Faculty Interaction

Prof Rudi Eigenmann

School of Electrical and Computer Engineering
and Interim Director, Computing Research Institute
Purdue University

Brief Project Description:
Evaluation of High Performance Computing platforms using realistic application codes. In particular, the SPEC HPC suite is being evaluated on parallel computers at ITaP.

Project: HPC Benchmarking and Performance Evaluation With Realistic Applications With rapid developments and changes in the High Performance Computing (software and hardware) benchmarking and performance evaluation is a critical research area. The goal of this research study is to assess the performance and understand characteristics of HPC platforms and their important applications. The study will help in the creation of yardsticks for research and the anticipation of needs of future HPC technology. We would like to address some of the important questions such as:

1) How do HPC platforms perform in solving today's important problems?
2) What are the characteristics of today's computational applications and how do these characteristics change in future problems?

In addition to using the SPEC HPC benchmarks (WRF, SEISMIC and GAMESS) the study will also make use of other real application benchmarks from the NSF real application benchmark suite and any other scientific applications meeting certain criterion for an acceptable benchmark. We will also make use of low level kernel benchmarks (SPEC CPU, STREAM, BLAS, MPI, HPC Challenge etc.,) and synthetic benchmarks such as NAS parallel benchmarks to understand full application performance. The study will look at the rankings of various platforms using Kernel benchmark (such as HPL) and the real applications.

References:

HPC Benchmarking and Performance Evaluation With Realistic Applications Brian Armstrong, Hansang Bae, Rudolf Eigenmann, Faisal Saied, Mohamed Sayeed, Yili Zheng. The paper was presented at the SPEC 06 conference.

Measuring HPC Performance With Real Applications Brian Armstrong, Hansang Bae, Rudolf Eigenmann, Faisal Saied, Mohamed Sayeed, Yili Zheng. The paper was submitted to the IEEE magazine Computing in Science and Engineering (CiSE).