October 30, 1998
Mr. Steven Howell
Committee on Science
United States House of Representatives
B-374 Rayburn House Office Building
Washington, DC 20515
Dear Mr. Howell:
I write in response to the questions forwarded by Representative Pickering concerning my testimony at the October 6, 1998 hearing on High-Performance Computing.
In addition to chairing the Department of Computer Science & Engineering at the University of Washington, I chair the Computing Research Association, whose members include the nearly 200 academic graduate departments and industrial research laboratories in the fields of computer science and computer engineering. It is on behalf of this broad constituency that I speak.
Representative Pickering forwarded nine questions. I will respond to each of them. I would first, however, like to ask your indulgence while I fabricate and respond to what I feel is the truly fundamental question.
0.How should the Committee view the role of the discipline of "computer and information science and engineering" (CISE)?
The Committee should view CISE as a fundamental discipline that, through its progress, brings about revolutionary and transformational changes in every aspect of our lives, and contributes in essential ways to America's world leadership and to the well-being of her citizens.
Because the Committee is concerned with "science," it is natural for the Committee to look at CISE as an enabler of other fields of science and engineering, and to focus mainly on those aspects of CISE that are of obvious relevance to these other fields.
But this is much too narrow a view of CISE. The fact that CISE is a fundamental enabler (and a revolutionary transformer) of other fields of science and engineering is important, but it's not the only reason that CISE is important.
When the Committee thinks about physics, or chemistry, or astronomy, or biology, or mathematics, the Committee doesn't focus mainly on what they can do to enable other fields of science and engineering. Rather, the Committee considers their broad direct and indirect impact on the nation and its citizens. Taking this broad view is even more important in the case of CISE:
- If you tie your support of CISE to its applications in other fields of science and engineering, you'll under-value (and under-support) many aspects of CISE that bring about revolutionary and transformational changes in other fields, such as commerce, education, employment, health care, manufacturing, government, national security, communications, and entertainment.
- More surprising, though, is that if you tie your support of CISE even to its much broader range of applications, you still run the very significant risk of stifling the true potential for revolution and transformation. The reason is that tomorrow's "killer applications" — the ones that really matter — will not be things that we can even envision today. They will be totally new applications made possible by extraordinary advances at the core of CISE. If your support of CISE is tied to extrapolations of today's applications — to the needs that we can envision — then work won't get done in areas of CISE that truly have the potential for revolutionary and transformational change, as opposed to evolutionary and incremental change. This is no different than other fields of science and engineering, but for some reason — perhaps, perversely, just because it is so darned useful — CISE seems to have a harder time achieving this recognition.
I want to emphasize that I am not "arguing against applications." Quite the contrary! All of us who work in CISE do so because we love to see our work applied — that's why we chose this field! All I am saying is that, in order to achieve the truly revolutionary and transformational potential of CISE, we must have a balanced investment strategy — one that supports the needs of today's applications and their extrapolations, and also supports the core of the field. Today's investment strategy is not balanced: it under-supports the core of the field.
This is precisely the point made by the Interim Report of the President's Information Technology Advisory Committee. PITAC argues for balanced support of CISE, including high-end computing, software, scalable infrastructure, and workforce and social issues. And within the high-end computing piece, PITAC argues for focused work on software and architecture, as well as for system acquisitions. PITAC gets it just right.
To summarize this very long-winded response to a question that you didn't ask:
- CISE is a fundamental discipline that, through its progress, brings about revolutionary and transformational changes in every aspect of our lives, and contributes in essential ways to America's world leadership and to the well-being of her citizens.
- Tomorrow's "killer applications" of computing will be things that we cannot even envision today. They will be totally new applications made possible by extraordinary advances at the core of CISE. This, a wise investment strategy will balance support for the needs of today's applications and their extrapolations, with support for the core of CISE. That's the only way that the revolutionary and transformational potential of CISE will be realized. For the health of the nation, support for research at the core of CISE must be dramatically increased.
