Partnerships in Advanced Computational Infrastructure

Ken Kennedy, Director, CRPC


The high-performance computing community is being invigorated by the excitement surrounding the new NSF program called "Partnerships in Advanced Computational Infrastructure" (PACI). Designed as a successor to the NSF Supercomputer Centers, the PACI program calls for proposals from consortia built around "leading-edge sites" offering hardware systems that are orders of magnitude more powerful than systems typically found at major research universities. The other consortium institutions, or "partners," would be expected to "facilitate research and experimentation with new hardware and software" and to "promote education and training." The partner sites would typically have smaller research configurations of the machines at the leading-edge sites or experimental systems of a completely different character. Resources to support these configurations would come from the base NSF funding of the partnership.

The strength of this program is that it requires the bidders, many of whom will be the existing NSF centers, to involve the computational science research community from the outset. This involvement is in many ways patterned after the successful organization used by the CRPC, with consortium members playing a significant role in establishing the vision and direction for the partnership. Since the budget is expected to be about the same as the current supercomputer center budget--approximately $65 million per year--and the cost of each partnership should be higher than a stand-alone center, one or more of the existing centers will be unsuccessful. This means that the competition should be fierce, as existing centers are struggling to maintain their positions and new centers seek to replace them.

A critical component in the evaluation will be the perceived quality of the partner institutions and the extent to which they fit into a coherent vision for the future. This has led to a flurry of activity over the past two months, as centers reach out to potential partners and vice versa. I expect this activity to increase as the April 15 deadline for preproposals approaches and each bidder seeks to put together the strongest possible team.

I am of two minds about all of this. First, the existing NSF centers have, for the most part, been quite effective in providing access to high-end supercomputing to the NSF research community. It is unfortunate that there are not enough resources to continue to expand all of them. It is also unfortunate that the vision of all the supercomputers knitted together into a single "metacenter" is dead, at least for the duration of the competition.

On the other hand, the intellectual ferment that this competition will generate over the next few months will bring new life to the HPCC program. Talented computational scientists all over the nation will be engaged in what might be one of the broadest planning efforts ever undertaken by our community, and the winning proposals will help define the character and direction of high-performance computing for the next decade. In the end, this has to be good for computational science and for the nation.


Table of Contents