Customer Reviews

Beowulf Cluster Computing with Linux (Scientific and Engineering Computation)

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

 

 

I seldom review books, but this one is so disappointing that I felt obligated too. The key thing to note about this book is that Thomas Sterling is not the author, but the "Editor" of the book. Sterling assembled a number of short, loosely related articles and tries to pass them off as a comprehensive book on beowulf clusters running on linux. This book offers no real guidance on building a beowulf cluster or working with an existing one. At best, this book only offers a very general overview of cluster computing.

My supervising professor bought this book for me to use in my senior project and I'm glad it wasn't my purchase. My major complaint is that the code examples in the PVM section are trash. I'm not sure if Al Geist (one of the co-authors of PVM) was rushed or thought that it would be good for students to find and correct errors in the code. Perhaps the best part of the book is Chapter 18 which details the experience at Argonne National Lab of setting up their (massive) cluster, Chiba City. There are many better books out there. Try Parallel Programming by Berry Wilkinson and Michael Allen, Parallel Programming with MPI by Peter Pacheco, or anything by William Gropp.

This second edition covers a large range of cluster topics. Firstly, in terms of why you might need a Beowulf. Then, how to assemble the hardware. And how to install linux on the nodes. The book is valuable not least because it encapsulates decades of knowledge about running clusters, not just of the Beowulf type. The chapters' authors hail from places like Argonne and Oak Ridge National Labs, where there have been research groups using many different types of clusters for years.

Aside from linux, the book describes the special purpose parallel language called MPI. It can be accessed from a C, C++ or Fortran program. Note this choice of languages. Typically the cluster is meant for intensive computational use, with UI issues a low priority. So the book focuses on those languages, and not how to hook Java or C# to MPI.

Of course, with each chapter written by a different author, there is somewhat of a discontinuity between chapters. Think of the book as more akin to a conference proceedings writeup.

I've read this book cover to cover, and then gone back to various parts throughout the course of building a large beowulf. It's the best resource I've found so far. I've recommended it other folks who have also had the same experience--it *explains* beowulfs in very clear and readable language. Excellent primer.