ADVANCED COMPUTATIONAL INFRASTRUCTURE AND RESEARCH
$80,220,000

The FY 2002 Budget Request for the Advanced Computational Infrastructure and Research (ACIR) Subactivity is $80.22 million, a decrease of $1.38 million, or 1.7 percent, from the FY 2001 Current Plan of $81.60 million.

(Millions of Dollars)

Totals may not add due to rounding.

The ACIR Subactivity provides access to, and support for, high-end computing for the national scientific community, and research on the development, use and applications of these computing systems. Within Advanced Computational Infrastructure (ACI), FY 2002 will be the fourth full year for the Partnerships for Advanced Computational Infrastructure (PACI) Program, which replaced the NSF Supercomputer Centers Program in FY 1998.

PACI consists of two partnerships, each consisting of a leading edge site and a number of partners. More than 60 geographically distributed partner institutions from 27 states and the District of Columbia are associated with PACI. The leading edge sites maintain a variety of high-end computer systems and together with their partners, they develop, apply and test necessary software, tools, and algorithms to further the growth of a set of interconnected resources consisting of advanced visualization and data handling capabilities linked with high-end computing capabilities. The Terascale Computing Systems, described in the MRE section, will be linked to the PACI centers.

PACI activities include:

• Access - making available a diverse set of advanced and mid-range compute engines, data storage systems, and experimental machine architectures.

• Enabling Technologies - developing parallel software and computation tools to enable effective exploitation of the partnerships' widely distributed, architecturally diverse, machines and data sources.

• Application Technologies - developing and optimizing discipline-specific codes and software infrastructures, making these broadly available to researchers.

• Education, Outreach and Training - ensuring awareness and understanding of how to use high-end computing and communications resources, and broadening participation in advanced computational science and engineering.

In FY 2002, funding for the ACI line item will be held at $73.71 million. Of this amount, $3.0 million will provide operations support for the two Terascale Computing Facilities. These funds will support operational costs of integrating the Terascale facilities with the existing PACI facilities. The Terascale Computing Systems are described under the Major Research Equipment account.

Advanced Computational Research (ACR) complements PACI activities through single-investigator or small-group research grants to advance the state of the art in high-performance computation. It has three principal technical thrusts: visualization, data handling, and parallel numerical algorithms. In FY 2002, ACR will decrease by $1.38 million. The impact on researchers will be partly offset by related projects funded in the ITR program.

ACIR-funded advances include Globus and Legion, two middleware components that play an increasing role in building computational, information, and access grids. Grids are connected resources that enable researchers to access the best resources over networks without complex human brokering for resources or requiring users to adapt data or software to unfamiliar computing environments. Such grids are developing methods for security and privacy, distributed storage, grid measurement and other techniques to enable seamless and efficient access to resources.

Researchers at the PACI partnerships are creating innovative ways to harness computing power to solve heretofore-unsolvable problems. Scientists at NCSA (located at the University of Illinois) with partners at the University of Iowa and Argonne National Laboratory demonstrated the power of combining computers at multiple locations. By linking over 1000 computers from around the world, computers from five different vendors operating together as a single parallel computer, they solved a quadratic assignment that has been unsolved for over thirty years. The specific problem in quadratic assignment is to find a lowest cost assignment of facilities to locations that will minimize the cost of moving material flows among locations. Quadratic assignment problems arise in such varied applications as locating factories, hospital layouts and designing computer chips.

Michael Klein and March Saitta of the University of Pennsylvania with colleagues at DuPont Research have used PACI facilities to model the behavior of knotted and unknotted polyethylene strands under strain. Their results have shown a fundamental difference in breaking behavior - knotted strands break at the entrance to the knot while unknotted strands break at the ends where the strain is exerted. Their work illustrates the power of high-performance computing to deepen our understanding of molecular scale phenomena.