Statement of Dr. Stephen Bryen
: "TESTIMONY OF DR. STEPHEN BRYEN ON U.S.POLICY ON HIGH PERFORMANCE COMPUTER EXPORTS TO THE HOUSE ARMED SERVICES COMMITTEE, OCTOBER 28, 1999
I appreciate the opportunity to address the committee today concerning the export of high performance computers. As the committee knows, I managed the technology security program for the Defense Department between 1981 until 1988. I have continued to maintain my interest and involvement in high technology and defense issues. My testimony today reflects my personal observations and assessment of the current situation. In my testimony the terms "high performance computer" ("HPC") or supercomputer are interchangeable.
High performance computers are a vital element in United States industrial, economic, political and military leadership. The United States developed the first high performance computers, and continues to lead the world in the development of high performance computer hardware and software.
High performance computers have many military, industrial and commercial applications in addition to their importance in supporting the development of the sciences.
High Performance Computers and Military Risk
My testimony today focuses on the export of high performance computers where there is strategic risk involved. There are a number of countries embarked on nuclear programs and advanced missile development. High performance computers play a vital role in such developments.
There is a great deal of confusion as to just how high performance computers can do more than somewhat less capable computers. It is often said that the United States developed most of its nuclear weapons and delivery systems on computers that were less capable than today’s high performance computers.
HPC’s and Nuclear Weapons and Missile Development
This is true, of course, but it misses the point. Today’s high performance computers offer the following benefits for nuclear and missile development programs.
the ability to perform tens of thousands of calculations in a much shorter period of time with less errors than previously possible, thereby reducing R&D time dramatically
the ability to make calculations using a computer that could only be done in the past by trial and error, thereby reducing R&D costs dramatically
the ability to test designs without having to build prototypes, saving time and expense and masking development efforts, thereby making it difficult to gauge an adversary’s intentions
the ability to develop new types of weapons –such as those than can penetrate deeply buried missile silos, or develop advanced neutron nuclear devices, thereby shifting the balance of power.
In regard to nuclear weapons, the Natural Resource Defense Council points out that high performance computers can be used to
--maintain the reliability and safety of nuclear weapons without having to rely on extensive nuclear testing;
--design and prototype nuclear weapons;
--"certify" changes to existing nuclear weapon types –thereby making it possible to modify the weapons or adapt them to other kinds of delivery platforms such as ballistic missile nuclear submarines;
--design weapons to be more transportable;
--design better nuclear weapons maintenance; improve performance margins and improve weapons shelf life.
Compelling evidence is beginning to emerge that United States transfers of high performance computing capability is helping potential adversary states achieve the benefits and uses described above. For example, the Hong Kong Standard newspaper reported on October 14th that China has completed laboratory simulations of a test launch of its latest multi-warhead intercontinental ballistic missiles, the DF-41. Pamela Pun, the Standard reporter, notes that "Mainland sources [told her that the] computer simulations of the launches of the sold-fuelled Dongfeng-41 ICBM have been completed and proved successful." If the story is accurate, and there is no reason to believe otherwise at present, then China almost certainly made effective use of the tremendous computer capability that has been transferred there by the United States.
Civilian and Military End Use
There is a convenient mythology that the United States only exports high performance computers to China for civilian use. However, this is merely an assertion not a fact, since the United States has performed only one Post Delivery Verification (PDV) in China between November 1997 and November 1998 according to the General Accounting Office. And even if the end use site is verified, it would not prove the argument, because almost all of China’s high performance computers are operating in national networks, including those of the Chinese Academy of Sciences (CAS), a major nuclear research agency closely integrated in China’s nuclear weapons programs. So long as the U.S. is prepared to export high performance computers to China to organization such as the CAS, it is certain the computers will be used for military purposes. But even if the computers are sent to a Chinese University or a government-owned toy factory, the machines will be available for military R&D.
It is too late to worry about limiting the use of high performance computers in China. The deed has been done, alas: we have deliberately and systematically transferred awesome computer power to an strategic adversary that is openly developing weapons systems targeted against the United States or United States assets abroad.
Impact of HPC Liberalization
There is another myth that the proposed additional liberalization on export controls for unlicensed computers to Tier III countries such as China will only lead to a modest increase in their capabilities. Actually, China is already receiving high performance computers significantly more powerful than those on the control list. The Department of Commerce is simply approving export licenses of machines with a throughput of 36,000 MTOPS or better, even though, as the committee knows, the current decontrol limit is less than 7,000 MTOPS.
