CI Evaluation Center

Overview

Very large computing systems have contributed enormously to answering the big questions in science and engineering; so much so that further breakthroughs in science and engineering will depend on continuing increases in computational ability and easily usable and accessible high-end systems. The National Science Foundation (NSF) Cyberinfrastructure Evaluation Center, led by researchers at RENCI and the San Diego Supercomputer Center (SDSC), uses a wide range of analysis tools to predict and evaluate the effectiveness of innovative and future systems in addressing scientific problems of interest to the National Science Foundation. The project seeks to advance the development of cyberinfrastructure by creating and validating performance models of strategic NSF computational science applications and by assessing the interdependence of alternative hardware, middleware and software implementations on application performance. Its fundamental goal is to develop and deploy techniques that will that will help scientists better understand their applications and help computer scientists develop systems better able to accommodate applications. In the long run, this should allow faster and better scientific simulations that enable new discoveries in many disciplines.

Researchers at RENCI SDSC are characterizing the existing NSF supercomputing application workload. The models that are being developed will then be used to evaluate alternative configurations of new systems, starting with SDSC’s Blue Gene/L and the Pittsburgh Supercomputing Center’s Red Storm. The Evaluation Center will then extend its modeling methodology to encompass additional systems and to assess the performance of grid-enabled applications. In addition, the Evaluation Center will develop the following:

  • Application signatures of strategic applications—these are high-level representations of an application’s behavior and resource requirements.
  • Machine profiles of emerging systems—these are high-level representations of the rates that systems can perform basic operations on behalf of applications.
  • Performance models of the interactions between applications and systems, identifying the performance sensitivity of applications to changes in architecture, algorithm, or implementation
  • Recommendations based on the models, quantifying the system configurations best suited to the application workload, the applications that have performance affinity to particular systems, and what middleware and/or application code tuning would reduce the time-to-solution.

Funding

National Science Foundation

Project Leaders

  • Allan Snavely, San Diego Supercomputer Center
  • Rob Fowler, RENCI

RENCI Participants

  • Todd Gamblin
  • Allan Porterfield
  • Dan Reed (Original PI)
  • Jeff Tilson
  • Ying Zhang

Partners

  • RENCI
  • San Diego Supercomputer Center

Links

National Science Foundation
San Diego Supercomputer Center