Dual-Use Technologies and Export Control in the Post-Cold War Era (1994) / Chapter Skim
Currently Skimming:
High-Performance Computing: Countrollability and Cooperation
High-Performance Computing: Countrollability and Cooperation
Pages 35-56
The Chapter Skim interface presents what we've algorithmically identified as the most significant single chunk of text within every page in the chapter.
Select key terms on the right to highlight them within pages of the chapter.
From page 35... ...
vary widely, and due to the extraordinarily rapid rate of technological advance and diffusion throughout the world, leading-edge technologies move into the mainstream only a few years after their introduction. In the past, control efforts have consisted of measures taken by the United States and its allies in the Coordinating Committee for Multilateral Export Controls (COCOM)
|
From page 36... ...
The Western community should acquire assurance that, under the current conditions of fragmentation and decentralization of lines of authority, the Russians have the ability to establish an effective, civilian, control regime. This paper examines the present nature and inherent controllability of high-performance computing technologies.
|
From page 37... ...
Trends in HPC Over the last half decade in particular, the high-performance computing sector has been highly dynamic, witnessing remarkable advances in performance throughout all classes of machines, component bases, storage devices, architectures, software and software environments, data transmission technologies, etc. While systems with traditional vector-pipeline architectures continue to evolve steadily, the emergence of a variety of commercial massively parallel machines with performances in some instances rivaling or exceeding those of traditional vector-pipelined supercomputers has signaled an important era of transition in the field in which massively parallel processors (MPP)
|
From page 38... ...
Two of the most significant hindrances have been the complex and cumbersome political and economic structures needed to support the development of complex technology, and, correspondingly, a technology base unable to support the development and manufacture of machines with worId-class performance. In particular, the Soviet/Russian microelectronics industry has been unable to achieve large-scale, reliable production of chips with less than I.5 micron technology (approximately 30,000 transistors per chip)
|
From page 39... ...
At the same time, many Russian policy barriers to international contact and cooperation have been lifted, making it likely that the Russian HPC community will in the future be more integrated into the world HPC community. Controllability of HPC High-performance computing systems have become a particularly problematic element of the export control regime.
|
From page 40... ...
In order to keep manufacturing costs Tow and leverage the enormous amount of research done in the workstation and personal computer sectors of the computer industry, Intel has sought to use commercial and non-exotic technologies to the greatest extent possible. For example, the iPSC/860, introduced in 1990, is constructed using the i860 commercial microprocessors, commodity CMOS memory components, the 5.25" disks used in most workstations, as well as widely used I/O, networking, operating system, and computer language standards.
|
From page 41... ...
Actual performance depends directly on the efficiency with which system resources are managed and data are moved from one location to another. Intel has invested hundreds of man-years and millions of dollars researching the most appropriate ways of managing system resources, taking advantage of the computing power the hardware offers, decreasing software development time, and providing computational results in a useful form.
|
From page 42... ...
RS/6000 CLUSTERS The RS/6000 workstation clusters recently announced by IBM represent an alternative path to high performance computing based on commercially available technology. IBM has widely advertised the fact that a cluster of RS/6000 workstations ~,nr~lantec!
|
From page 43... ...
Gordon Bell, the well-known former Vice President for R&D at DEC and founder of the Gordon Bell Prize for achievement in parallel computation, states that "Important gains in parallelism have come from software environments that allow networks of computers to be applied to a single job. Thus every laboratory containing fast workstations has or will have its own supercomputer for highly parallel applications.
|
From page 44... ...
Workstations, with global production rates approaching half a million annually from a dozen or more vendors, are not easy to control; neither will computing clusters based on them be. While leading vendors are cooperating with government policies to limit direct sales of restricted technology effectively, the installed base is so large that controls are "leaky" at best.
|
From page 45... ...
Application Domains The applicability of high-performance computing applications to military concerns varies considerably. Confidence-builA~ing measures should initially focus on applications which have little importance to the military and gradually move towards those which are marginally important.
|
From page 46... ...
