January 28, 2010

Thoughts, Observations, Beliefs & Opinions About the NSF Supercomputer Centers

by Sidney Karin

There is no such thing as an NSF (Supercomputer) Center and there never has been. There should be. What there are, in the words of Ed Hayes, then comptroller of NSF, are “NSF ASSISTED Supercomputer Centers.”

This is a double edged sword. The directors of the NSF centers have historically had considerably more latitude and agility in their decision making and in the operations of their organizations than the directors of their peer organizations, sponsored by other federal agencies have had. This has led to much success in the past; in the pursuit of new avenues of research, development of innovative technologies, creation of research partnerships, fostering of relationships with both vendor and user industrial organizations and the raising of funds from outside sources.

The other side of the coin is that NSF has neither provided sufficient funding nor has it provided any other kind of support when centers found themselves in one sort of difficulty or another. In my direct experience, and to my direct knowledge of activities at other centers, NSF funding has been inadequate to provide the direct support of what used to be called the base program. Each center has raised funds from industry partners, state governments, local universities, and foundations.

These funds have been necessary to the successful operation of the base program and essential to the added value that the centers have created. This again is in contrast to the process at peer centers funded by other federal agencies. In my opinion this has been a worthwhile tradeoff for the so called NSF Supercomputer Centers. I would not have traded places with any of my contemporaries in other organizations. Nevertheless, it is possible to preserve the majority of the benefits while eliminating much of the negatives.

It comes as no surprise that the benefits of a research endeavor often go beyond the planned benefits of the research. Indeed, they frequently arise instead of the planned benefits. The so called NSF Supercomputer Centers have consistently provided modern state of the art computational infrastructure to the academic research community. Indeed, they have gone beyond that in the provision of early instances of leading edge computational and peripheral systems and the introduction of alternative approaches to computational science and engineering. More to the point, the centers have produced other results of enormous impact to the larger national and international community. MOSAIC is at the top of this list, but there have been many other successes. Note also the fundamental role of these centers in the establishment of the NSFnet and the transition to the commercial internet as we now know it.

These ancillary benefits have arisen precisely because of the flexibility and discretion afforded the centers. Indeed, in 1976 when DOE allowed such flexibility to its centers, LANL forged a deal with Seymour Cray that led to the establishment of the modern supercomputer industry. Later, when DOE would no longer allow such flexibility to its centers, the NSF centers were by default given the opportunity to work directly with vendors in the development and deployment of first of a kind computational systems. Some of these first of a kind systems became one of a kind systems while others flourished as is the nature of a research enterprise. The nation has benefited greatly from this prototype and test bed process at the NSF centers. In addition, the available flexibility and discretion afforded the centers was evident in numerous examples of the emergence of new research emphasis, new research directions, innovative software and technology development and deployment, and dozens, if not hundreds, of spin off commercial enterprises.

In recent years I have been saddened to observe (from a distance) the substantial reduction in this centrally important aspect of the program. Flexibility and agility are greatly reduced. Large system procurements seem far more appropriate to acquisition of business data processing systems for applications such as payroll and accounts receivable than for the advancement of science. The greatest accomplishments of NSF supercomputer centers program would not, and could not, have taken place under current procedures.

It should be obvious that I am calling for a return to the original successful model that was put in place when the centers were first established. But all was not perfect in that model either. In particular, despite the obligations of the Cooperative Agreements, NSF often acted capriciously and undependably in the actual provision of funds. This was a severe destabilizing influence.

Finally, the capriciousness of the NSF funding support was not limited to failure to live up to signed cooperative agreements. It extended to in effect compelling each center to recompete for its very existence on an annual basis. This has had a debilitating impact on center staff at all levels and upon the level of success of the centers. What is needed is some form of institutionalization that would remove the fear of termination and the attendant enormous efforts put forth to prevent termination, all at the expense of productive efforts in furtherance of the center’s missions and the academic research enterprise overall.


Reprinted with permission of Sidney Karin, Professor of Computer Science and Engineering, University of California, San Diego. The original article was published in December 2009 by the National Institute for Computational Sciences (NICS).