Information & Communication
- MIMD (Multi Instruct
- Multi Data Stream)
- SIMD (Single Instruct
- Multi Data Stream)
- VLIW (Very Large Instruction Word)
- Systolic arrays
- Specialized parallel coprocessors
- Parallel data storage architecture
Parallel processing is the capability to conduct simultaneously a large number of computing functions offering significant advantages in terms of speed and capacity. There are now several companies marketing so-called massively parallel processors (MPP), consisting of from tens to thousands of individual processors and memories interconnected by a variety of methods within the same machine. Successful exploitation of particular multiple processor architectures such as systolic arrays and hypercubes remains a real challenge to the U.S. research community. In addition to technical difficulties with memory capacity, another difficulty facing the MPP field is the lack of appropriate software. These systems are still very difficult to program efficiently, in spite of considerable research investment in the relevant software over the past decade or more. This lack may limit the use of such machines to a small (but important) niche market. Even so, massively parallel processing offers considerable promise, and research into this area continues to flourish both in universities and commercial companies.
Among the specific parallel processing computing technologies, the one that has drawn the most interest over the past few years has been the so-called MIMD (for multiple instruction, multiple data stream) machines. These consist of from tens to thousands of individual processors and memories interconnected by a variety of methods within the same machine. Many startup companies, often with federal government support, have been formed to exploit this technology. This technology has the advantage over the traditional vector supercomputer in that it is cheaper to build and scales more easily to larger capacity. In particular, there is a consensus that this is the only technology with the potential to reach teraflop speeds (a million million floating point instructions per second) in the foreseeable future.
Massively parallel computing has direct applications in national defense warfighting and weapons control, war gaming, in the Partnership for the Next Generation Vehicle, and the Global Climate Change and Human Genome research programs. While commercial applications are further off than research applications, parallel processing can have a secondary impact on the capabilities of U.S. science and engineering to maintain world class status.
The U.S. has a technology lead in nearly all aspects of high- performance computers. Japanese computer firms continue to lag their U.S. counterparts in parallel computer technology and are facing mixed prospects. Japanese computer firms-- primarily Fujitsu, NEC, and Hitachi--possess strong capabilities in some key supporting technologies. These firms have, for example, outstanding semiconductor component and circuit interconnection capabilities that could give them a distinct advantage over many of their U.S. counterparts in the hardware area. Japanese capabilities in components and board-level interconnection designs give them tremendous freedom to design innovative architectures. However, Japanese firms must overcome some tough technical hurdles before they can develop commercially successful parallel computer systems.