[an error occurred while processing this directive]

NERSC 3 Greenbook

Next: User Requirements for the Up: User Requirements for the Energy Research Computational Community Previous: List of Figures

Executive Summary

The purpose of this report is to document the computational science being done at the High Performance Computing Access Centers - especially the National Energy Research Scientific Computing Center (NERSC) - of the U. S. Department of Energy, Office of Energy Research. NERSC and the other centers provide scientists from the Office of Energy Research with a very high-level shared computing facility, including at this time a massively parallel computer, a cluster of vector multiprocessors, a mass storage system, visualization facilities, a user support organization, networking and software. This report includes discussions of the ways in which the HPCAC's should evolve in order to meet the needs of Energy Research scientists to continue to do forefront science.

A large shared computing center continues to be essential to the scientific mission of the Office of Energy Research as it provides the researchers with resources that could not be supplied locally. We document this role (Chapter 3) by providing samples of computational applications performed at NERSC. Brief descriptions of current and/or future scientific efforts are presented from each of the five major offices within the Office of Energy Research: High Energy and Nuclear Physics, Basic Energy Sciences, Fusion Energy Science, Biological and Environmental Research, and Computational and Technology Research.

The computing hardware normally needed at the forefront of scientific research has evolved from the ``vector supercomputers'' of the late 1970's and 1980's to presently a mixture of massively parallel computers employing distributed memory, symmetric multiprocessors employing shared memory, and vector multiprocessors. We describe (Chapter 4) these architectures and their anticipated near-term evolution, and document the importance of each of these architectures to the computational needs of Energy Research. The massively parallel machines, based on commercially mass produced chips, can offer extremely large memory at the most attractive price-performance ratio. The vector multiprocessors still offer the fastest turn-around time for many of the scientific codes. Symmetric multiprocessors (SMPs) and clusters of SMPs are developing to provide the same advantages of massively parallel machines and provide more flexibility in use at the same time. Mass storage, visualization facilities and networks, which are essential complements to raw CPU power, are briefly described with a discussion of the future needs in these areas for Energy Research. Nothing useful happens in a computer without good software, and we document (still Chapter 4) our requirements for base software, for mathematical and applications libraries, and especially for a unified production environment that supports distributed computing across the high performance computing centers, the Energy Research laboratories, and affiliated university groups. We also discuss research and services at the computing centers.

Based on the presentation of computing endeavors within Energy Research, and of the hardware and software environment at the Centers and as available commercially now or anticipated, we make these specific recommendations:

• Expand the high performance computing resources available at NERSC. The procurement (already underway) must ensure an appropriate balance of computational equipment to support the wide range of large-scale applications that use the current facilities.
• Focus efforts to convert appropriate applications to parallel systems. To meet this requirement, NERSC must have a highly qualified and diverse parallel computational science and applications group to work with the core Energy Research computational efforts.
• Strengthen the networks. Reliable, robust and fast access to the computational resources is essential.
• Encourage collaborations and integration among ER physical, computational, and computer science programs. NERSC can act as a collaboratory meeting place for Energy Research scientific efforts to form ``virtual laboratory groups.''
• Encourage software development for the advancement of Energy Research missions. NERSC should serve as a facilitator and where appropriate as partner in these activities.
• Encourage large-scale applications by enhancing the effective use of the Capability computing systems. Supercomputers should be used primarily for the applications that require the specific and special capabilities of the resource. NERSC should support both Capability and Capacity computing, and is in a unique position to foster the development of a unified production environment for distributed computing.
• Encourage expansion of data intensive computing capabilities. This is an emerging area for many experimental efforts collecting, storing, refining and analyzing large (multi-terabyte) data sets.

NERSC 3 Greenbook

Next: User Requirements for the Up: User Requirements for the Energy Research Computational Community Previous: List of Figures