Where Industry and Security Intersect
What's New | Sitemap | Search
Mr. Chairman, thank you for the opportunity to testify before the Committee on export policy for high performance computers. The debate over computer export policy has been marked by a good deal of smoke and heat, but perhaps not as much light as one could wish. I will try today to explain on the Administration's policy. Its fundamental premises are that like it or not, rapid technological progress has rendered control of high performance computers increasingly difficult, and that it is more important to our national security to have a healthy computer industry supplying state of the art products to our military and intelligence services than it is to attempt to control the uncontrollable and jeopardize our companies' futures in the process.
Four factors have shaped our thinking. The first is technological change. Improvements in microprocessor design have allowed high performance computers to become ever smaller, cheaper and faster. At the same time, improvements in microprocessors have made routine desk top computers capable of performing at were considered supercomputer levels a few years ago. The second factor is global diffusion. We must assess realistically our ability to control the distribution of computers when they are produced in the thousands or even tens of thousands and sold from a variety of sources around the world. Third is the growth of parallel processing , which greatly increases computer performance, and the concomitant ability of users to easily upgrade performance. Finally, there is our conclusion, based on research and the 1995 and 1998 studies, that computer power is a secondary consideration for many applications of national security concern. These factors -- rapid technological change, limited controllability, scalability and limited national security application -- have shaped our efforts to keep our policy in tune with today=s technology and international security environment.
In doing so we have kept in mind the nature of the computer market, which is a vital element of U.S. economic strength. We are world leaders in the very competitive computer market with
$2 billion a year in revenue, and this leadership helps us across the board in the information technology sector. The policy adopted by the U.S. in 1995 affected more than ten billion dollars in exports, which supported 140,000 jobs annually. If misapplied, export controls can profoundly damage this important sector, put these jobs at risk and seriously damage our national security by crippling our companies just as our national security establishment=s reliance on them grows.
The competitive and increasingly global market has strong implications for export controls. Roughly half of the computers made in the U.S. are exported, and the sales fall in the ranges below:
-- Computers capable of up to 400 MTOPS have been sold in the millions.
-- Computers capable of 400 to 1000 MTOPS have been sold in the tens of thousands.
-- Thousands of computers capable of 1000 to 5000 MTOPS range have been sold.
-- A few thousand computers capable of 5000 to 10000 MTOPS have been sold.
-- Some hundreds of machines capable of more than 10000 MTOPS have been sold.
Some of these computers can be reconfigured by their users to have much higher performance, and in the future, in response to market demands, more and more computers will be scalable. Our fundamental reality is that computers which are available in the thousands in markets around the world cannot be effectively controlled, even if they are built in the U.S. or based on U.S. technology. The 1995 study predicted many of these developments, and everything we have learned since then confirms them.
Let me turn now to the issue of technological change. Technological change means that computer performance is constantly improving, creating unavoidable pressure on export controls. In few other areas has the pace of technological change been so rapid and so dramatic as computers. Five years ago, the U.S. controlled as a supercomputer machines with a performance of 195 MTOPS. Today's average desktop PC is more powerful, and the software which can be run on it more sophisticated.
The engine of change is the microprocessor. Computer chips are produced in the millions in plants in the U.S. and overseas. You are all familiar with Moore's Law, which states that the performance of chips doubles every eighteen months. These performance increases are the result of both improved design and improved manufacturing techniques. As of August 1998, chips capable of roughly 500 MTOPS are being produced in the millions and chips capable of 1800 MTOPS are being produced in the tens of thousands. Although the U.S. is the most advanced producer, plants around the world can make these chips. Within twelve months, if industry projections are correct, we can expect to see chips capable of 2000 MTOPS enter into mass production. When this occurs, we will see sales of 2000 MTOPS computers numbered in the thousands, making the limitations on our ability to maintain effective controls even more obvious than they are now.
Other technological changes have made it easier to upgrade performance. These include the increased sophistication of software, the increased availability of interconnect technologies which offers substantial improvements in performance and which may allow numbers of low level workstations to be clustered together to give high performance. The spread of parallel processing, which allows many microprocessors to work simultaneously on the same problem, has also reduced the controllability of high performance computers.
Rapid advances in microprocessors, software, interconnects and parallel processing mean that the performance levels once associated with giant machines can now be obtained by smaller and relatively inexpensive computers. The implications of technological progress go beyond performance. High performance computers are smaller, cheaper, simpler to install and maintain and more reliable. These attributes are desirable in the marketplace, but they degrade our ability to control.
Another element of technological change could be called attainable performance or scalability. Manufacturers have sought to build platforms which can be easily upgraded through the addition of new boards. This allows users to buy computers at one performance threshold and then increase the performance later through upgrades. Some computers are designed to allow these upgrades to be performed without even turning off the machine and with system software that
automatically adjusts to the higher performance levels. The result is that it is possible to buy a number of systems that perform well below 2000 or even 1000 MTOPS, and thus do not require a license for export, and then upgrade these machines to 5000 or 6000 MTOPS or more.
