Features:

INTERVIEW WITH JAMES R. MCGRAW, DEPUTY DIRECTOR, ISCR, LLNL
By Tim Curns, Assistant Editor, HPCwire

HPC: SC2003 is supposed to have the largest number of exhibitors to date, with more than 200 booths. In addition, attendance is on track to match last year's Baltimore record-breaking number. This all comes in the midst of a slow or dragging economy. Why do you think this is? Why do you think attendance is as good or better than in Baltimore, where there was a surge in attendance? What's happening to conference momentum?

JAMES R. MCGRAW: Our conference history shows that we have a dedicated, committed, strong core of attendees. This group's presence makes the SC conference the one place that professionals in supercomputing and high-speed networking need to be. Each year, we work extremely hard to create a high quality technical program that will keep these folks coming back. I think the current trends show we're keeping up with our attendees' needs and interests.

On the exhibit floor, SC has redefined the scope of what we need to include to be of most benefit to attendees. Industry exhibitors cover the full range of hardware and software products to enable an Integrated Computing Environment -- desktop to teraflops. This is an unparalleled opportunity to see all of the vendors and compare products, including networking, mass storage, computer security, software and, of course, supercomputers. Not only can you view the products, but you can talk to the people who develop and build them. Research exhibitors provide a unique balance in that they are using much of these capabilities in bleeding-edge applications. Attendees can find out what they need to do to effectively use these systems, by leveraging the experiences of the research exhibitors.

Last year, we expected a surge in attendance, due to the proximity to the Washington, D.C., area. Historically, that has been our greatest draw. This year, Phoenix is clearly a drawing-card due to its great climate, easy access for travelers and also because Arizona has a growing interest in technology as a key industry for its future.

However, I think that all of the recent headlines and interest in supercomputing in the last 18 months is probably the most important factor. Everyone wants to learn more about Japan's Earth Simulator system, DOE's ASCI systems, Computational Grids and most recently, the announcements about a supercomputer made out of Mac's. This conference is a place you can go to learn more about these systems and get to a better understanding of what they can really do.

HPC: Five or six years ago, supercomputing seemed to be dragging a bit, and there was even talk about whether the conference was still relevant. Why do you think that this year's conference is on track to be the most successful ever? What's happening in the supercomputing industry to account for such an explosion of interest?

JM: Supercomputing and high-speed networking are enduring needs; we have always been able to identify critical problems that cannot be solved with the currently available technology. However, at various times, other issues and other technologies can capture the imagination and attention of the community. In the last six to eight years, the "dot-com" industry was booming and demanded the attention of hardware and software experts, because of the sheer size and potential impact on the economy. Clearly, that situation has now changed, but who knows for how long?

Our thinking about supercomputing has also evolved over that same time period. We went through a period where many companies were proposing radically different approaches to supercomputing (Cray to Thinking Machines). About six to eight years ago, that started to settle out into a very few winners, which may have LED to a little complacency and less interest and attention because there seemed to be a clear path. With the arrival of the Earth Simulator, Linux Clusters and the Mac Supercomputer, we are seeing more options for high performance with very different cost profiles. So, the future is more interesting and more challenging, trying to determine which types of systems are best for the new applications challenges.

Our approach to the exhibit floor is also attracting greater participation from private industry. Supercomputer users recognize the need for Integrated Computing Environments and more vendors are seeing the value and role they can play in this market segment.

And, finally, I think we are seeing renewed interest in supercomputing on the political front. The National Academy of Science is undertaking a study of "The Future of Supercomputing." The National Coordination Office's High-End Computing Revitalization Task Force (HECRTF) is "developing an interagency R&D roadmap for high-end computing core technologies, a federal high-end computing capacity and accessibility improvement plan, and discussing issues (along with recommendations where applicable) relating to federal procurement of high-end computing systems." The National Science Foundation is taking steps toward developing an enhanced cyber-infrastructure and developing an Extensible Terascale Facility to demonstrate the potential of Grid computing. These examples of activity indicate renewed interest and increased funding for supercomputing and high-speed networking, which naturally attracts more interest and more attention.

HPC: Much of the preparation for SC2003 is done by volunteers such as yourself and other members of the SC2003 Executive Committee. What is it like preparing for such a conference? How do you organize people from all over the country, and how long before the conference do you start planning?

JM: This conference would not be possible without the dedicated support of more than 200 volunteers. Over the years, we have developed an effective "training" scheme, in that most key volunteers are selected over two years prior to their conference dates, so they can serve as deputies for one year and observe the full process in action, before they have to take charge. Volunteers also help out tremendously by coming back often and taking on different roles in various years. Right now, we have a truly amazing set of volunteers who are cross-trained in a number of areas. Almost all of our committees have an excellent mixture of experience and new members. My preparation began three and a half years ago when I was selected as General Chair for SC 2003. As one past chair described it, my primary job is to pick good people to be in charge of all of the various areas of the conference and then get out of their way and let them do their jobs. It has been a privilege working with all of these folks. They are terrific.

