on 6 December 1998
This is another example of the best of the O'Reilly computing series. It provides a good introduction and overview of the current state-of-the-art in high performance and parallel computing. There is a clear historical perspective on the development of the hardware, leading to the current generation of parallel processing on microprocessor based systems, with some anticipation of the next few years likely developments at both processor and system level. The book is very up-to-date, and that may be its one long term weakness - new editions may have to be produced regularly to keep up to date and current (I had the 2nd Edition). There are some small signs towards the end of the book that it has been rushed into print when some minor lapses in the proof-reading of the volume become apparent, but this does not spoil the message. It is a price worth paying for such an up-to-date perspective.
Hardware and architectures are usefully discussed, with a lead-in to advice on benchmarking and an introduction to the associated high performance computing languages and extensions available. The authors show a positive emphasis on the real needs of the end user and/or system evaluator, without indulging in much academic puritanism, or any obvious bias in favour of particular systems. If anything, I would have preferred a few more subjective opinions about what have been good and bad systems and combinations, but the authors would rightly point out that different systems can behave very differently depending on their application.
For someone involved in purchasing or specifying HPC kit for engineering or similar applications, this would provide extremely valuable background information and knowledge. For someone like me, who has moved into the HPC field out of necessity rather than having a strong formal or educational background in the subject it is particularly useful.
Whilst some of the programming subtleties are a bit beyond my needs (or understanding), I regard the information in this book as a strong addition to my general knowledge in this field, and will help me make more informed decisions when called to evaluate HPC hardware and software in future.