An Efficient Data Distribution Technique for Distributed Memory Parallel Computers
スポンサーリンク
概要
- 論文の詳細を見る
For programming massively parallel distributed memory machines, selecting an appropriate data decomposition is of critical importance to the performance of the data-parallel program for these machines. The important considerations for a good data distribution pattern are that it should maximize the system performance by balancing the computational load and by minimizing remote memory accesses. Most of the existed data-parallel language present the regular distribution, i.e. block, cyclic or block-cyclic distribution, as in Fortran D, Vienna Fortran and HPF. In this paper, we propose a technique, called linear data distribution, which partitions array onto processors according to parallel lines, such as rows, columns or slant lines, that reduce the communication overhead. We determine array distribution pattern based on analysis of the array subscript references in loop nests.
- 一般社団法人情報処理学会の論文
- 1997-03-12
著者
-
Yamashita Yoshiyuki
Department Of Chemistry Faculty Of Engineering Science And Research Center For Photoenergetics Of Or
-
Yamashita Yoshiyuki
Institute Of Information Sciences And Electronics University Of Tsukuba
-
Guo M
The School Of Computer Science And Engineering University Of Aizu
-
Guo Minyi
Doctoral Program In Engineering University Of Tsukuba
-
Nakata Ikuo
University Of Library And Information Science
-
Yamashita Yoshiyuki
University Of Tsukuba
-
GUO Minyi
University of Tsukuba
関連論文
- Spectroscopic and Theoretical Studies of Interface States at Ultrathin Oxide/Si Interfaces
- Interface States at Ultrathin Chemical Oxide/Silicon Interfaces Obtained from Measurements of XPS Spectra under Biases
- Interface States for Si-Based MOS Devices with an Ultrathin Oxide Layer : X-Ray Photoelectron Spectroscopic Measurements under Biases
- Message Scheduling for Irregular Data Redistribution in Parallelizing Compilers(Parallel/Distributed Programming Models, Paradigms and Tools, Paradigms and Tools, Parallel/Distributed Computing and Networking)
- A Parallel Implementation of Multi-Domain High-Order Navier-Stokes Equations Using MPI(Scientific and Engineering Computing with Applications)(Hardware/Software Support for High Performance Scientific and Engineering Computing)
- New Method for Observation of Interface States in the Semiconductor Band-Gap : XPS Measurements under Biases(Interfaces by various techniques)
- Direct Spectroscopic Evidence of Bias-Induced Shifts of Semiconductor Band Edges for Metal-Insulator-Semiconductor Diodes
- Efficient Implementation of Multi-Dimensional Array Redistribution
- An Efficient Data Distribution Technique for Distributed Memory Parallel Computers (特集:並列処理)
- An Efficient Data Distribution Technique for Distributed Memory Parallel Computers
- Efficient Loop Partitioning for Parallel Codes of Irregular Scientific Computations(Software Systems)
- Optimization Techniques for Parallel Codes of Irregular Scientific Computations(Code Generation and Optimization)
- Negation Technique for Context-Free Grammars