The administration proposal to move this to 12,300 MTOPS is misdirection, pure and simple. Machines are simply being licensed to China that are more powerful than 12,300, and the potential for harm to our national security is much greater than the Congress and the American public has been led to believe. This committee should focus on the real exports, not make-believe numbers produced to create an illusion of greater security.
I have focused on China because China has the largest industrial base and already has a well-developed nuclear program and a scientific establishment that is adept at exploiting the value of the high performance computers going to China. Others, such as Pakistan and India, will benefit from these developments as well.
Officially the administration says that the reason for changing the computer controls has to do with the rapid increases in computer power. Desktop computers will soon have as many MTOP’s as today’s supercomputer. Therefore, it is important to "keep pace" with changes in technology, otherwise all we will do is restrict American exports and hand over the business to others.
Definition of High Performance Computers
The current definition of a high performance computer is, indeed, inadequate. The inadequacy is based on the use of a single, badly flawed, parameter to define a high performance computer and differentiate it from a fast, not as high performance, desktop computer. That is not the way to evaluate the power of modern computers. I sometimes have the impression that the administration’s fixation on the single-parameter definition is intended to facilitate dangerous exports to places such as China, perhaps the largest single customer for American supercomputers. This is altogether regrettable and totally unnecessary.
A high performance computer that has strategic importance is one with characteristics that are decisively different from a desktop computer. Among these characteristics are:
HPC’s have tens or hundreds of processors while desktop machines have one or two processors. In the next few years some microcomputers will have 4 or 8 processors, still well short of the tens to hundreds of processors in an HPC machine;
Most of the latest HPC’s are parallel processors. Most desk top machines are not parallel processors;
HPC’s of modern design feature tight coupling between the processors and the memory of the machine; all desk top machines have inferior coupling between the machine processor and memory;
The best HPC’s have random, non-interruptible communication between the machine’s processor and memory; desk top machines have interrupt in their operating system and do not feature such communication;
HPC’s have much more powerful memory access than desktop machines –in fact this is one of their most important features. Memory access takes into account the size of the memory that can be addressed, and the speed memory can be addressed. Taken together this defines the memory bandwidth of HPC’s and distinguishes them from desktop machines. Main memory in an HPC is fully addressable; not so in a desktop.
HPC’s cannot be properly measured by "MTOP’s." The MTOP measurement does not properly register the power of a supercomputer.
There are other less technical distinctions that also need to be factored in. For example, multi-threaded software designed for an HPC won’t run on a desktop, even many desktops linked together and running an operating system such as Linux. A Chinese military operative in many cases is trying to make use of software acquired either legally or illegally from U.S. organizations, such as the Department of Energy. There is no way he can run the software on a Linux box or boxes. He needs a machine "just like" that used by the DOE labs.
Therefore, if we stick with the current single definition of controls on computers, MTOPS, our export control system will not control the right computers and, ultimately, will fail in its strategic purpose even with an administration that is serious about enforcing the rules. It will also continue to require computer producers of desk top type machines to pay a price for strategic export controls that is unreasonable.
Need for a Strategic Evaluation
Linked to this is the fact that the current export control parameters and regime are operating on one wing and a prayer. The wing is nearly broken and, it seems, no one hears the prayer. Why?
The reason is easy to figure out. The current export controls are not based on any coherent strategic evaluation. Our military experts in the government have not been consulted about the threat, have not evaluated the technological and military intentions of countries such as China, and have not been called on to propose controls that might protect America’s strategic interests.
For example, if China can perfect the warheads and penetration aids on its new class of missiles, will that defeat our efforts to build a missile defense, or significantly add to the cost and complexity? Are certain computer exports likely to contribute to China’s ability to rapidly develop such systems? Questions of this kind must be answered, and quickly. Running an export control system without this kind of database is a dangerous mistake.
It is urgent for the Congress to demand such an analysis be carried out. In the meantime it is sensible and prudent to freeze the export of all licensed high performance parallel computers, and block any increase in liberalizing the current de-control, until the Defense and intelligence community can answer such questions and provide coherent and responsible advice on managing the export control system.
Finally I would urge that the Joint Chiefs of Staff be asked to make a serious evaluation of all technology transfers to China, and propose export controls that will leverage America’s advantage in the future. Given the harsh reductions in the American defense budget, the under-investment in military technology and new military systems, and extraordinary delays in getting a serious national ballistic missile defense system deployed, we cannot count on our ability to come through in a clutch as we did in World War II. It is time for a responsible technology security policy."