Building Linux Clusters [With CDROM]

While I can sympathize with some readers who may have been expecting a different book that covered the topic of cluster building in a different way, I would like to respond with several facts to clarify issues surrounding "Building Linux Clusters":1) The book, as published, is not the final draft given to the editors at O'Reilly. When the book was ready to be put into O'Reilly's publishing system the then-editor of this series left to pursue a career at VALinux; the copy of the book put into the system is missing was over 2 months old and missing 50-100 pages of changes/additions/updates and other corrections that are not reflected in the printed book. This was discovered by me only when the book was delivered to stores -- at which time (obviously) it was too late to do anything about it. 2) Despite the issues surrounding which version of the book went into production, the book sold over 10,000 copies out of a printing of 12,000: O'Reilly decided not to issue a 2nd edition for a number of reasons, including the fact that many of the facilities I documented in the book (such as MPI, PVM and tools for cluster management, parallel rendering, etc.) were subsequently available pre-installed in many major Linux distributions which obviated the need for a dedicated book on cluster building. 3) Many, many readers have had great success with the book and have written to tell me so, including several readers who were able to build 100+ node clusters in a matters of a few hours using the software provided with the book. 4) A large number of software-related problems experienced by readers came about as readers attempted to build new distributions (RedHat7, RedHat8, etc.) based on the original software (based on RedHat6.2) provided with the book. Where possible I proved many hundred of hours of assistance to such readers.5) Updated software and other tools were made available both at O'Reilly's web site as well as my own to help readers who found that certain tools would not work on their configurations and have been downloaded and successfully used thousands of times.

Most of the O'Reilly books I have looked at don't read as well as this one does. I am at working building a super computer now. While the author is involved in stock market work I don't find him condesending or snobby or like he doesn't have time for me. He tells the straight story of how to build a super computer with all the practical details you could want. I read it from the school library first and thought I wouldn't buy it myself but I have and now am really enjoying reading it. I am learning all kinds of things in the process.

This is an excellent first book about building a Beowuff type cluster. The book is not meant to be a primer on parallel computing, networking or computer architecture. However, the reader is introduced to all the necessary concepts for connecting COTS parts into a useful computing entity. Like so many other "How-To-Do" books tackling complex projects, it must rely on other sources of information to fill in the gaps -- it is like a good roadmap that will not indicate each individual pothole but will point out the Grand Canyon and other necessary sights. This focus produces a crisp reading book. The author introduces a large range of useful tools and packages that will aid in the care and feeding of the cluster. Unfortunately, my personal preference when introduced to a single tool for a particular problem (and knowing that in this field that there are numerous tools written to address the same problem), I wish to know more about the introducer -- after all, who is this person to say that this is the best tool for the problem? I can't help but feel that the author has held one or two significant positions, contributed to magazines or conferences in this field, or at least contributed to the Usenet news groups (and of which nothing of the sort is alluded to on the back of the book). But, this personal feeling of incompleteness is nothing compared to the usefulness of the book.

Building Linux Clusters by David HM Spector describes how Beowulf software can be used to turn network clusters of ordinary PC's into a virtual supercomputer. The first four chapters are a good introduction to the concept of effeciently connecting numerous computers together to act as a single computer. The author goes into useful detail about how a multiprocessor environment can use parallelism to get the most out of CPUs, high speed cache memory and slower main memory, refresh rates, wait states, bus speeds, and other machine level topics such as syncronization. There is detailed information about what kinds of hardware is and is not recomended for clusters.The software used to harness the parallel power of the cluster is also introduced in detail, discussing different technologies such as Symmetric Multi-Processing (SMP), Multiple Instruction Multiple Data (MIMD), and others. The basics of Ethernet, TCP/IP, and routing are reviewed and applied to Clustering. There is lots of intersting information about network speed and configuration as is applies to clusters, such as Cube and Hyper-cube clusters. Software included with Beowulf to provide RAID support accross the hard drives on the cluster is explained, and comes with good advice about how to handle storage on the cluster as a whole rather than on individual machines.There are examples of applications that run on clusters, such as MUDs, POVRAYs, and MP3 crunchers, and parallel programming and compilers.This book was easy to read and very informative, and has put me well on my way to building my own hetrogeneous cluster. I have always wanted to study parallel computing and programming, and now I will have a chance.