Parallelization of pipelined algorithms for sets of linear banded systems

被引:10
|
作者
Povitsky, A [1 ]
机构
[1] NASA, Langley Res Ctr, ICASE, Hampton, VA 23681 USA
基金
美国国家航空航天局;
关键词
Thomas Algorithm; linear banded systems; pipelined algorithms; parallel computations; implicit numerical methods; MIMD computer;
D O I
10.1006/jpdc.1999.1568
中图分类号
TP301 [理论、方法];
学科分类号
081202 ;
摘要
Solution of independent sets of linear banded systems is a core part of implicit numerical algorithms. In this study we propose a novel pipelined Thomas algorithm with low parallelization penalty. We introduce two-step pipelined algorithms (PAs) formally and show that the idle processor time is invariant with respect to the order of backward and forward steps, Therefore. the parallelization efficiency of the PA cannot be improved directly. However. the processor idle time can be used if some Lines have been computed by the time processors become idle. We develop the immediate backward pipelined Thomas algorithm (IB-PTA). The backward step is computed immediately after the forward step has been completed for the first portion of lines. The advantage of the IB-PTA over the basic PTA is the presence of solved lines, which are available for other computations, by the time processors become idle. Implementation of the IB-PTA is based on a proposed static processor schedule that switches between forward and backward computations and controls communication between processors, Computations are performed on the Gray T3E MIMD computer. Combination of the proposed IB-PTA with the "burn from two ends" algorithm shows low; parallelization penalty. (C) 1999 Academic Press.
引用
收藏
页码:68 / 97
页数:30
相关论文
共 50 条
  • [2] Direct parallel algorithms for banded linear systems
    Arbenz, P
    Gander, W
    ZEITSCHRIFT FUR ANGEWANDTE MATHEMATIK UND MECHANIK, 1996, 76 : 119 - 122
  • [3] PARALLEL ALGORITHMS FOR BANDED LINEAR-SYSTEMS
    WRIGHT, SJ
    SIAM JOURNAL ON SCIENTIFIC AND STATISTICAL COMPUTING, 1991, 12 (04): : 824 - 842
  • [4] Direct parallel algorithms for banded linear systems
    Arbenz, P.
    Gander, W.
    Zeitschrift fuer Angewandte Mathematik und Mechanik, ZAMM, Applied Mathematics and Mechanics, 76 (Suppl 1):
  • [5] Decomposition and Parallelization of Linear Programming Algorithms
    Karbowski, Andrzej
    PROGRESS IN AUTOMATION, ROBOTICS AND MEASURING TECHNIQUES: CONTROL AND AUTOMATION, 2015, 350 : 113 - 126
  • [6] Fast and stable algorithms for banded plus semiseparable systems of linear equations
    Chandrasekaran, S
    Gu, M
    SIAM JOURNAL ON MATRIX ANALYSIS AND APPLICATIONS, 2003, 25 (02) : 373 - 384
  • [7] Parallelization of the Kalman Filter for Banded Systems on Multicore Computational Platforms
    Rosen, Olov
    Medvedev, Alexander
    2012 IEEE 51ST ANNUAL CONFERENCE ON DECISION AND CONTROL (CDC), 2012, : 2022 - 2027
  • [8] Parallelization of Algorithms for Linear Discrete Optimization using ParaPhrase
    Rossbory, Michael
    Reisner, Werner
    2013 24TH INTERNATIONAL WORKSHOP ON DATABASE AND EXPERT SYSTEMS APPLICATIONS (DEXA 2013), 2013, : 241 - 245
  • [9] Concurrent Separation Logic for Pipelined Parallelization
    Bell, Christian J.
    Appel, Andrew W.
    Walker, David
    STATIC ANALYSIS, 2010, 6337 : 151 - 166
  • [10] PARALLELIZATION OF WHILE LOOPS ON PIPELINED ARCHITECTURES
    TIRUMALAI, PP
    LEE, M
    SCHLANSKER, MS
    JOURNAL OF SUPERCOMPUTING, 1991, 5 (2-3): : 119 - 136