[Updated on Apr. 18] Integration of CiNii Articles into CiNii Research


Bibliographic Information

Other Title
  • スウチ ケイサン ノ タメ ノ ヘイレツ ケイサンキ セイノウ ヒョウカ ホウ
  • A Performance Evaluation Method of Parallel Computers for Numerical Analysis
  • 高性能計算

Search this article



A performance evaluation method of parallel computers for numerical analysis which enables to predict the performance consistently from a program level to a loop level is offered.The character of a processing time model of loops used this method is that the loop level performance evaluation is executable by introducing model parameters including normalized efficiency by theoretical peak performance flop/s etc.In case of a program level the evaluation is carried out by using both the model and the model parameters decided by actual measurement of time.As an example,this performance evaluation method is applied for molecular dynamics program parallelized with the particle decomposition approach,and the performance of vector parallel computer VPP300 is evaluated.As a result,it is shown that the performance of a program level depends on shielding distance and observation times of physical variables in addition to the number of particles and the number of processors.The performance of a program level is greatly improved when the number of particles increases because the most time-consuming loop in case of a small number of particles is exchanged for another loop which uses processing units efficiently when the particles increase.This performance evaluation method allow us performance evaluation consisting of the three view points which are use of computer,development of program anddesign of computer.


Citations (3)*help

See more


See more

Related Articles

See more

Related Data

See more

Related Books

See more

Related Dissertations

See more

Related Projects

See more

Related Products

See more



Report a problem

Back to top