Building and Matching Applications to NRL Supercomputers

Author

Rosenberg, Robert ; Bique, Stephen ; Anderson, Wendell ; Lanzagorta, Marco

Author_Institution

US Naval Res. Lab. (NRL-DC), Washington, DC, USA

fYear

2009

fDate

15-18 June 2009

Firstpage

479

Lastpage

484

Abstract

We describe our experiences in building and executing codes from the Department of Defense (DoD) High Performance Computing Modernization Program (HPCMP) TI-09 Benchmark suite on the Naval Research Laboratory (NRL) HPC supercomputers. These include a 256 node SGI Altix 3000, an 864 core Cray XD1 and a 1,536 core SGI Altix ICE. These platforms constitute a variety of architectures where we can compare code performance based on commodity and custom components, including processors, interconnects, memory sizes and disk input/output (I/O) rates. Based on this code performance, we provide recommendations for compiler and linker options as well as a basis for determining which platform is best suited for a particular code.

Keywords

mainframes; naval engineering computing; parallel machines; program compilers; 864 core Cray XD1; Department of Defense; HPC supercomputer; NRL supercomputer; Naval Research Laboratory; SGI Altix 3000; SGI Altix ICE; building codes; commodity component; custom component; disk input-output rates; executing codes; high performance computing modernization program TI-09 benchmark; memory size; Benchmark testing; Computational fluid dynamics; Computer architecture; Ice; Mathematical model; Program processors; US Department of Defense;

fLanguage

English

Publisher

ieee

Conference_Titel

DoD High Performance Computing Modernization Program Users Group Conference (HPCMP-UGC), 2009

Conference_Location

San Diego, CA

Print_ISBN

978-1-4244-5768-7

Type

conf

DOI

10.1109/HPCMP-UGC.2009.79

Filename

5729437