Scientific Computing Corner:

The Strategic Applications Collaboration (SAC) and Strategic Community Collaboration (SCC) Program at SDSC - An Overview

—Amit Majumdar

Introduction

The Strategic Applications Collaborations (SAC) and the Strategic Community Collaborations (SCC) programs have been in place at SDSC since 1999 and allow SDSC's computational science experts to work closely with computational science researchers for a longer time period that ranges from a few months to a year. The goal of the SAC and SCC programs is to enable users' computational science research in a significant way by utilizing SDSC's hardware, software, and human resources. The SAC and SCC programs span high performance computational, data, and visualization science projects utilizing SDSC resources. As a part of the Cyberinfrastructure Partnership (CIP), SDSC staff also collaborate with NCSA staff on various projects.

SAC/SCC Selection Process

All SAC/SCC projects have the potential for enabling scientific capability for the user. In addition, in order to qualify for the SAC/SCC program, a project must combine some of these criteria:

• Fit well with SDSC's vision of the HPC+Data strategy, such as

  • Produce, use, and share large amounts of data and/or data collections

  • Utilize the whole pipeline of compute, data, and visualization resources and human expertise at SDSC

• Have the highest possibility of concrete and specific success and a well defined roadmap

• Be led by PIs who are interested in

  • close collaboration with SDSC SAC/SCC staff and other collaborators from SDSC
  • incorporating contributions from SAC/SCC staff in their production codes

• Lead to work with community codes and software packages

• Involve new communities such as social sciences or economics

• Span across all the NSF directorates as well as other government research funding agencies such as NIH, DOE, or NASA

Example of SAC and SCC projects

Over the past years, SDSC computational experts have worked on SAC and SCC projects in various disciplines such as astronomy, biochemistry, bioinformatics, chemical engineering, chemistry, civil engineering, climate science, geosciences, imaging science, linguistics, mechanical engineering, neuroscience, nuclear medicine, and space physics. In addition to the above projects, SDSC's computational experts continuously gather single processor, parallel scaling, and I/O performance results of various micro benchmarks, benchmark kernels, and applications on SDSC resources. These results provide guidelines to current and future users regarding choice of HPC resources for their applications.

Geoscience SCC: SDSC staff collaborated with researchers from the Southern California EarthquakeConsortium (SCEC) to perform a large scale simulation of a 7.7 magnitude seismic wave propagation on the San Andres fault. Detailed single processor and parallel performance analysis, MPI I/O implementation, and visualization were done as a part of this effort. This simulation ran on 240 processors of the IBM DataStar machine for five days consuming 20,000 hours and generating about 50 terabytes of output data which was used for visualization.

Such simulations provide potentially immense benefits in saving both many lives and billions in economic losses Tom Jordan University of Southern California Director of SCEC

Turbulence SAC: SDSC staff collaborated with Prof. PK Yeung of Gerogia Tech towards the goal of performing Direct Numerical Simulation (DNS) on 4096³ grid. A parallel 3D FFT module with 2D domain decomposition was developed, and the scalability is now being tested. The resultant library of 3D FFT with 2D decomposition will be made available for general users. Effort is also underway to port and test scalability of this on SDSC's BlueGene machine.

Turbulent fluctuations in this fluid jet promote efficient mixing. Large simulations by P.K. Yeung on SDSC's DataStar are giving new understanding of the complex mechanisms of mixing and contaminant dispersion, including for slow-diffusing substances that have been difficult to study. Image courtesy of K.R. Sreenivasan, U Maryland.

Bioinformatics SAC: SDSC staff collaborated with Prof. David Baker and his research group from U. Washington. The Rosetta code is a prominent protein structure prediction code. SDSC staff parallelized the code using posix-compliant file locking, and the resulting code scales to a large number of processors. This code is now running on both SDSC and NCSA resources consuming millions of hours. Currently, effort is underway to improve single processor performance of this code.

Image source: http://depts.washington.edu/bakerpg

For more information about the SAC and SCC programs, contact Amit Majumdar via e-mail at majumdar@sdsc.edu.