|
|
|
|
|
|
|
This book stresses the nature of design tradeoffs through the use of extensive design studies and examples, and the use of analytic tools in making these tradeoffs. |
|
|
|
|
|
|
|
|
There is a debate in the engineering design community on the relative merits of evaluation by analysis versus evaluation by simulation. Both techniques have a place, but surely analytic analysis takes precedence, at least in an academic treatment of the subject. The design space for computer architectures is enormous. Unless the space can be shrunk by making tradeoffs using analytic approximations and analytic evaluations, it would be hopeless to arrive at an optimum design. As pointed out by a number of researchers, simulation based upon small kernel programs frequently gives erroneous results. The problem with analysis, of course, is that analysis can be very complex. The underlying assumptions of the analysis (a certain probability distribution, etc.) are frequently not met. I have chosen a first-order analytic approach that we use reasonably consistently throughout this text. This approach targets the most obvious bottleneck or constraint in the system, represents that constraint in analytic form, and then studies the effect of that bottleneck on the rest of the system. This, coupled with conservative design data (what I have called the design target data, following the usage of Alan Smith), enables reasonably sophisticated studies of tradeoffs in complex design situations. |
|
|
|
|
|
|
|
|
The last chapter, Processor Studies, could be a subtitle for the entire book. This chapter brings together many of the exercises and studies used throughout the chapters into a study of two complex design situations. As the reader will see, minor variations of many of the parameters in these studies could lead to significantly different conclusions than those presented. Concepts, data, and evaluation form the bedrock of any reasonable design study. |
|
|
|
|
|
|
|
|
This book has a lengthy history. It was started about ten years ago, and has undergone ten ''editions" over these years as course notes for EE382 at Stanford. Advances in the field kept me always one year away from a final edition. But, in doing so, it helped me to rethink both the content and presentation of materials many times, and, as a result, I have tried to abstract the essentials and stress design methodology. Indeed, the text avoids extensive compendiums of current features of various processors or technologies, just as it stresses concepts that underlie these processor designs. |
|
|
|
|
|
|
|
|
How to Contact the Author |
|
|
|
|
|
|
|
|
Comments on the book and error detection may be directed electronically to flynn@ee.stanford.edu, or mailed by way of the Electrical Engineering Department, M/C 9030, Stanford, CA 94305. |
|
|
|
|
|
|
|
|
http://www-flynn.stanford.edu will point to a current page about this book. |
|
|
|
|
|