| 基于属性约简的MPI运行时参数优化 |
| |
| 引用本文： | 朱炜,王云岚. 基于属性约简的MPI运行时参数优化[J]. 计算机与现代化, 2014, 0(1): 14-17,21 |
| |
| 作者姓名： | 朱炜 王云岚 |
| |
| 作者单位： | [1]西北工业大学计算机学院,陕西西安710129 [2]西北工业大学高性能计算中心,陕西西安710072 |
| |
| 摘 要： | 目前MPI实现提供上百种可供调整的运行时参数,但默认的运行时参数不能发挥程序应有的性能。为了给应用程序提供接近最优的运行时参数,本文提出基于粗糙集理论中的属性约简建立优化模型。首先通过属性约简方法为不同基准程序产生参数组合值,再根据未知应用程序的程序特征通过所建立的优化模型为其预测参数值。实验表明,该方法可以有效地为未知应用程序预测参数值,并且比使用默认参数程序性能提升约20%。
|
| 关 键 词： | 粗糙集理论 优化模型 预测 Rosetta |
Optimization of MPI Runtime Parameters Based on Attribute Reduction |
| |
| ZHU Wei,WANG Yun-lan. Optimization of MPI Runtime Parameters Based on Attribute Reduction[J]. Computer and Modernization, 2014, 0(1): 14-17,21 |
| |
| Authors: | ZHU Wei WANG Yun-lan |
| |
| Affiliation: | 1. College of Computer Science, Northwestern Polytechnic University, Xi＇ an 710129, China; 2. High Performance Center, Northwestern Polytechnic University, Xi＇an 710072, China) |
| |
| Abstract: | Currently, MPI implementations provide hundreds of tunable runtime parameters, but the detault parameters can not a- chieve ideal application performance. To provide the near-optimal runtime parameters for applications, attribute reduction in rough set theory is adopted to build an optimization model. First, the method generates values of parameters for different bench- marks through attribute reduction, then, predictes parameter values for unknown input application according to their features based on the model built. Experimental evaluations show the method proposed in this paper can predict values of parameters effec- tively, and achieve approximate 20% performance enhancement. |
| |
| Keywords: | rough set theory optimization model prediction Rosetta |
| 本文献已被 CNKI 维普 等数据库收录！ |
| 点击此处可从《计算机与现代化》浏览原始摘要信息 |
|
点击此处可从《计算机与现代化》下载全文 |
|
