. The challenge series defines scientific computation problems without dictating program code or algorithms, thus maximizing freedom of implementation as architectures, languages, and methods evolve over the generations.</li>
<li>Implemented template versions of subset of EPB 1 working set codes. The templates methodology has evolved as a consequence.</li>
<li>Tested benchmarks on multiple parallel platforms and collected execution statistics. </li>
<li>Initial industrial contacts for testing EPB 1.</li>
</ul>
</p><p>
<p>Significance: The EPB 1 benchmark suite offers the first comprehensive set of controlled codes for detailed evaluation of parallel computing systems as they relate to the computational requirements of the Earth and space sciences community. The templates methodology being developed to specify these codes will be the first ever applied to wide-scale benchmarking and will achieve unequaled portability across diverse parallel architectures while sustaining control of application workload. The EPB suites will permit meaningful comparative analysis for future procurement decisions and parallel architecture research.
</p><p>
Status/Plans: Refine templates methodology and complete template version of working set codes. Work with industrial partners to advance benchmark suite as an important tool for comparative studies of parallel systems. Employ a subset of the EPB 1 working set for the ESS CAN. Release the industrial-grade EPB 2 to the HPC community.
</p><p>
</p>
<hr>
<p>Point of Contact:
<ul>
Dr. Thomas Sterling
<br><a href=)
Center of Excellence in Space Data and Information Sciences (CESDIS)
Goddard Space Flight Center/Code 930.5
tron@chesapeake.gsfc.nasa.gov
(301) 286-2757
Objective: Develop a benchmark set for parallel computing systems that will reflect the computational needs of the Earth and space science community.
Approach: Determine the range of disciplines comprising the ESS computational community and identify key applications that reflect their computing requirements. Develop a benchmarking methodology that achieves portability across parallel platforms while retaining control of essential computing demand. Enable multigeneration tracking of HPC architecture impact on Earth and space science. Engage in an extensive in-house testing period prior to release to general community. Incorporate inline software instrumentation to generate metrics of performance (see picture, 150K).
Center of Excellence in Space Data and Information Sciences (CESDIS)
Goddard Space Flight Center/Code 930.5
tron@chesapeake.gsfc.nasa.gov
(301) 286-2757
Return to the PREVIOUS PAGE