HPC: Please comment on the decision to select Donna Cox as keynote speaker. How did you finally decide on selecting Ms. Cox?

JM: I chose Donna Cox as the keynote because her research and her message are well-aligned with this year's conference theme of "Igniting Innovation." We want this conference to be known as a place where attendees come to stimulate their thinking and create new innovations, whether those innovations are in the technologies that underlie this field or in innovative uses of the technology to solve important problems.

Donna Cox is both an artist and a research scientist at the National Center for Supercomputing Applications at the University of Illinois- Urbana-Champaign. Her talk, entitled "Beyond Computing: The Search for Creativity," examines how our human abilities of innovation and creativity interact and react to new capabilities provided through computing, visualization and data mining. I am thrilled that Donna accepted our invitation to give our keynote speech.

HPC: Any other examples of how you see SC2003 igniting innovation when we meet in Phoenix?

JM: This year, we are also including SC Global, an activity first launched at SC 2001. The SC Global 2003 program has two major thrusts: the content of the presentations and the technology required to produce the geographically distributed program. SC Global 2003 will feature presentations by speakers from 20 remote locations, seven countries, and five sovereign tribal nations, distributed across four continents. We will link the Phoenix Civic Plaza Convention Center with Access Grid nodes worldwide. SC Global 2003 plans to present a very dynamic technical program through a combination of panel discussions, "Birds of a Feather" meetings and showcases events. One goal of this portion of the technical program is to demonstrate how the Access Grid brings together the right people and the right data at the right time in order to perform a complex task, solve a difficult problem or simply discuss issues that are pressing at that moment in time.

The SC conference series has a proud tradition of reaching out to faculty and students, working in partnership with the Shodor Educational Foundation, NSF, ACM's SIGArch and IEEE Computer Society. The SC2003 Education Program engages undergraduate faculty and K-12 teachers in four days of hands-on workshops to learn about high performance computing and communications tools and resources appropriate for their classroom. The participants selected to participate in this program will learn about approaches to using modeling and visualization in the classroom, including numerical modeling, algebraic modeling, dynamic modeling systems and agent modeling. The learning goes beyond just the conference week; participants will meet again for one week next summer. We expect to have at least 15 regional sites from which to choose to receive further training. The goal is to have the educators develop curricula that will expose their students to the supercomputing and high-speed networking.

SCinet, the state-of-art, on-site network designed and built especially for the annual SC conference, enables a rich environment for real-time demonstrations, communications and collaboration. SCinet works with applications developers who attempt demonstrations using most or all of the provided bandwidth. A high-speed network testbed provides access to major national networks and testbeds, and a virtual conference capability with international participants. As in prior years, an elite team of researchers, exhibitors, communications carriers and networking equipment suppliers will work with talented volunteers from universities, government and industry to assemble and operate SCinet, making the SC2003 conference one of the best-connected sites on the planet.

HPC: While SC2003 may have seemed like a full-time job, you also have your responsiblities at Lawrence Livermore National Lab. You are perhaps best-known for your work in SISAL, an innovative approach to parallel computing. Would you care to reflect on that?

JM: SISAL is a research language developed to explore the question of how well a functional language might serve as a vehicle for parallel computing. Work began in the mid-80s and continued for about 10 years. At that time (and still today) most parallel programming languages employ an underlying sequential execution model (execute statements one at a time in order) with extensions to permit different sections of code to execute in parallel.

A new type of computing paradigm was emerging in "data flow" machines, which turned everything upside down because individual operations executed as soon as all of their inputs were available. To deal with the chaos, this type of machine, functional languages like VAL and ID came into being. The original SISAL developers (of which I am one) advanced the theory that these highly parallel, functional languages may also be an excellent programming paradigm for the more conventional parallel machines, particularly when it looked like future platforms could go from four and 16 parallel processors up to large numbers (like 512 and 1,000).

Research in SISAL did produce some important results, including putting to rest the worry that functional languages would have to waste far too many resources copying data to ever produce efficient parallel programs. We were able to write and execute some very fast parallel codes. In the end, lack of funding, technical concerns over handling I/O, and the worry that programmers would not be willing to adopt the radical style of functional programming led to the decline of the research.

However, I must say I was gratified to hear Burton Smith of Cray Research give a talk on this subject not long ago. He suggested that if we really want to explore new parallel programming paradigms, we should look closely at the results produced by functional languages like SISAL. He believes that we need to consider pursuing some approaches that mix functional languages with concepts from relational programming. In my opinion, that could provide us with some excellent capabilities for efficient and portable parallel codes.

Burton's suggestion characterizes the high performance computing and networking community. We build on past experience to invent the future, whether it's the re-emergence of vector computing in the world's fastest supercomputer, the resurgent idea of building systems specially designed for scientific computing or giving a new generation of computing experts an opportunity to meet and talk with some of the leading lights in the field.