The Federal Role As Early Customer

David B. Nelson

David B. Nelson is Executive Director in the Office of Energy Research, U.S. Department of Energy. For more complete biographical information, please see his presentation earlier in this session.

Federal agencies—and I'll use Department of Energy (DOE) as an example because I'm most familiar with it—have for many years played an important role as an early customer and often as a first customer for emerging high-performance computing systems. This is true not only in federal laboratories but also in universities with federal funding support, so the federal role is diffused fairly widely.

Let's talk about some of the characteristics of the good early customer.

First, the customer must be sophisticated enough to understand his needs and to communicate them to vendors, often at an early stage when products are just being formulated.

Second, the customer must work closely with the vendors, often with cooperative agreements, in order to incorporate the needs of the customers into the design phase of a new product.

Third, the customer is almost always a very early buyer of prototypes and sometimes even makes early performance payments before prototypes are built to provide capital and feedback to the vendor.

Fourth, the early buyer must be willing to accept the problems of an immature product and must invest some of his effort to work with the vendor in order to feed back those problems and to correct them. That has historically been a very important role for the early good buyer.

Fifth, the early buyer has to be willing to add value to the product for his own use, especially by developing software. Hopefully, that added value can be incorporated into the vendor's product line, whether it is software or hardware.

And sixth, the early buyer should be able to offer a predictable market for follow-on sales. If that predictable market is zero, the vendor needs to know so he doesn't tailor the product to the early buyer. If the early buyer has a predictable follow-on market that is not zero, the vendor needs to be able to figure that into his plans. For small companies this can be extremely important for attracting capital and maintaining the ability to continue the product development.

The High Performance Computing Initiative authorizes an early-customer role for the federal government. The six characteristics of a good customer can be translated into five requirements for success of the High Performance Computing Initiative.

First, there must be a perceived fair process that spreads early customer business among qualified vendors. As we learn from Larry Tarbell's description of Project THOTH (Session 12) at the National Security Agency, competition may or may not be helpful. In fact, it has been our experience in DOE that when we are in a very early-buy situation, there is usually no opportunity for competition. But that doesn't mean that fairness goes out the window, because various vendors will be vying for this early buyer attention. They ought to. And by some mechanism, the spreading of the business among the agencies and among the user organizations needs to take place so that the vendors can perceive that it's a fair process.

Second, the early buyer agency or using organization must provide a patient and tolerant environment, and the agency sponsors need to understand this. Furthermore, the users of the product in the early-buy environment need to exercise restraint in either bad-mouthing an immature product or, equally bad, in making demands for early production use of this immature product. We have seen instances in agencies in the past where this aspect was not well understood, and mixed expectations were created as to what this early buy was supposed to accomplish.

Third, these early buys must be placed into sophisticated environments. Those agencies and those organizations that are going to participate in this aspect of the initiative need to be able to provide sophisticated environments. The last thing that a vendor needs is naive users. He accommodates those later on when his product matures, but when he is first bringing his product to market, he needs the expertise of the experienced and qualified user.

Fourth, early buys should be placed in an open environment, if possible. The information as to how the computer is performing should be fed back at the appropriate time, not only to the vendor but also to the rest of the user community. And the environment should involve outside users so as to spread the message. The organization that is participating in the early buy or the prototype evaluation needs to have both a willingness and an ability to share its experience with other organizations, with other agencies, and with the broader marketplace. Clearly, there's a tension between the avoidance of bad-mouthing and the sharing of the news when it's appropriate.

Finally, we need to have patient capital—the federal agencies must be patient investors of advanced technology and human capital. They must be willing to invest today with an understanding that this product may not return useful work for several years. They must be willing to find users to try out the new computer and feed back information to the vendor without being held to long-term programmatic accomplishments. In many cases, these sophisticated institutions are leading-edge scientific environments, and scientists must make tradeoffs. Is a researcher going to (1) publish a paper using today's technology or (2) risk one, two, or three years of hard work on a promising but immature technology?

As we put the High Performance Computing Initiative together, we kept in mind all of these aspects, and we hope we have identified ways to achieve the goals that I laid out. However, I don't want to leave you with the idea that this is easy. We, all of us, have many masters. And many of these masters do not understand some of the things that I've said. Education here will help, as in many other places.