NASA STTR 2008 Solicitation


PROPOSAL NUMBER: 08-1 T5.01-9954
RESEARCH SUBTOPIC TITLE: Benchmark Numerical Toolkits for High Performance Computing
PROPOSAL TITLE: Benchmark Numerical Toolkits for High Performance Computing

NAME: CFD Research Corporation NAME: The University of Texas at Austin
STREET: 215 Wynn Drive, 5th Floor STREET: 10100 Burnet Road
CITY: Huntsville CITY: Austin
STATE/ZIP: AL  35805 - 1944 STATE/ZIP: TX  78758 - 4445
PHONE: (256) 726-4800 PHONE: (512) 471-5809

PRINCIPAL INVESTIGATOR/PROJECT MANAGER (Name, E-mail, Mail Address, City/State/Zip, Phone)
Ravi Kannan
215 Wynn Drive, 5th Floor
Huntsville, AL 35805 - 1944
(256) 726-4858

Expected Technology Readiness Level (TRL) upon completion of contract: 5

TECHNICAL ABSTRACT (Limit 2000 characters, approximately 200 words)
Computational codes in physics and engineering often use implicit solution algorithms that require linear algebra tools such as Ax=b solvers, eigenvalue, optimization, or graph algorithms. Developers face major challenges in selecting linear algebra tools that can support their algorithms, numerical schemes, meshes, and computing hardware and to minimize the time, space and complexity. The existing libraries such as PETSc or LAPACK are "stretched" to the limits by new generation application codes which create big, unsymmetric, often dense, and poorly conditioned matrices. One of the obstacles in effective utilization of linear algebra libraries is lack of benchmark quality representative test cases and benchmarking toolkits for these types of problems. This project will develop, demonstrate and deliver a comprehensive numerical test suite for benchmark evaluation of linear algebra solvers for computational application software on High Performance Computers. Unlike existing benchmarks on static, Ax=b matrix, problems CFDRC-TACC team proposes new generation of dynamic, discipline specific and multidisciplinary functional benchmarks accounting for sparse/dense and unsymmetric matrices using web accessible benchmark matrix/problem generators. Our team has excellent expertise and tools (multiphysics solvers, sparse/dense solver libraries, benchmark cases, related projects, and understanding of NASA engineering/scientific challenges) as well as HPC resources to achieve this goal.

POTENTIAL NASA COMMERCIAL APPLICATIONS (Limit 1500 characters, approximately 150 words)
NASA scientists and engineers are involved in the development of advanced computational codes that utilize linear algebra libraries for applications in physics, engineering, biotechnology, information processing and other disciplines using parallel computing. This project will develop and deliver tools for automated testing and benchmarking of codes in fluid dynamics (CFD), structures mechanics (FEM), heat transfer and thermal exchange, electrostatics, electromagnetics image processing, and other codes. These tools will cut cost and time of maintenance of existing NASA software, help porting legacy tools to new parallel computers, and aid the development of next generation scientific and engineering computational tools.

POTENTIAL NON-NASA COMMERCIAL APPLICATIONS (Limit 1500 characters, approximately 150 words)
Linear algebra libraries have a pivotal role in scientific software, as they often dictate the programming style and structure, take a major portion of the memory and CPU time (often > 90%) and require frequent and rigorous testing/benchmarking each time a new algorithm is implemented or new parallel computing hardware is used. One of the most cumbersome and time consuming tasks is the benchmarking of coupled software-library on related test problems on serial and parallel computers. The benchmarking toolkits and technical services established in this project will help developers and users of computational software in the development, evaluation and effective utilization of the software on parallel computing hardware.

NASA's technology taxonomy has been developed by the SBIR-STTR program to disseminate awareness of proposed and awarded R/R&D in the agency. It is a listing of over 100 technologies, sorted into broad categories, of interest to NASA.

Computer System Architectures
Database Development and Interfacing
Software Development Environments
Software Tools for Distributed Analysis and Simulation

Form Generated on 11-24-08 11:59