- Even if you focus just on the ways in which CISE enables other fields of science and engineering — a very narrow view — it's important to understand that this involves much more than high-end computing! Datamining, the web, algorithms for computational astrophysics, and the deep intellectual partnerships that characterize the confluence of biology and computation ("the genome is just bits"), are examples. More than by supercomputing, science and engineering are being transformed by the use of "ordinary" computers in extraordinary ways for information and communication.
- Even if you focus just on high-end computing — an even narrower view — it's important to understand that this involves much more than hardware: "It's the software, stupid" as we discussed at the hearing.
1. The PITAC Interim Report recommends that NSF be given the leading role in coordination of the proposed $1 billion research initiative.
1.1. Is NSF the right agency to be given this responsibility? Are there reservations about NSF's ability to perform this role?
I am presently part of a PITAC working group seeking to refine the notion of "lead agency." (Remember, it was an interim report!) We have spoken with the heads of a wide range of Federal agencies: NSF, DARPA, NIST, Energy, and others. What I am comfortable saying, and what I believe many others would support, is the following:
- First and foremost, PITAC's recommendation that half of the recommended increment should go to NSF is appropriate. I support this recommendation strongly. As Ray Kammer from NIST said to us last Thursday, "NSF invests the money in university research programs, and this is where the greatest under-investment is today." Another important attribute of NSF is that its mission is broader than that of other Federal agencies, and is well in line with today's national priorities.
- Computer and information science and engineering has benefited enormously over the years from the involvement of multiple Federal agencies with multiple missions and multiple "styles." There is strength in diversity! This strength — this diversity — must be preserved. But coordination, oversight, and representation at the highest levels of government are essential. In my view, the right approach would be the appointment of an Associate Director for Information Technology in the White House Office of Science and Technology Policy (OSTP). Information technology has accounted for fully 1/3 of the total growth in U.S. production since 1992, and continued rapid progress in information technology is essential for achieving America's 21st century aspirations. Information technology deserves, and requires, this level of coordination, oversight, and representation. (Among other duties, the OSTP AD for IT would "adopt" the current National Coordination Office.)
- I envision multiple IT research "thrusts," with collections of Federal agencies participating in each one. I envision NSF having special responsibility for the "core" of the field — serving as the lead agency for a dramatically expanded program of fundamental research directed at the "core" of the field.
NSF is absolutely capable of performing this role. In my view, though, NSF has tended to view CISE too much as an enabler of other fields of science and engineering, as described in my response to Question 0. (NSF can be viewed as a "mission agency" — its mission is all of science and engineering. So it is natural for NSF to focus on those aspects of CISE that support its broad mission — "What can CISE do for chemistry?") Not enough support has been focused on the core of the field. It is essential that the funding of NSF's CISE Directorate be dramatically increased, and that the lion's share of this increase be directed towards programs that support computing research, as opposed to programs that provide infrastructure for all of science and engineering. NSF must dramatically increase the Foundation-wide proportional investment in the core of CISE.
1.2. Does NSF have the breadth of expertise and sufficient staff to carry out this function effectively?
The NSF CISE Directorate is badly under-staffed, particularly considering that CISE is both a discipline unto itself, and the core of a wide variety of multi-disciplinary initiatives. Staff quality is fine, and increased funding will have a highly positive impact — individuals will want to join the CISE Directorate because they will be able to shape the nation's future. I, and the community, have enormous confidence in the incoming Assistant Director for CISE, Dr. Ruzena Bajcsy, and in the new Director of the Foundation, Dr. Rita Colwell.
2. The PITAC Interim Report recommended that steps be taken to ensure that academic researchers be given access to leading edge scientific computers.
2.1. Is it generally agreed that the computers available at NSF supercomputer centers are falling seriously behind the state of the art for high performance computers?
Let me begin by reminding the Committee that PITAC recommended an increased focus on four areas. High-end computing was one of those four areas. And the provision of improved access to today's state-of-the-art scientific computers was one part of that one area. Because the Committee is responsible for the health of many areas of science and engineering that require immediate access to the latest high-end computers, it is natural for the Committee to have a special interest in this issue. But it is critical to keep it in context. The PITAC Interim Report has the right balance.
