How to Build a Beowulf attempts to walk the reader through the process described in the title -- namely, building a Beowulf cluster, all the way from choosing hardware to assembling the systems, installing Linux, configuring the "Beowulf" type environment, and even how to use a parallel programming environment to take advantage of your new parallel computing system.

The book succeeds in doing exactly what it sets out to do. Having been published in 1999, certain information is dated (such as hardware pricing, or the need for floppy disks to install the cluster, as a few examples), but overlooking these simple cosmetic issues, the book is extremely useful for someone relatively new to clustering.

The first two chapters are overview. Chapter one is a brief introduction to the history of Beowulf computing and a bit about the book. Chapter two goes into the overview of a Beowulf system, hardware, software, message passing, and systems management.

The next four chapters get into building the system. First comes a description of the node hardware the reader should look into getting in chapter three. Chapter four is an explanation of what Linux is, how it works, some of the things the reader will be doing, and so on.

Chapter five is devoted entirely to networking hardware and software. The hardware section looks at the different types of network that can be set up. This information is interesting for a brief overview of the different types of networking, though these days, it is clear that basically everyone who wants to build their own personal cluster is using ethernet. The software section of this chapter looks into aspects of networking such as TCP/IP, sockets, high level protocols, distributed objects (CORBA and the Java RMI), distributed file systems, and remote access and command execution.

After the previous three chapters lay out all the pieces one will need to build a cluster, the last chapter of this section, chapter six, titled "Managing Ensembles," is about how to build and configure a cluster (ensemble) system. This book does something that I found refreshing, namely basically saying "install Linux and then do this..." leaving the installation of Linux as an exercise for the reader. More detail is gotten into for such topics as cloning your system, imaging your slave nodes, account management, and some simple security topics (ssh, restricting host access, IP masquerading). Again, we find a little outdated information here. A decent segment of this chapter is about how to clone your slave nodes. At the time, there were very few reliable software packages that could do this, but today, there are a number of options out there to image a system that can do everything that is described here in an automated fashion. However, the basic idea behind what is done in the book and what these software packages do is the same, so if you are curious how they work, a quick read will give you the basics.

The next three chapters are the last major section of this book. Once the cluster is built, the next hurdle is using it. Chapters seven through nine deal with parallel computing and programming. Chapter seven is title "Parallel Applications." As explained in the chapter introduction, "...a beowulf [may] be the user's first experience with parallel computing. The purpose of this chapter is to offer some guidelines, examples, and advice in designing parallel applications suitable to Beowulf systems." Time in this chapter is spent covering categories of parallel algorithms, process-level parallelism, and some overheads you may encounter.

Chapter eight is devoted to explaining MPI, the "Message Passing Interface." The author does a good job explaining MPI, and includes examples and sample code to help guide the reader through. Chapter nine is then devoted to "Programming with MPI - A Detailed Example." This chapter is devoted to the writing, and analysis, of MPI programs.

The last chapter is a conclusions chapter, talking about what the author sees in the future for Beowulf computing.

Overall, I very much enjoyed this book. This is the book I wish I had read when I first wanted to learn about Linux clustering. Unfortunately, I did not; I did not get the opportunity to read this book until I had been playing around with clustering for a few years. Consequently, much of the information provided were things I had already learned, one way or another. However, even though some of the information in this book is dated, the concepts are sound, and for someone just starting out, this book is an excellent read. I would highly recommend it.