. Other application domains having a greater, but indirect, relevance to military capability are crucial to economies in general, and the Russian economy in particular.
|
From page 47... ...
. Finally, there exist application domains which have great military importance, but marginal economic importance.
|
From page 48... ...
A higher risk might well be suitably controlled at public, centralized, tightly managed, international computing centers sponsored and managed by individual governments or international agencies such as the United Nations for the purpose of providing advanced computing resources for non-military researcb~in specific application domains. Over time, the center could expand the user community, possibly offering time to the international community on a competitive basis.
|
From page 49... ...
Third, the Russian government has stated that it intends to maintain significant military capability and the ability to reconvert enterprises to military production if necessary. Under these circumstances it is difficult to envision a civilian authority which could effectively control the diversion and proliferation of high performance computing technologies, even though in principle structures and procedures like COCOM could be established within Russia as they have been within Hungary.
|
From page 50... ...
Confidence building measures involve a series of steps which increasingly reduce hard controls and then soft controls. Soft controls in most cases wall have to be in place longer than hard controls to provide objective verification that violations have not occurred.
|
From page 51... ...
The Supercomputer Safeguards Plan requires extensive soft controls to be used in conjunction with the hard controls. These include maintaining in a secure fashion usage logs and inspecting them daily, detecting attempts to gain unauthorized access, recording execution characteristics of enr.h program con anal the monitoring of CPU and T/O usage.
|
From page 52... ...
As their contribution, in the interests of mutual cooperation in advanced high-performance computing technologies, the Russian scientists can adapt these methods to Western machines. At the first stage of a joint project, a team of Russians would undergo training at a Western university in software development for a particular Western massively parallel system.
|
From page 53... ...
A second example could be oriented towards the creation of a prominent computer center which would provide computer time to individuals conducting civilian research in a variety of application domains. At the first stage, a low-end, general-purpose machine from a leading Western supercomputer manufacturer could be installed at a prominent Russian university or Academy of Sciences computer center under the exclusive control of representatives of Western export control organizations and the computer's manufacturer.
|
From page 54... ...
27-47. Gartner Group, Inc., High-Performance Computing and Communications: investment American Competitiveness, prepared for the U.S.
|
From page 55... ...
TABLE I: RECENT U.S. HIGH PERFORMANCE COMPUTERS | Year Company Machine Proceeds VECTOR PIPELINE SUPER COMPUTERS Convex C-3 Series 1-8 34.4-960 MFlops Cray Research Y-MP C-90 16 16 GFlops Y-MP/EL 1-4 133-532 MFlops 1992 Cray Research Y-MP M-90 2-8 666-2664 MFlops MASSIVELY PARALLEL COMPUTERS 1991 Intel Paragon XP-S 66-4096 5-300 GFlops Thinking Machine CM-5 32-1024 4-128 GFlops KendallSquare KSR1 32-1088 40Mflops/node 1992 nCUBE nCUBE 2E,2S 8-8192 27-34000 MFlops OTHE 1991 IBM Power System 8-32 1280 MFlops Convex.HP Meta Series IBM RS/6000 Cluster WORKSTATIONS 1992 Sun Microsystems SPARCStationlO 4 400 MIPS SPARCStation2000 2-20 100-2190 MIPS ADD-IN BOARDS 1992 Transtech TTM110 1 60 MFlops PARAStation 4 240 MFlops Sky Computers SKYbolt 16 960 MFlops CSPI Supercard Quad860 4 320 MFlons 55
|
From page 56... ...
TABLE 2: RECENT SOVIET HIGH PERFORMANCE COMPUTERS Year Machine Processors Performance Status 1992 El'brus 3 1-16 34.4-960 MFlops near prototype ElektronikaSSBIS 1-2 16GFlops prototype 1990 MKP 1-2 1 GFlops prototype 1988 PS-2100 64-640 1.5 Gips (32bit) series production 1985 El'brus 2 1-10 125 MIPS series production 56
|
Key Terms
This material may be derived from roughly machine-read images, and so is provided only to facilitate research.
More
information on Chapter Skim is available.