CONVEX COMPUTER CORPORATION WRITTEN STATEMENT Presented to U.S. Senate Commerce, Science and Transportation Subcommittee on Science, Technology and Space CONVEX supports S. 272, the High-Performance Computing Act of 1991, as we believe it will assist U.S. industry in maintaining leadership in computing technology. We strongly believe this legislation can positively contribute to one of the biggest threats facing the United States today: the loss of our international competitiveness in all technology related businesses. In addition, it will directly stimulate the supercomputing industry. Europe and Japan have targeted information technologies for particular attention, and unless decisive steps are taken to ensure our continued leadership, the U.S. could be surpassed in a technology field that we largely pioneered and which is vital to our economic future. The real American competitiveness question involves making our nation's industries competitive. The use of supercomputers is mandatory to maintaining America's competitive edge in all of our key industries, such as aerospace, automotive, electronics, pharmaceuticals, petroleum, etc. -- not just in supercomputing manufacturing. We believe the actions called for in S. 272 -- particularly the acceleration of the development of computer systems and subsystems, the stimulation of research on software technology, and the application of high-performance computing to "Grand Challenges" - - are not only appropriate goals, but vital to maintaining the U.S. lead in supercomputers and utilizing supercomputer technology in our high-tech industries and research. Supercomputers are the fundamental building blocks that contribute to almost all disciplines across the broadest spectrum of science and technology. In the 1990's, the way America can stay competitive is literally to put supercomputing in the hands of the "masses." Supercomputers are to the modern technologist what the invention of the microscope was to biologists and the telescope was to astronomers. In fact, supercomputers enable scientists and engineers to solve problems for things that are too small, too large, too quick, too slow, or too dangerous to observe directly. This use in industry results in new products that are more innovative, safer, and get to market more quickly. Their use in research results in fundamental breakthroughs in science that change how we see the world. The supercomputer is the one common tool across all U.S. scientific and technological activities that, if put in the hands of engineers and scientists throughout the United States, can dramatically sharpen the competitive output of the United States. Of course, Japanese industry and research institutions totally understand and believe these concepts. From our perspective, they have been the fastest nation to purchase CONVEX's latest technology. Until just recently, there were more of CONVEX's top- of-the-line supercomputers in Japan than in the United States. American researchers and engineers believe these concepts also, but access to supercomputer tools has been limited. S. 272 can be the catalyst to change this trend. CONVEX's assessment of the competitive position of the high- performance computer industry in the U.S. relative to that of Japan is as follows: The high-performance computer market is an international market in which Cray dominates the high-end of the market, and CONVEX dominates the mid-range market. The Japanese computer manufacturers, NEC, Fujitsu, and Hitachi, have high performance, fast hardware products. But while this is the case, U.S. high performance computer companies currently maintain the lead in supercomputing for the following reason: supercomputing is not about hardware, it's about solving complex problems. The U.S. supercomputer companies are ahead of foreign competition because we understand there are aspects to supercomputing solutions: o Balanced, high-performance hardware: There is more to real performance than pure megaflops or gigaflops performance. Unfortunately, that' s how performance is commonly measured but these definitions must be properly interpreted. There is much more to useful performance than peak speed, such as software performance, memory performance, and 1/0 performance. Users care only about the performance of their applications -- the problems they specifically solve with their machines -- and this type of performance is determined by dozens o attributes. In terms of speed, the Japanese have high peak performance, but that's only a part of the supercomputing solution. o Software technology -- Operating systems (UNIX) and compilers: Maintaining the lead requires being proficient at several software standards. Companies such as CONVEX and Cray recognized the emergence of the UNIX standard long ago and designed their machines for UNIX -- now considered a requirement in supercomputing. Japanese systems have historically been based on IBM standards and only now are attempting to migrate to UNIX. Also superior compiler technology is critical to computing performance and productivity. American companies and research institutions lead in this areas, as well. o Application specific software: Most of the supercomputers in use today, especially in industry, utilize third-party written software applications rather than custom-written software applications. The majority of that third-party software is developed by U.S. based organizations. CONVEX considers having both a broad array of application software available on its machines and having agreements/relationships with the software developers, as critical elements of its competitive strategy and success. American suppliers are leading in this crucial area. o Service and support -- taking care of the customer: This is a critical component in supercomputing solutions. American companies' reputations in the area of service and support are superior. American suppliers utilize direct sales and support organizations in all major markets and, as such, are closer to the customer. Outside of Japan, Japanese manufacturers typically use distributors or OEMs for sales and customer support. It would be naive to believe that U.S. companies will always be able to maintain the supercomputer lead for the reasons cited above without continual development and diligence in these areas. The Japanese can -- and will, in time -- develop these necessary strengths. Although CONVEX has been selling its supercomputers successfully to the Japanese for almost six years now, we also realize that when, or if, the Japanese companies decide that the price/performance market niche that CONVEX currently dominates is a viable and sizable market for Japan, the competitiveness threat posed by Japan can become very serious. The biggest threats posed by the Japanese to American supercomputer companies are: o The size of the big three Japanese companies is over $89 billion, which provides substantial financial staying power. This gives them the ability to mask the success or lack of success of their supercomputer products versus U.S. supercomputer companies, whose existence relies solely on the success of their supercomputers. o Furthermore, they can afford to not be profitable in the supercomputer market segment for a very long period of time and can buy market share by excessive and unreasonable discounting, while public U.S. companies are