3. High Performance Computing Research centers (HPCRCs)

HPCRCs are a cornerstone of the HPCC Program. HPCRCs are the home of a variety of systems: early prototypes, small versions of mature systems, full-scale production systems, and advanced visualization systems. The current production systems, capable of hundreds of gigaflops of sustained throughput, will be succeeded by teraflops systems. These systems are being used on Grand Challenge class applications that cannot be scaled down without modifying the problem being addressed or requiring unacceptably long execution times. The largest of these applications are being run on multiple high performance computers located around the country and networked in the gigabit testbeds.

An interdisciplinary group of experts meets at these centers to address common problems. These include staff from the HPCRCs themselves, hardware and software vendors, Grand Challenge applications researchers, industrial affiliates that want to develop industry-specific software, and academic researchers interested in advancing the frontier of high performance computing. Funding is heavily leveraged, with HPCC agencies often contributing discretionary funds, hardware vendors providing equipment and personnel, and affiliate industries paying their fair share. Industrial affiliation offers a low risk environment for exploring and ultimately exploiting HPCC technology. Two of these industrial affiliations are:

The Oil Reservoir Modeling Grand Challenge in which more than 20 companies and several universities participate
The High Performance Storage System Project in which more than 12 companies and national laboratories participate

Production-quality operating systems and software tools are developed at these centers, thereby removing barriers to efficient hardware use. Applications software tailored to high performance systems is developed by early users, many of whom access these systems over the Internet, and increasingly over the gigabit testbeds, from their workstations. Production-quality applications software is often first run on HPCRC hardware. The wide range of hardware at HPCRCs makes them ideal sites for developing the conventions and standards that enable and test interoperability, and for benchmarking systems and applications software.

Production-quality applications software often is run first on computing systems at HPCRCs.

The major HPCRCs are:

NSF Supercomputer centers --

Cornell Theory center, Ithaca, NY
National center for Supercomputer Applications, Champaign-Urbana, IL
Pittsburgh Supercomputer center, Pittsburgh, PA
San Diego Supercomputer center, San Diego, CA

Tens of thousands of users from more than 800 institutions in 49 states and 111 industrial partners have computed on systems at the NSF centers. Currently there are 8,000 users and 78 partners. The centers are developing a National Metacenter Environment in which a user will view multiple centers as one. The National center for Atmospheric Research (NCAR) in Boulder, CO, also receives HPCC funds.

NSF Science and Technology centers

center in Computer Graphics and Scientific Visualization -- Brown University, Providence, RI; CalTech, Pasadena, CA; Cornell University, Ithaca, NY; University of North Carolina, Chapel Hill, NC; University of Utah, Salt Lake City, UT
center for Research on Parallel Computation, Rice University, Houston, TX

NASA centers --

Ames Research center, Mountain View, CA
Goddard Space Flight center, Greenbelt, MD

DOE centers --

Los Alamos National Laboratory, Los Alamos, NM
National Energy Research Supercomputer center, Lawrence Livermore National Laboratory, Livermore, CA
Oak Ridge National Laboratory, Oak Ridge, TN

The DOE centers accommodate more than 4,000 users from national laboratories, industry, and academia.

Major systems at HPCRCs include one or more of each of the following (the number of processors in the largest machine at an HPCRC is shown in parentheses):

Convex

- C3880 (8 vector processors)
Cray Research

- C90 (16 vector processors)

- T3D (512 processors)

- YMP (8 vector processors)
Digital Equipment Corp.

- Workstation Cluster
Hewlett-Packard

- H-P Workstation Cluster
IBM

- ES9000/900 (6 vector processors)

- PVS

- SP1 (512 processors)

- Workstation Cluster
Intel

- iPSC 860 (64 processors)

- Paragon (512 processors)
Kendall Square Research

- KSR 1 (160 processors)
MasPar

- MasPar 2 (16,000 processors)

- MasPar MP-1 (16,000 processors)
nCube

- nCUBE2
Thinking Machines

- CM2 (32,000 processors)

- CM5 (1,024 processors)

Smaller versions of some of these scalable high performance systems have been installed at more than a dozen universities. The HPCRCs also use a variety of scientific workstations, such as those from Silicon Graphics and Sun Microsystems, for numerous tasks.

FY 1995 Plans

NSF will install new scalable parallel hardware and hardware upgrades, enhance Metacenter resources, and establish several more regional alliances.
DOE will install two different 150 gigaflop machines at two sites.
NASA will establish a prototype high performance computing facility comparable in nature but not in performance to the ultimate teraflops facility. It will be configured with advanced high performance machines, early systems or advanced prototypes of important storage hierarchy subsystems, and sufficient advanced visualization facilities to enable system scaling experiments. NASA Grand Challenge researchers in Federal laboratories, industry, and academia will access these advanced systems using the Internet and gigabit speed networks. These researchers will provide a spectrum of experiments for scalability studies. Prototype systems and subsystem interfaces and protocol standards will be established and evaluated, accelerating the understanding of the character of future teraflops computing systems.
NOAA will acquire a high performance computing system for its Geophysical Fluid Dynamics Laboratory at Princeton, NJ to develop new scalable parallel models for improved weather forecasting and for improved accuracy and dependability of global change models.
EPA will acquire a scalable parallel system to support more complex multipollution and cross-media (combined air and water) impact and control assessments.

Next Section
Table of Contents
Home Page