Simulation Guided Memory Analyzer
A toolkit designed to help programmers understand the precise memory references in scientific programs that are causing poor use of the memory subsystem.
Date Posted: March 17, 2003
|
|
 |
|
|
Update: April 21, 2006
The SIGMA instrumentation engine has been updated to properly handle certain types of Fortran 77 and Fortran 90 applications. Several bugs have been fixed, and performance has been improved.
What is Simulation Guided Memory Analyzer (SIGMA)?
SIGMA is a toolkit designed to help programmers understand the precise memory references in scientific programs that are causing poor use of the memory subsystem. Detailed information such as this is useful for tuning loop kernels, understanding the cache behavior of new algorithms, and investigating how different parts of a program compete for and interact within the memory subsystem.
How does it work? SIGMA provides an infrastructure for analysis of applications' bottlenecks, problems, and inefficiencies that are due to the memory hierarchy. Some of the key features in SIGMA include the following:
- Memory analysis infrastructure and a family of tools for understanding the memory subsystem
- A focus on detailed statistics to complement existing hardware counters
- An ability to handle applications written in Fortran and C
- An infrastructure for asking "what if" questions based on perturbation of data structure and/or architectural parameters that could be helpful in providing directions for improving performance of programs.
The approach taken by the SIGMA performance software includes these steps:
- It instruments the application automatically by implementing a pre-execution tool that locates and instruments all instructions that refer to memory locations.
- It runs the instrumented application to capture complete information about memory use, producing a compact trace.
- It processes the compressed memory reference trace with post-execution simulation and analysis tools, thereby providing programmers with tuning information.
|
|
|
| | About the technology author(s):
Luiz DeRose earned a Ph.D. in computer science from the University of Illinois at Urbana-Champaign. He is the Tools Group Leader at the Advanced Computing Technology Center (ACTC) at IBM Research. Dr. DeRose has more than 15 years of experience in high-performance computing. For the last ten years, he has been working in the area of performance tools and software support for high-performance computation, participating in the design and development of performance tools such as FALCON, SvPablo, and the HPM Toolkit. Before joining IBM, he worked at the National Center for Supercomputing Applications (NCSA), the Pablo group at the University of Illinois, and the Center for Supercomputing Research and Development (CSRD). Dr. DeRose may be reached through e-mail.
Simone Sbaraglia earned a Ph.D. in mathematics at the University of Rome in 2003, with thesis work on "Mathematical Methods and Models for Technology and Society." Dr. Sbaraglia is currently a postdoctoral fellow at the Advanced Computing Technology Center (ACTC) at IBM Research, working on design and development of parallel programing tools. He is the co-architect of the Simulation Guided Memory Analyzer (SIGMA) tools, which he started when working as a co-op student at ACTC. Prior to joining IBM, he worked at the National Research Council, Institute for Applied Computing, in Italy, where he designed and developed a graphical interface for the simulation of the propagation of brakes and the diffusion of chemical polluting in porous materials. In addition, he there conducted research in the design and implementation of prototypes for optimal asset-liability management with constraints for insurance companies, and he also conducted research in efficient implementation of algorithms for the approximation of optimality problems arising in financial theory. Dr. Sbaraglia may be reached through e-mail.
K. Ekanadham is a research staff member in the Parallel Systems Group at IBM T. J. Watson Research Center, Yorktown Heights, N.Y. Dr. Ekanadham's reserach interests include computer architecture and application/system interfaces to machines.
|
|
|
| |