forced to live by quarter to quarter reporting, which represents the results of a single technology focus. o The big three Japanese computer companies also dominate the semiconductor industry, including advanced semiconductor research and development required to build supercomputers. o The cost of capital differs substantially for U.S. versus Japanese companies. In light of these factors, staying competitive in today's global supercomputer market will take a concerted effort by American companies, as well as cooperation and constructive stimulation by government. Certainly, the High-Performance Computing Act of 1991 will be a positive contribution in this direction. Comments on the bill. S. 272 General Comments CONVEX enthusiastically supports this legislation and commends it to you for your favorable consideration and swift passage in the House. We fully support the idea of a "National High-Performance Computing Program." There are several provisions of the bill on which I would like to comment and highlight. The High-Performance Computing Advisory Panel The federal government has played a prominent role in the American supercomputing success story and S. 272 again demonstrates this leadership. In several areas of the bill, cooperation between government and industry is called for to review progress made in implementing the plan and making necessary revisions. In particular, the bill calls for the establishment of a High- Performance Computing Advisory Panel consisting of representatives from industry and academia to assist with these tasks. I want to highlight this concept as being extremely important to achieving the objectives of the bill. The results of the expenditures for equipment and research called for by the bill must ultimately be the development of competitively superior commercial products. The strategic plan that is put into place by this bill should have this as a fundamental objective. Government is better qualified for some aspects of the task, and industry is better qualified for others. Partnership between the two will allow the plan to utilize the best capabilities of both. CONVEX has exposure to applications, research and product developments occurring all over the world, and in the broadest of scientific areas. We volunteer to help in whatever ways we can. The National Research and Education Network (NREN) CONVEX fully supports the bill's provision calling for the creation of a multi-gigabit-per-second National Research and Education Network (NREN). It is our perspective that in the past, too much emphasis was placed on providing limited access to too few centralized machines. Supercomputing must be made available to, and meet the needs of, a broad base of users through widely distributed supercomputer systems placed closer to the ultimate user. This would not supplant the centralized machines, but rather complement them. I suggest that in establishing NREN, it should not only be envisioned as a multi-gigabit per second backbone network, connecting only a small number of very high-speed, centralized computer systems. Let's think of it as a distributed network of computing and telecommunications services, serving the widest possible number of scientists and engineers from government. industry and academia. The National Science Foundation's national supercomputer centers represent a case in point. The program has been a success, but we can learn from what those users are additionally asking for: supercomputing close to the user. Let's supplement and complement the national supercomputer centers with affordable, open, accessible supercomputing facilities, available in departments and dedicated to products across the nation. Let's put a broad range of supercomputers, distributed data bases, and other research and production facilities, in the very laps of those who need them to help maintain and regain America's preeminence in many disciplines. Software In the last ten years, only about 300 high-end supercomputers have been sold by U.S. companies to industry and to research institutions. From CONVEX alone, over 600 high-performance computing systems have been shipped in only five years. American industry needs distributed, affordable supercomputing power to remain competitive. These companies, large and small, are voting with their checkbooks for this means of providing supercomputing. They are using supercomputing in production environments, not just in their research laboratories. They need supercomputers to bring new and improved products to market faster. Supercomputers are a valued competitive weapon for all of these companies. The full utility of supercomputers can only be reached through software. The sophisticated supercomputing user community desperately needs improved software development tools, computer- assisted software engineering (CASE) capabilities, and better algorithmic methods. With this improved state-of-the-art software, we can move forward with attacks on the Grand challenges enumerated in the bill. CONVEX wholeheartedly supports the software tasks and goals of the bill. Care should be taken to ensure that resources are not wasted by reinventing what may already exist in industry or somewhere in the world. ~t' s concentrate on improving software technology, but adhering to industry standards wherever possible, and avoiding proliferating proprietary solutions to software problems. Basic Research and Education CONVEX strongly supports the provisions of the bill in the areas of basic research and education. Only the largest and richest corporations can afford to have very much of their resources dedicated to basic research. Most of the industry, and I count CONVEX in this group, must use its limited research and development resources in the development and production of the next generations of our commercial products. So we need a fertile source of basic research if the supercomputer industry and the nation are to progress. Again, this must be treated as a partnership. We must create effective, efficient, fast-acting technology transfer mechanisms so that our basic research can be fully utilized. We. therefore. recommend that the bill specifically call for the creation of a separate. responsible Technology Transfer Program Office to insure that basic research is translated into products to be used to further all of our goals. In the area of education, the United States needs a great deal of assistance to help us remain competitive. The bill's provisions to educate and train additional undergraduate and graduate students in software engineering, computer science, and computational science and to provide researchers, educators, and students with access to high-performance computing are extremely worthwhile. However, the intent of the bill should be applied across the board in the supercomputing industry and should include mechanical engineers, packaging engineers, chemical engineers and others. Summary In summary, I recommend this bill to you. The amount of funding called for by this bill is indeed small when compared to the significant economic benefit the program will bring to U.S. industrial competitiveness. It is essential that the United States remain aggressive in the area of supercomputer technology. This bill will combine the resources of U.S. industry, government, and universities to meet the challenge of foreign competition.