Let me also point out that access to high-end computing does not principally serve researchers in CISE — it principally serves those in other fields of science and engineering.
With that said, the answer to your question is an emphatic "yes": the capabilities of the NSF PACI centers are behind the state-of-the-art.
2.2. What should be done to carry out the recommendation?
Any response must be measured in the context of the overall PITAC recommendations. Improved access to today's state-of-the-art scientific computers is one part of one of the four PITAC recommendations. To disproportionately fund this element would be a disservice to the nation.
The NSF PACI centers are important for at least three reasons. First, they provide scientific computing to scientists and engineers in a broad range of disciplines. Second, they have established outstanding partnerships spanning universities and industry. Third, they have established outstanding partnerships that bring together computer scientists and computational scientists — those inventing the future of computing, and those applying today's computing to move their own disciplines ahead.
I don't believe that a mission-focused center will be able to combine these three crucial elements. It is conceivable that a mission agency could create a new center with these elements as its explicit goal. It certainly seems, though, that the most effective, and most cost-effective, approach would be to increase the capabilities at the two NSF PACI centers.
It is not clear that the PACI centers must be as close to the absolute bleeding edge as the Department of Energy ASCI machines. While recognizing the importance of access to state-of-the-art systems by the broad range of scientists and engineers, one must calibrate for "ASCI envy."
3. Is the academic community getting sufficient access to high-end computers? Will the recommendation in the PITAC Interim Report lead to improved access?
No, access is insufficient. But (to repeat my refrain), so is support for research at the core of computer and information science and engineering — in fact, much more so. And remember that access to high-end computers supports all of science and engineering, not just (not even mainly, or much) researchers in CISE). Yes, the recommendation in the PITAC Interim Report will lead to improved access.
4. Has the balance between basic vs. applied research shifted too far towards the applied side, as the Interim Report suggests? If so, how can the proper balance be restored?
The PITAC Interim Report makes a distinction between long-term and short-term IT research, and does not frame the question as one of basic versus applied research, which is less useful.
In CISE, both long-term and short-term research are crucial. Short-term CISE research receives "natural attention" for the reasons outlined in the PITAC Interim Report: the pressing needs of the mission agencies for ever more advanced applications of computing, and the general tendency to view CISE mostly as an enabling technology, and to support mostly the CISE research which is tied to applications. The problem is that support for long-term CISE research has not kept pace, threatening the foundation for tomorrow's revolutionary and transforming uses of computing — the kind that lead to order-of-magnitude productivity boosts, entirely new industries, and dramatic improvements in quality of life. So yes, the balance has shifted, but it is through under-investment in long-term CISE research, rather than through over-investment in short-term CISE research.
We can redress the under-investment in long-term CISE research by following the prescription set forth in the PITAC Interim Report: dramatically increase funding for long-term research at the core of computer and information science and engineering, and pursue innovative research models such as projects of broader scope and duration, virtual centers for "Expeditions into the 21st Century," and "Enabling Technology Centers."
5. The Interim Report recommends an infusion of $1 billion over 5 years to enhance Federally sponsored, long-term research on information technologies.
5.1. Is there general agreement that the Federal R&D portfolio is under-investing in long-term research?
Yes, without question. The level of support for long-term CISE research is badly inadequate by any measure. It will not lead to the robust, reliable, scalable computing and communications systems that are needed to meet the complex needs of the 21st century. Without dramatically expanded investment in fundamental CISE research, the nation's ability to meet future IT-intensive goals — in both the public and private sectors — will be severely compromised.
It is important to understand the critical role in innovation that the Federal IT research portfolio has played in the past, and must play in the future. The National Research Council "Brooks/Sutherland Report" on the HPCC program, of which I was an author, makes this case compellingly. Research investments made 10, 15, and 20 years ago are bearing fruit today. It is essential that we sow the seeds of fresh ideas so that we will be able to harvest them 10, 15, and 20 years hence.
5.2. Is the existing research community adequate to carry out this proposed new research initiative?
The research community is adequate to carry out a program of research that is significantly expanded in scope from the one currently being funded. This is evident from the large number of outstanding proposals going unfunded, and from the fact that the low level of funding is one contributor to the current short research horizon.
