previous | contents | next

40 Part 1½ Fundamentals Section 2 ½ The Computer Space

Table 1 Discipline/Environment-Based Functional Segmentation

Scientific ~, engineering, and design

Commercial environment

Manufacturing control environment

Communications, office, and publishing

Transportation systems


Home (using TV set)

~ Implies continuous program development.

Adapted from Bell, Mudge, and McNamara [1978].

tion times was to count the number of multiplications in a program; all else could be neglected. The arithmetic unit has developed to where the floating-point multiply is hardly more expensive than the floating-point add. This requirement for fast arithmetic, however, has really been directed at the logic design level, not at the ISP or PMS level. Thus, the main effect at the ISP is the adoption of long word lengths, floating-point data-types (in addition to integers), and an extensive repertoire of arithmetic operations in the ISP. The main PMS effect is the emphasis on the classic "statistical clerk" PMS design.

The press for increased arithmetic processing has led in recent times to the development of various forms of Pc concurrency, as in the look-ahead of the IBM System/360 Model 91 (Chap. 18) and the n-instruction buffer of the CDG 6600 (Chap. 43). This might be considered a unique functional specialization for scientific computation. Although the needs for scientific computation initiated the exploration of concurrency and parallelism, these functions are applicable in all computers above a certain power, whatever the task domain. Indeed, even microcomputers prefetch instructions. Physical limits on component speed and signal propagation will make these techniques universally attractive.

A better case for permanent specialization can be made in the special-algorithm computers, which compute the fast Fourier transform or do vector operations. Here we finally have systems whose whole design is responsive to a narrow class of problems. This may extend to the very special kinds of Pc parallelism exhibited by the ILLIAC IV (Chap. 20), although there is substantial attempt at generality in such systems.

Whereas early scientific computers dealt mainly with numerical data-types, their use has grown to include text and graphics. In engineering applications, the scientific computer has evolved to a sophisticated notebook for keeping specifications, designs, and scientific records.


In the early days of electronic computing it was felt by many that there was a major functional separation between business computing and scientific computing.1 Scientific problems were "large computing-small input/output"; business problems were "small computing-large input/output." Historically, the IBM 701 scientific computer, for example, used the Pc to control everything dynamically, actually catching the bits from running tapes on the fly (by executing well-timed small loops). These design efforts for business computers resulted in the IBM 702 (and subsequently the IBM 705, 708, and 7080). This machine had two major innovations for IBM: it used characters, and it had a PMS structure that permitted more flexible and voluminous input! output. The latter feature was immediately incorporated into scientific computers, e.g., into the 709, and then into all large

1Such feelings are still extant. Whatever the validity of such feelings, the important consideration is their effect on a particular period of computer development.

previous | contents | next