会议专题

A Whole High Performance Benchmark Tuning Process on a Beowulf Cluster

The purpose of this study is to find and validate a practical and workable HPL benchmark tuning process. As the Beowulf architecture becomes popular, the tuning process research on this architecture turns significant and is of great reference. In this work, background knowledge is first introduced and some related works are listed. Then a HPL tuning process is described in detail and implemented on a Dell cluster of Beowulf architecture. A serial of tests in small size are performed to determine MPI. BLAS and so on. Later, four other important parameters—N, NB, P and Q—are tuned. Then other less important parameters are selected. Finally, a fine-tuning step is necessary to further improve the performance. Experimental results show that this tuning process is practical, effective and time-saving.

HPL Beowulf cluster Performance tuning benchmark

Yang Wang Yongquan Lu Pengdong Gao Chu qiu

High Performance Computing Center,Communication University of China,Beijing,100024,China Information High Performance Computing Center,Communication University of China,Beijing,100024,China

国际会议

2011 3rd International Conference on Computer and Network Technology(ICCNT 2011)(2011第三届IEEE计算机与网络技术国际会议)

太原

英文

444-448

2011-02-26(万方平台首次上网日期,不代表论文的发表时间)