PERCs
High Performance Computing (HPC) shares many aspects with sequential programming. Where it differs is in scale, parallelization, and problem difficulty. While a niche field there is enormous variability in terms of organizations, resources, types of coding and problems, and lifespan of code.
PERCS stands for Productive Easy to Use Robust Computing system. It is the project name for IBM’s effort in the DARPA HPCS program challenge to build a peta-scale computer by 2010. Our goal is to understand human behavior with parallel systems—particularly parallel programming—with an eye to improving programmer’s work with tools. In addition this needs to be measured from a baseline of 2002 equipment and competencies.
We have approached this problem in a number of ways: fieldwork, empirical studies, and complexity metrics.
Field Work: In 2005 we collected data from Lawrence Livermore National Lab (LLNL), Pittsburgh Supercomputing Center (PSC), and inside IBM. In 2007 we collected data from the Department of Energy’s National Energy Research Center for Scientific Computing (NERSC). In 2008 we hope to expand to other users of high performance computing systems.
Empirical Studies: In 2005, working with the PSC, we conducted a study comparing language uptake and effectiveness with student programmers. The three languages were C+MPI, UPC, and X10. Currently we are working on establishing the 2002 baseline for comparison to new tools.
Complexity Metrics: Need Catalina to do this… or refer to a separate page.
For more information
About our productivity efforts on PERCS contact: Christine Halverson krys@us.ibm.com or Catalina Danis danis@us.ibm.com
If you’d like to read more about the PERCS project overall see:
http://domino.research.ibm.com/comm/pr.nsf/pages/news.20030710_darpa.html
If you’d like to read about our first study for PERCS see:
Danis, C. and Halverson C.A. (Feb 2006) The value derived from the Observational Component in an Integrated Methodology for the study of HPC Programmer Productivity. Workshop on Productivity and Performance in High-End Computing. 12th International Symposium on High-Performance Computer Architecture, Austin TX.
Other papers about Productivity and PERCS
Danis, C. (Nov 2006) Forms of collaboration in high performance computing: exploring implications for learning. Computer Supported Cooperative Work (CSCW 06), Banff, Alberta, Canada. pp 501-504
