Science-Driven System Architecture

NERSC's mission includes providing the largest-capability production computing resources to a broad array of scientific disciplines. The Science Driven Systems Architecture group combines the resources of NERSC with the Computational Research Division's Future Technologies group and Scientific Computing Group to better understand the requirements of the NERSC workload, assess emerging system technologies, and use the NERSC workload requirements to drive changes in computing architecture that will result in better HPC system architectures for scientific computing in future generation machines.

One of the primary tasks of this group is to continually assess available HPC system solutions using a combination of application benchmarks and microbenchmarks. In addition, the team is working in the following areas.

• Improve the selection process for new systems through stewardship of NERSC SSP benchmarks.
• Establish a systematic process for using workload characterization data to select future SSP benchmarks.
• Improve understanding of the NERSC workload through development of non-invasive data collection tools and data mining methods for extracting useful information from accumulated workload data.
• Apply advanced performance modeling methods to characterize system performance, application requirements, and to synthesize predictive models using the convolution of the application and system signatures
• Work together with the NERSC User Services Group to refine "best practices" in parallel programming and I/O techniques to fully exploit the performance potential of current systems.
• Encourage vendors to design systems that are better suited to scientific computation by supplying them with better analysis of algorithm requirements (current and projections of future requirements), system bottleneck analysis, and better understanding of the NERSC workload.