PART 2
The Instruction-set Processor: main-line computers
To have a "main line" of computers is to have a family that predominates through the generations. Predominance can probably best be measured by the percentage of distinct computers produced within the family, as opposed to outside it. Members of the family need not all be identical; especially evolution over time can be tolerated. But it must be the case that there is at any moment a "standard" design which is seen as emerging from the just prior "standard" design.
Within these definitions there indeed has been a main line in computer systems. It is based on the Burks, Goldstine, and von Neumann memorandum, reprinted as Chap. 4. The most striking characteristic is the evolution from 1 address organization (1), through index-register (1 + x) to general-register (1 + g) organization. Left outside the main line have been multiple-address organizations, character machines, and stack machines. This seems to be an appropriate description, even though a character machine (variable-length character string), the IBM 1401, probably holds the record for number of machines produced (when each model of the IBM System/360 is counted as a separate computer).
A second characteristic feature has been the PMS structure, which has evolved from a single P to a Pc-nPio structure. This has not been uniform within the family, since it applies only to the larger members; the small machines, such as the PDP-8 (Chap. 5), have no separate Pio's. It might seem that all computer systems, both within and without the family, have evolved in this same way. But this disregards the history of computer development. For a while, in the early fifties, there were seen to be two main lines of potential development: scientific computers, featuring large computation and small input/output, and business computers, featuring small computation and large input/output. The latter started to develop into the Pc-nPio structure (with the IBM 702) but, instead of a separate line developing, scientific computers (with the IBM 704 and UNIVAC computers) adopted the more powerful input/output structure. Again, despite its success, the 1401 has not bred a new generation of computer systems in its image, either within IBM (where one might argue that the overriding consideration was to have a uniform series) or by IBM's competitors.
A third characteristic of the main line is the use of binary as opposed to decimal as the basic radix of the machine. This affects both the arithmetic and whether logical processing (on bit vectors) can be done. The issue seems almost settled in the third generation, with smaller machines being binary and larger machines having multiple data-types. The last serious venture into a large pure decimal machine was the UNIVAC LARC, delivered in 1960. In retrospect, the difference in organizations between binary and decimal machines seems small enough so that we have included them all in the same section.
There are a number of striking features that are characteristic of the main line but do not differentiate it from any of the alternatives that have actually been produced. These features include the stored-program concept; the use of sequential
87