ISCAS OpenIR  > 中科院软件所  > 中科院软件所
分布式并行数值代数软件的数据重分布及测试
李基凯
Major计算机科学
1998
Degree Grantor中国科学院软件研究所
Degree Level博士
Place of Degree Grantor中国科学院软件研究所
Keyword并行 代数 分布 测试
English Abstract数据重分布是提高并行数值代数软件效率的一个关键技术。在并行机的实际应用中,线性代数在相当大一部分高性能科学计算中起着核心作用,随着并行技术的成熟和推广,研制一批类似BLAS(Basic Linear Algebra Subroutines)、LAPACK(Linear Algebra Package)的并行的通用数值代数软件包提上日程。这些通用并行软件包也应象以往的串行程序那样:高效、灵活,然而,由于分布式多级存储系统的日趋复杂、存储速度严重滞后处理器的处理能力,存储系统的瓶颈效应明显,通用并行软件要想达到高效、灵活的目标非常困难,事实证明:适当的采用数据重分布技术可以有效地提高并行程序总体性能,并实现高效与灵活的均衡。本文在分析、测试、评估PBLAS(parallel basic linear algebra subprograms)的基础上将详细地分析数据重分布技术。在移植并行软件的过程中为了保证移植软件的正确性,我们针对一个已有的串行C语言测试工具CTT进行了改进:简化其结构、增强其对信号等的处理,使之具有对BLACS、PBLAS等并行软件进行测试的能力。
AbstractIn practice, numerical linear algebra programs play an important role in parallel computing. With the development of parallel computer, there is a urgent need to develp linear algebra with flexibility and efficence. To achive such target, one should pay particular attention to data distribution. This paper is based on the work that we porting and improving PBLAS. The PBLAS (Parallel Basic Linear Algebra Subprograms) is a component of the ScaLAPACK library. It is targeted at distributed vector-vector, matrix-vector and matrix-matrix operations with the aim of simplifying the parallelization of linear algebra codes, especially implemented on top of the sequential BLAS. In this paper we will describe data distribution formally, and display that efficiency and flexibility are not antagonistic features of the block cyclic mappings. We also will give a brief introduction of a Parallel C language testing tool that we developed on the basis of CTT, a sequential testing tool.
Pages45
Language中文
Content Type学位论文
URIhttp://ir.iscas.ac.cn/handle/311060/5656
Collection中科院软件所_中科院软件所
Recommended Citation
GB/T 7714
李基凯. 分布式并行数值代数软件的数据重分布及测试[D]. 中国科学院软件研究所. 中国科学院软件研究所,1998.
Files in This Item:
File Name/Size DocType Version Access License
N98869.pdf(2296KB) 限制开放--Application Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[李基凯]'s Articles
Baidu academic
Similar articles in Baidu academic
[李基凯]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[李基凯]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.