More importantly, nearly 1,000 new Ph.D.'s are granted in computer science and computer engineering every year. If research funding increases, more of these graduates will choose to embark on academic careers, or to remain in such careers. (I have lost 4 faculty members in the past 3 years to industrial positions. I have lost 0 to other academic positions.) This in turn will increase our capacity to educate coming generations of researchers.
The PITAC Interim Report suggests a 5-year ramp-up in support, consistent with the need and the ability to grow the field. The nation must make this investment. There is crucial work to be done if we are to achieve America's aspirations for the 21st century.
6. PITAC recommends that NSF be given the leading role in coordination of the proposed $1 billion research initiative. Didn't the NSF CISE Advisory Committee recently point out that the Directorate needed to increase staffing levels just to carry out its current program?
Yes. In my opinion, NSF has under-invested not only in CISE research, but also in CISE Directorate staffing. But this is an easily curable problem! Also, as I noted above, increased research funding will have a highly positive impact on both the quantity and the quality of NSF CISE Directorate staff — individuals will want to join the CISE Directorate because they will be able to shape the nation's future.
7. Some of today's testimony suggests that our focus on massively parallel machines has drawbacks as well as advantages. Does this focus need to be changed?
I feel that the focus already has changed, in precisely the appropriate way. No action is necessary.
Yesterday, all high-end computing was done on vector machines. Vector machines can't cut it for today's high-end applications. Today, all high-end computing is done on parallel machines. The Federal research program in very-large-scale integrated circuits, computer architecture, software, and parallel systems has made this possible. Stay the course. As the PITAC Interim Report says, we must invest in the applications and tools to make these new machines usable.
8. Concerning the information infrastructure, the Interim Report contents that we cannot extend safely what we currently know to more complex systems.
8.1. Will the current information infrastructure — particularly the Internet — be able to handle future traffic? No. Both PITAC and the National Academy have expressed grave concerns on this point. The Internet is in no way a true "information utility." New discoveries are needed to make it scalable and robust. An aggressive program of research is needed. This is what the PITAC Interim Report recommends — that's part of what "Scalable Infrastructure" is about. I applaud PITAC for the wisdom of this recommendation.
8.2. What can be done to improve its capacity and dependability?
Invest in research. Federal investments in research created ARPANET. Federal investments in research and research infrastructure transitioned ARPANET to NSFNET and thence into the Internet (and its host of multi-billion-dollar businesses) as we know it today. Federal investments in research are essential to evolve the scalable, robust information infrastructure that the nation requires. We need to get cranking on this.
8.3. Where does the Next Generation Internet program fit into the Interim Report's proposals?
The NGI program involves a range of activities. This far, the level of support for fundamental research into evolving the nation's information infrastructure has been fairly modest. In making its overall funding recommendations, PITAC considered the scope and nature of existing programs such as NGI, the DoE ASCI program, and others.
9. What is your view on Federal restrictions on the import of high-performance computers? What is the effect on computational science and engineering?
There are some old programs that work best on vector machines. The Japanese make the fastest vector machines. These vector machines simply aren't scalable, so ultimately the old programs are going to have to be re-designed to run on scalable parallel machines in order to keep pace — re-designed with the help of some of the research that the PITAC Interim Report advocates. In the mean time, let the Japanese machines in — you surely don't want to encourage (or subsidize) American companies to complete in a dying niche market!
The very fastest machines — the machines at the bleeding edge, such as the Department of Energy ASCI systems — are scalable parallel machines, in which America holds a commanding lead thanks to past Federal research investments. And all markets except for the bleeding edge are economically sustainable. Modern systems short of the bleeding edge meet the vast majority of needs of computational science and engineering. I thus do not believe that the Federal restrictions have a significant impact on the general conduct of computational science and engineering.
I would like to thank the Committee for the opportunity to respond to these additional questions. I look forward to working with you in the future.
Sincerely,
Edward D. Lazowska
Chair, Board of Directors,
Computing Research Association
Document last modified on .