#### DMCA

## Using Computational Similarity to Analyze the Performance Data of the NAS Parallel Benchmarks (1995)

Citations: | 1 - 1 self |

### Citations

10 | NAS parallel benchmarks results 10-94
- Bailey, Barszcz, et al.
- 1994
(Show Context)
Citation Context ...rks have been developed at the NASA Ames Research Center. In the last three years extensive performance data has been reported for parallel machines based on the NAS Parallel Benchmarks (NAS PB) [5], =-=[6]-=-. But unfortunately we have tostay with the table of the results without being able to reduce the amount of information. This could be done by introducing a limited number of parameters which hopefull... |

9 |
editors). The NAS parallel benchmarks
- Bailey, Barton, et al.
- 1993
(Show Context)
Citation Context ...nchmarks have been developed at the NASA Ames Research Center. In the last three years extensive performance data has been reported for parallel machines based on the NAS Parallel Benchmarks (NAS PB) =-=[5]-=-, [6]. But unfortunately we have tostay with the table of the results without being able to reduce the amount of information. This could be done by introducing a limited number of parameters which hop... |

6 |
Amdahl’s law and the statistical content of the NAS parallel benchmarks
- Simon, Strohmaier
- 1995
(Show Context)
Citation Context ...mation. This could be done by introducing a limited number of parameters which hopefully will be characteristic for the system and the algorithms e.g. the hard- and software used. In previous studies =-=[2]-=-, [3] we showed that there is strong statistical evidence that only four of the eight NAS Parallel Benchmarks are independent of each other. This is 1san indication that a more condensed description o... |

2 |
Strohmaier: Statistical Analysis of
- Simon, E
- 1995
(Show Context)
Citation Context ...n. This could be done by introducing a limited number of parameters which hopefully will be characteristic for the system and the algorithms e.g. the hard- and software used. In previous studies [2], =-=[3]-=- we showed that there is strong statistical evidence that only four of the eight NAS Parallel Benchmarks are independent of each other. This is 1san indication that a more condensed description of the... |

2 | Extending the Concept of Computational Similarity for Analyzing Complex Benchmarks
- Strohmaier
- 1995
(Show Context)
Citation Context ...ney showed a possible way toexpress the scaling behaviour of a whole class of applications by the same set of parameters. Along the line of his concept we developed a generalized version of his ideas =-=[4]-=-. We did this in a way which, in our opinion, is especially well suited to study the timing behaviour of complex applications. The method is also usable if analytical timing relations are not deducabl... |

2 |
Amdahl: Validity of the single processor approach to achieve large scale computing capabilities
- M
- 1967
(Show Context)
Citation Context ...previous chapter. However, we assume that their arguments might be positive and that these cases contribute to the times measured. 4 Fitting Amdahl's Law to the NAS PBs For deriving his well known law=-=[9]-=-, Amdahl used only two kinds of works namely serial and parallel. This gives the timing relation T0 1 (N� p) = 1 + 2 p (31) with the additional condition 1 + 2 =1. Identifying 2 as parallelization rat... |

1 |
Computational Similarity
- Hochney
- 1994
(Show Context)
Citation Context ... NAS Parallel Benchmarks are independent of each other. This is 1san indication that a more condensed description of the measured data should be possible. With his concept of computational similarity =-=[1]-=-Roger Hockney showed a possible way toexpress the scaling behaviour of a whole class of applications by the same set of parameters. Along the line of his concept we developed a generalized version of ... |

1 |
Getov: Performance Characterisation of the Cache Memory E ect
- S
- 1994
(Show Context)
Citation Context ... >dcrit (22)swith u 1 ca = 1 p2 and u 2 ca = 1 p Here again a single operation in uences the i of di erent characteristic functions ui. A more realistic description of cache behaviour may be found in =-=[8]-=-. 3.13 Overlap between communication and computation Here we look at the communication work which can only partially be overlapped with computational work depending on the problem size per processor. ... |