previous | contents | next

viii PREFACE


compatibility vary in importance from year to year. Finally, while some of the relationships, such as the time-space tradeoff in adder design, are well under stood, others, particularly those involving marketing factors, are not.

Because no theory exists to undergird this multidimensional design problem, we believe that there is no substitute for an extensive, critical understanding of the existing examples of designed and marketed systems. Therefore, this book uses the case study approach. For examples, we have used the thirty DEC computers that have been built over the twenty years that the company has existed, plus some PDP-1 1-based machines built at Carnegie-Mellon University. Carnegie- Mellon's machines explore interconnect structures that we feel will form the basis of future generations.

The association between DEC and Carnegie-Mellon has produced not only some interesting machines to examine but also some of the written material for this book. People in universities can and do write, whereas engineers directly involved in design work are less inclined or encouraged to publish their work.

A substantial portion of the material contributed by DEC authors is historical. We strongly believe that historical information is worth the expense in terms of writing, reading, and learning; machine design principles and techniques change slowly. In fact, the machines currently being designed are based on principles that have been understood and used for years, and we are often asked, "Are we running out of design issues?" Yes, we feel technology provides the forcing function for new designs, not new principles.

Learning about design is always important. Although new designs often appear to be a reapplication of old principles, in the process of being reapplied they change and go beyond their first application. Design is learned by examining and emulating previous designs plus incorporating general principles, new use, and new technology. Indeed, the microcomputer developments draw (or should draw) extensively from the minicomputers. As we build new structures, we should be able to avoid the pitfalls of the immediate past design.

We have intentionally restricted our scope to DEC computers. The reason is obvious: we can speak with first-hand knowledge. If we had used other companies' designs, our data would have been less accurate, and some factors, e.g., design styles, would have been omitted. The main reason, however, is a key part of the philosophy of the book. To understand machine design evolution, the effects of changes in the underlying technologies, and time-invariant principles, we must analyze a family beginning at birth and follow it over several generations of technology. Four series of DEC computers allow such an analysis. DEC computers also provide an opportunity to study another dimension of computer engineering - the coexistence of complementary (and sometimes competing) products. Particular design efforts must compete for resources (design talent, manufacturing-plant capacity, and software, marketing, and sales support). DEC computers have, in general, been designed to be complementary and to avoid overlapping or redundant products. Thus, another set of constraints can be seen at work in the design space.

previous | contents | next