These technical developments pose real problems for controllability. Faster chips available in the millions; smaller, cheaper and more reliable computers with performance up to 7000 MTOPS and computers which can be exported without a license and then upgraded to HPC performance, all have created serious limitations on our ability to control computer exports.
Foreign availability -- the availability of high performance computers built by foreign manufacturers with foreign parts and technology -- was a key determinant of our export policy during the Cold War. It now makes little sense and is of secondary importance in determining policy. We cannot realistically control the many thousands of U.S.-made computers sold freely in Europe, Asia and elsewhere. Many countries we sell to do not have re-export controls -- in fact the New York Times recently quoted an official from a close European ally as saying that they advise their exporters to ignore U.S. re-export controls. We know there is a flourishing market in secondhand high performance computers overseas -- some can be ordered directly over the Internet. As a result, even though the U.S. today dominates the market for high performance computers, there is a performance threshold below which we cannot realistically expect to maintain control of computers unless we restrict sales to even our closest allies.
That is why we have focussed on controllability - whether licensing can be effective in restricting access to high performance computers. The studies in 1995 and this year suggest that HPCs are becoming less and less controllable because they are becoming smaller, cheaper, more powerful, and more reliable, requiring less vendor support. The availability of fast, well-designed microprocessors has allowed manufacturers to build more and better computers. Machines capable of 4000 to 5000 MTOPS are small and easily transported. Computers well above 2000 MTOPS are freely available on the global second hand market. We cannot realistically expect to keep the organizations responsible for weapons development in states of concern, organizations that are technically sophisticated, well funded and which enjoy strong government support, from clandestinely obtaining computers with a performance much below 10,000 MTOPS.
Computers are not a choke point for military production. High performance computers have attained a symbolic importance in our export control debates which their real utility may not warrant. The fundamental premise of critics of our policy is that high performance computers are essential for making advanced weaponry. This critique is wrong. The weapon systems found in the U.S. arsenal today - the tanks, airplanes, missiles and ships -- were built with computers whose performance was below 1000 MTOPS. These were the supercomputers of the 1980's, but today you can find more capable machines on many office desktops.
We have found that the amount of computing power needed to design and manufacture modern weapons, once you get over a few hundred MTOPS, is not significant. For example, the level of computational power used to develop all the bombs in the current U.S. nuclear arsenal is less than that found today in many workstations. Other factors -- skill in software design, access to sophisticated manufacturing techniques, experience in building weapons and good test data -- are much more important than a high performance computer.
There are a number of applications -- precise weather forecasting, computational fluid dynamics, and particle dynamics in particular -- where high levels of computing power are significant, and our policy attempts to identify those and, where possible, protect them. This differs from those who have argued that high performance computers will give countries like China the ability to leap forward in military production. While HPCs no doubt provide some incremental benefit, as would a wide range of items, we do not believe they constitute a choke point in weapons development, and, as stated previously, even if they did, there are serious limitations on our ability to control them at all but the highest levels. There is no evidence that you need a high performance computer to make most modern weapons, including nuclear weapons, or that having access to high performance computers alone will give you improved military-industrial capabilities.
In fact, none of the nonproliferation regimes, the Missile Technology Control Regime, the Nuclear Suppliers Group, or the Australia Group consider computers important enough to control. The members of these regimes decided that computers are not essential for production of these weapons systems. The only regime which controls computers is the Wassenaar Arrangement, which inherited the old economic warfare controls aimed at the Soviet Union. These controls did not, in the end, work very well in preventing the Soviet bloc from getting its hands on widely available computers. They were helpful in keeping big machines that require extensive support out of enemy hands but failed to stop computers available in the thousands sold freely at relatively low prices around the globe without significant vendor support. We would do well to remember this lesson because what we are experiencing now is essentially an acceleration in the rate that ever-higher performing computers are becoming widely available. In other words, we can still control the high end, but whereas that used to be measured in hundreds of MTOPS; now it is measured in ten thousands.
The U.S. currently dominates the high performance computer market, in part because of the realistic computer export policy we adopted in 1995. Tighter controls would penalize U.S. firms for winning the high performance computer competition. Our firms' strength has driven most producers from all but the low end of the market and discouraged others from entering. Export controls, like any government intervention, can, however, reverse that situation. Controls at too low a level act as a subsidy for our foreign competitors, damage our national security, and cost the American economy exports and jobs.
Maintaining our status as world leader in information technology and computer manufacturing is critical to both our economic growth and our national security. Exports account for roughly half the revenues of U.S. computer companies. Ill-advised export controls would put this vital sector at risk and at the same time compromise our security by making it harder for the Pentagon to obtain the cutting edge technology it needs. Events since 1995 have confirmed we are on the right course, and I hope the Congress will support the Administration as it moves into a new review of control parameters.
In April of 2002 the Bureau of Export Administration (BXA) changed its name to the Bureau of Industry and Security(BIS). For historical purposes we have not changed the references to BXA in the legacy documents found in the Archived Press and Public Information.