previous | contents | next
Chapter 6 Structure
We now turn from function and performance, which provide design constraints and objectives, to the dimensions of structure, which provide the space in which the design is actually cast. A structural dimension is one in which the designer can attain any of the values along the dimension by relatively direct means. Thus a machine is completely specified by listing all its values along the structural dimensions. From this, the system's function and its performance within that function can be determined.
What dimensions should be selected for structure? The view point is distinctly different from that of performance, where one averages and combines many features to summarize effective output. This tends to obscure structure. For structure, one wants maximally independent aspects which are easily obtained if selected as a design choice. For example, a computer designer who had only a single dimension to describe a computer would undoubtedly select the logic technology used in the Pc and K's; this tells a good deal about many aspects of the computer's structure. In fact, the technology and the average number of bits processed per second by the Pc are correlated, and so each can be used to predict the other, though only imperfectly. If one is interested in performance, the effective number of bits per second is preferred; if one is interested in design, technology is preferred.
The computer space in Table 1 in the introduction to this section presents our choice of the major structure dimensions. There are fewer rationales to validate the choice of dimensions here than there are for performance. Nevertheless, there are a few hallmarks. Perhaps the most important is redundancy (the opposite side of the coin from independence, mentioned above). Several dimensions of structure may covary, so that giving any one of them is tantamount to giving the others. This covariation need not come from physical dependence; it may arise from the nature of an appropriate design and good engineering practice. Such a cluster of covarying dimensions is likely to indicate an important dimension (which one among the correlates is to be used is a secondary matter.) Table 1 in the introduction to this section is organized in terms of such clusters, with one of each selected as the main representative and placed at the left. The following subsections discuss each of the seven clusters of covarying dimensions in turn.
Among the technology dimensions are generation, component complexity, and date. These dimensions, which were briefly mentioned in the introduction to Sec. 2, will be explored in more detail below. Also listed are Pc speed (operations per second) and cost (dollars per million operations), both of which vary directly (or inversely) with logic technology. The Pc operation rate is strongly correlated with logic technology, as we have indicated in the computer space. Our discussion about technology and generations is also about operation rate. The principal reason for the higher operation rate is faster logic technology. Technology also has a secondary effect on increasing speed. More reliable devices allow large computers to be built. Smaller devices allow higher device densities, thus decreasing stray capacitance and inductance and shortening transmission delays. Smaller components also allow increased interconnection density.
Operation rate is relatively highly correlated with total performance. If we hold the structure and parallelism constant, the simplest way to increase performance is by increasing the clock rate. The increase in the performance/cost ratio over the past three decades of computers' evolution has made their primary gains through higher operation rates.
We have indicated only a few of the dimensions that are correlated with technology. In fact, the only dimensions in Table 1 of the section introduction that are independent of technology are the word length and the Pc addresses per instruction. All the rest show dependence on technology. For some, such as memory speed and size, there is a direct correlation. For others, such as PMS structure and parallelism, the development of more complex versions-the leading edge, so to speak-depends on technology, but there is free use of all versions that are in existence at any given time. There are still other dimensions of importance, not shown in Table I of the section introduction, that have also changed with technology, e.g., electric power consumption.
A comparison of the machines in a common computer family will reveal both variations and factors independent of technology. The simple two-parameter model involving Pc microcycle time (a function of technology) and Mp memory pause time (a function both of technology and system design) in Chap. 5 is applied to the System/360, System 370 (see Chap. 52), and PDP-11 (see Chap. 39) computer families. The model is able to explain most of the variation between the family members. And in the case of the System/360 and PDP-11 families, the dominant term is Pc microcycle time, which is almost wholly determined by technology.
Throughout this section we have referred to technology as the dominant factor in the computer. Does this mean that computer development waits upon new fundamental windfalls? We have been lucky in getting the transistor and, to a lesser degree, the integrated circuit from external efforts. However, core memories were invented for the computer and resulted because of need. Read-only memories have also resulted both from development at
previous | contents | next