This book sets out the principles of parallel computing in a way which will be useful to student and potential user alike. It includes coverage of both conventional and neural computers. The content of the book is arranged hierarchically. It explains why, where and how parallel computing is used; the fundamental paradigms employed in the field; how systems are programmed or trained; technical aspects including connectivity and processing element complexity; and how system performance is estimated (and why doing so is difficult). The penultimate chapter of the book comprises a set of case studies of archetypal parallel computers, each study written by an individ- ual closely connected with the system in question. The final chap- ter correlates the various aspects of parallel computing into a tax- onomy of systems. [...]... subject of parallel computing and to potential users who want to avoid the many possible pitfalls in understanding this new and complex field 1 Introduction Before attempting to understand the complexities of the subject of parallel computing, the intending user or student ought, perhaps, to ask why such an exotic approach is necessary After all, ordinary, serial, computers are in successful and widespread... this: In order to understand parallel computing, it is necessary to understand the relationships between problems and systems One starting point might be to consider what application areas could benefit from the use of parallelism However, in order to understand why these are suggested as being appropriate, it is first necessary to know something about the different ways in which parallelism can be applied... first such change is between the string of pixels at point A and the two-dimensional array of data (image) at point B, while the second change is between the image data at point C and the vector data at D We would thus expect that, between A and B, and between C and D, devices which can handle two data formats are required, whereas between B and C and after Z>, single format devices are needed Further,... are processed and decisions are made - at the rate at which one computing element executes its own segment of program When processing has been going on for some time, all the processors are working in parallel and the speedup is proportional to the number of processors There are therefore (at least) two ways of implementing functional parallelism and applying it to the problem in hand, and the resulting... data parallelism could be exploited, and at several places functional parallelism could be of benefit In the following sections we shall see how each of these approaches might be used, but it is necessary to continually bear in mind that a programmed parallel computing system comprises three facets - hardware (self-evidently), software (which enables the user to take advantage of the parallelism) and. .. of analysis and evaluation The potential users of a system are as likely to be set on the wrong path as the right one toward fulfilling their own set of practical aims This book is an attempt to set out the general principles of parallel computing in a way which will be useful to student and user alike The approach I adopt to the subject is top-down - the simplest and most fundamental principles are... approach achieves parallel embodiment in structures which are similar to those found in the brain, in which parallelism of data and function exist side by side In programmed systems, however, the two types of parallelism tend to be separated, with consequent impact on the functioning of the system There are therefore three basic approaches to parallel computing which we will now examine - parallel cognition... Results Figure 1.9 A function parallel computation system the proper places on the system, and to coordinate their interactions, but the user need know nothing of this process It is of interest to note that, for both data parallel and function parallel systems, the identification of parallelism is much more easily carried out at the level of a flow chart (Figures 1.5 and 1.7) than within a program... least aprima facie case for the utility of parallel computing In some application areas, parallel computers may be easier to program, give performance unobtainable in any other way, and might be more cost-effective than serial alternatives If this case is accepted, it is quite reasonable that an intending practitioner in the field should need to study and understand its complexities Can the same be said... is likely to involve different data types, and hence differently configured items of parallel hardware, if maximum optimisation is to be achieved 1.1.13 Parallel function Naturally enough, if we seek to implement functional parallelism in a computer, we need a tool which will enable us to analyse the areas of functional parallelism As in the case of data parallelism, we begin with a re-examination . various aspects of parallel computing into a tax- onomy of systems.