Book Review
Parallel Scientific Computing in C++ and MPI: A Seamless Approach
to Parallel Algorithms and Their Implementation
George Em Karniadakis and Robert M. Kirby II
Cambridge University
Press, New York, 2003
616 pp., $150.00 hb , $55.00 pb (both
include CD-ROM)
ISBN 0-521-81754-4 hb , ISBN 0-521-52080-0 pb
Amazon | Barnes & Noble
Reviewed by Babak Makkinejad
see all book reviews  New
and challenging problems are being encountered in the areas of data mining, bioinformatics,
and computational fluid dynamics that require a very large computational capacity.
The availability of commodity hardware components such as motherboards and memory
chips, together with free software such as Linux, the GNU compilers, and the
Message-Passing Interface (in which message-passing is used to control the flow
of the computation), has put massively parallel machines such as Beowulf clusters
within reach of medium-sized companies and academic departments. As parallel
computing continues to merge into the mainstream of computing, it is becoming
important for students and professionals to understand the application and analysis
of algorithmic paradigms to both the (traditional) sequential model of computing
and various parallel models.
Parallel Scientific Computing in C++ and MPI , by George
Em Karniadakis and Robert M. Kirby II, is a valiant effort to introduce parallel
scientific computing to the student in a unified manner. The textbook offers
the student with no previous background in computing three books in one. There
is a textbook on the analysis of algorithms, a textbook on parallel programming
using MPI 1.x, and an elementary book on programming using a subset of C++ as
a better “C”.
Karniadakis is a professor of applied mathematics at Brown University, working
on simulations of turbulence in complex geometries. Kirby is an assistant professor
of computer science at the University of Utah, specializing in large-scale scientific
computing. This textbook, largely based on Karniadakis's courses at Princeton
University, Brown University, and MIT over the past 15 years, is thus slanted
toward computational fluid dynamics. It is strong as a traditional algorithms-based
textbook for an introductory course in numerical analysis at the late-undergraduate
or early-graduate level. It examines such core topics as dense and sparse matrix
computations, linear systems, finite differences, and fast Fourier transforms.
The text assumes a solid technical background including calculus, linear algebra,
and differential equations.
The initial chapters explain how and why parallel computing began, present
an overview of parallel architectures, and introduce MPI 1.x. The authors follow
this by discussing the powerful divide-and-conquer paradigm and develop the basics
of each topic, such as root finding and approximation with sequential and MPI-specific
implementation details and much useful (but not optimal) C++ code. Chapters 3,
5, and 6 are the heart of the book, where approximation of functions, explicit
and implicit discretization, and MPI are discussed in detail. The authors are
very careful in establishing the foundation of each algorithm, and considerable
care is taken in explaining and estimating the accuracy of each numerical technique,
its stability, and its convergence with benchmarks. Each chapter also includes
advice on common programming pitfalls, “gotchas,” and exercises. There are, in
fact, 162 homework problems throughout the book.
The authors state the following: “Our book treats numerics, parallelism, and
programming equally and simultaneously.” They do not achieve their stated purpose
of treating these topics equally in their discussions of C++ and MPI 1.x. Readers
looking for examples of how encapsulation, inheritance, exception handling, templates,
and polymorphism can be used to control the complexity of developing, debugging,
maintaining, and tuning parallel software using MPI will not find it in this
book. For a more thorough discussion of MPI 1.x from a software development point
of view using “C,” one might consult Parallel Programming with MPI by
Peter S. Pacheco (Morgan Kaufmann Publishers, 1997). However, neither text treats
MPI 2.0 features such as multithreading or C++ bindings for MPI.
In spite of falling short of its ambitious goals, this textbook is useful
for those who would like to know how to write parallel programs using MPI or
who wish to go beyond such cookbook texts as Numerical Recipes in C++: The
Art of Scientific Computing by William
H. Press , et al. (Cambridge University Press, 1988).
Biography
Babak Makkinejad, a consultant
with EDS, received his Ph.D. in theoretical physics from the University of Michigan
in Ann Arbor. He has worked in the areas of computational physics, computer graphics,
image processing, and enterprise software development. |
