Date: Mon, 8 Jul 1996 11:57:35 +0200 From: sondjaap@horus.sara.nl (Jaap Hollenberg) Subject: Aspects-of-Computational-Science Aspects of Computational Science A Text Book on High-Performance Computing NCF the Dutch National Computing Facilities Foundation (a subsidiary of the Netherlands Organisation for Scientific Research, NWO) has published a text book on high-performance computing entitled "Aspects of Computational Science". The Editor in Chief is Aad van der Steen from the Academic Computer Centre in Utrecht; the complete Editorial Board comprises eight Dutch scientists active in the field who produced new and gathered old material. The book contains a course in the broad field of "high-performance computing", intended for students, graduate or post doctoral from universities and vocational high schools and for technical and research workers. The course is self contained but some global knowledge (e.g. of a high-level programming language) is assumed. The material can be used by lecturers in courses on computers, computer science or computational science, but can also be used by students directly without further guidance. The chapters in the book are interrelated, but lend themselves quite well for separate study. These chapters are: Architecture of high-performance computers ------------------------------------------ In chapter 1 the types of high performance computer architectures are given in a systematic overview. Interconnection structures and memory concepts are discussed. Also the main units from which a system is built are described, such as a CPU, memory components, pipelines, I/O-components, etc. Processor arithmetic -------------------- Chapter 2 deals with the finite representation of numbers (processor arithmetic). Different floating-point representations are in use on the various computer systems, despite the existence of an IEEE floating point standard. Consequence among others are different results and accuracies and different accumulative errors on different machines for the very same problem, using the same code and input. Also vectorisation and parallel evaluation may yield results different from scalar calculations as results of the order of the operations. Programming languages for high-performance computing ---------------------------------------------------- Chapter 3 gives some insight in the process of translating programs into machine readable code. This is done with by compilers that use as input lines of code written according to the rules of a programming language. Programming languages are still subject to many discussions and fundamental changes. Even for the most widely used languages is this the case, and the main cause is the emergence of highly parallel computers. The chapter is not intended as a course in high level languages, C or Fortran. Standard mathematical software ------------------------------ Chapter 4 concerns standard numerical software. Insight is given in the kind of software programs and packages that can be considered as standard tools, the most widely used subroutine libraries and their characteristics and some general technical information about standard numerical tools. Vectorisation and parallelisation of algorithms ----------------------------------------------- Chapter 5 discusses the important aspects of vectorisation and parallelisation, both from a basic and from a practical point of view. General considerations (Amdahl's Law) as well as numerous examples in linear algebra, differential equations and non-numerical algorithms are given. Selected parallel applications and practical elements ----------------------------------------------------- Chapter 6 contains selected applications. It is meant to elucidate the theory by examples of real applications from various application fields. Program development: Preparation, production, presentation ---------------------------------------------------------- Chapter 7 finishes the course material with some general guidelines, do's and don'ts regarding software engineering and maintenance, some remarks about visualisation, and examples of the use of computer algebra in computational science. For more information write to Stichting NCF Postbus 93120 2509 AC Den Haag The Netherlands e-mail: aerts@NWO.nl The book can be ordered by transferring the amount of Dfl 100 net per copy to the Dutch Postbank account number 177018 of the Netherlands Organisation of Scientific Research (NWO), mentioning "NCF textbook 02600100-7525", your name and delivery address and the number of copies desired.