Customer Reviews

Parallel Scientific Computing in C++ and MPI: A Seamless Approach to Parallel Algorithms and their Implementation

5 star:		(1)
4 star:		(1)
3 star:		(1)
2 star:		(0)
1 star:		(0)

 

 

The authors attempted to combine introductory material in C++, numerical methods and parallel computing. That is quite a brave endevaour. They certainly break the new ground writing an introductory material for a "simulation scientist", but I believe they have achieved mixed success.

On the one hand, the material they present on all subjects is really top quality, packed with 100% usefull information. Bibliography is also very good and usefull. But the organisation of the book is quite confusing. They introduce all the topics toogether - throughtout the book. Hence each chapter introduces some numerical algorithms, few new concepts in C++ and eventually MPI. I beleive a novice would experience serious difficulties following it. For example, authors introduce objects before introducing curly braces "{}" as scope delimiters and before semicolon ";" as statement delimiters.

Further, very soon after introducung the very basic concepts in C++, the authors move on to BLAS. BLAS is usefull, of course, but a novice in C++ may wonder why does he needs libraries written in Fortran, if C++ is a language of the choice for numerical computations. (At least it is claimed so by the authors).

Another confusing example is the one of memory memory access. In section 2.2.6 Memory Management, (pg. 41) the authors introduce basic concepts of memory management and how can loop constructs influence the efficiency of the code. Very usefull indeed, no question about it. But very soon bellow, in section 2.2.8 Exploiting the Structure of the Sparse Matrices, they come up with the claim (pg. 58): "... optimization-savy individuals, as the old saying goes, often miss the forest for the threes" :-( Hence, a novice reader might think: "Well, why do I need to worry about the memory management explained just 17 pages above?".

My most serious critic of this book by far (and I hope the authors will read this) are the contents pages. The contents list only chapters and first level sub-chapters. Secind level chapters are not present!!! That makes the book very hard to use as a reference material. That is really a pitty, since there is some good material in it which is hard to find and might stay hidden. (For example, the chapter I mentioned above: 2.2.6 Memory management is NOT in the contents, so I had to browse slowly throught the book to find it and refer it here). I suggest the authors introduce: "Contents at a glance" (the present one) and a "Detailed Contents", where one could find references to all the chapters in the book. The contents is THE reason why I gave this book 3 stars instead of 4. One it lost on the confusing organisation of the book.

I think the authors should have organised the book in four parts: 1 - Numerical algorithms, 2 - C++ and 3 - Parallel computing with MPI, 4 - Advanced topics. Part 1 could introduce numerical algorithms and have pointers to their implementations in Part 2 and corresponding parallel implementations in Part 3. Part 2 and 3 could have started with introductions, which a reader already familiar with those subjects, could skip. Part 4, could bring advanced topics, such as optimisation, BLAS, etc.

Bottom line, it is:

- brave and usefull endevaour,
- full of excellent material,
- organized confusingly,
- and has a very poor contents.

Buy it if you are simulation scientist or teacher, but prepare to struggle with its organisation and contents.

This book is great in describing some of the most important concepts and algorithms needed for the beginning numerical analyst. The book claims that it can be picked up by a complete novice and teach C++, MPI, and scientfic computing. I would say that the math goes very quickly and not quite as rigorous as necessary for the typical novice. The C++ is pretty basic but still the book leaves the reader a sense of confusion. This is largely because the book treats a large amount of the library functions as black boxes. The MPI starts very basic and gradually introduces the major concepts.

My recommendation for anyone reading the book is to supplement it with a good linear algebra book (such as Demmel) and book on C++ (The C++ programming language). For further study on MPI, Using MPI would be a good supplement. That way whenever you have a concept that isn't fully described, you have a source to get it from.

The book gets a high rating for going over the right content and doing so in a applied manner that gives the reader the skills to become a numerical analyst.

The book contains advanced numerical mathematics algorithms and
fundamental elements of parallel computation.
It will be useful for those academic instructors who believe that students should be shown the entire solution process
from mathematical problem definition to computer implementation. It has been used as a textbook at several leading American and European universities.
The authors professors Karniadakis and Kilby are innovators who demonstrate that combining education of applied mathematics with computer science is possible and extremely useful for students and their future employers.