究与发展》 EI CSCD 北大核心 2012年第S1期245-251,共7页 Journal of Computer Research and Development
基金 国家"核高基"科技重大专项基金项目(2009ZX01028-002-003 2009ZX01028-001-003) 国家自然科学基金项目(60736012 60921002 61050002 60803029 61173001 61003064 61100163 61070025)
关键词 分布式共享存储 嵌入式集群 目录缓存一致性 共享存储虚拟网络 MPI DSM embedded cluster directory based cache coherency shared memory virtual network MPI
  • 相关文献

参考文献12

  • 1王焕东,高翔,陈云霁,胡伟武.龙芯3号互联系统的设计与实现[J].计算机研究与发展,2008,45(12):2001-2010. 被引量:22
  • 2Duranton M.The challenges for high performance embedded systems. Proc of the9th Euromicro Conf on Digital System Design . 2006
  • 3Sukaridhoto S,,Sasaki Y,Ito K,et al.Development of a compact cluster with embedded CPUs. Proc of the6th Industrial Electronic Seminar . 2004
  • 4Fowler M R,Stipidis E,Ali F H.Practical verification of an embedded Beowulf architecture using standard cluster benchmarks. Proc of ICSEA’’08 . 2008
  • 5Kontothanassis L,Stets R,Hunt G,et al.Shared memory computing on clusters with symmetric multiprocessors and system area networks. ACM Transactions on Computational Logic . 2005
  • 6Buntinas D,Mercier G,Gropp W.Design and evaluation of Nemesis,a scalable low-latency message-passing communication subsystem. Proc of the6th IEEE Int Symp on Cluster Computing and the Grid . 2006
  • 7Hu Weiwu,Wang Jian,Gao Xiang,et al.Godson-3:A scalable multicore RISC processor with x86emulation. IEEE Micro Magazine . 2009
  • 8Liu Jiuxing,Chandrasekaran B,Wu Jiesheng,et al.Performance comparison of MPI implementations over Infiniband,Myrinet and Quadrics. Proc of the2003ACM/IEEE Conf on Supercomputing . 2003
  • 9A. Gara,M. A. Blumrich,D. Chen.Overview of the blue gene/l system architecture. IBM Journal of Research and Development . 2005
  • 10A. Ahmed,,P. Conway,,B. Hughes,,F. Weber.AMD Opteron Shared Memory MP Systems. Proceedings of the 14th HotChips Symposium . 2002

二级参考文献8

  • 1Wei-WuHu Fu-XinZhang Zu-SongLi.Microarchitecture of the Godson-2 Processor[J].Journal of Computer Science & Technology,2005,20(2):243-249. 被引量:52
  • 2Hu W, Wang J, Gao X, et al. Micro-architecture of Godson 3 multi-core processor [EB/OL]//Proc of the 20th Hot Chips. 2008 [ 2008-11-20]. http://www. hotehips. org/he20/main page. htm.
  • 3HyperTransport Technology Consortium. HyperTransport ^TMI/O I.ink Specification Revision 1. 03[M/OL]. 2001 [2008-11- 20]. http://www. hypertransprot. org/default. elm? page = HyperTransportSpecifieationslx.
  • 4ARM. AMBA AXI Protocol vl. 0 Specification [M/OL]. 2004[2007-05-10]. http://www. arm. com/products/solutions/ axi_spec. html.
  • 5Lamport L. Time, clocks, and the ordering of events in a distributed system [J]. Cornmunnieations of the ACM, 1978, 21(7): 558-565.
  • 6Gharachorloo K, Lenoski D, Laudon J, et al. Memory consistency and event ordering in scalable shared-memory multi processors [C] //Proc of the 17th Int Syrup on Computer Architecture (ISCA'90), Los Alamitos: IEEE Computer SOciety, 1990: 28-31.
  • 7Culler D, Singh J, Gupta A. Parallel Computer Architecture [M]. San Francisco: Morgan Kaufmann, 1996.
  • 8胡伟武,water.chpc.ict.ac.cn,施巍松,water.chpc.ict.ac.cn,唐志敏,water.chpc.ict.ac.cn.A Framework of Memory Consistency Models[J].Journal of Computer Science & Technology,1998,13(2):110-124. 被引量:1

共引文献21

同被引文献7

引证文献1

二级引证文献3

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