How to Build a Beowulf: A Guide to the Implementation and Application of PC Clusters (Scientific and Engineering Computation) [Paperback]

Thomas Sterling (Author), Daniel F. Savarese (Author), Donald J. Becker (Author), John Salmon (Author)

Book Description

Publication Date: May 28, 1999 | Series: Scientific and Engineering Computation

Supercomputing research--the goal of which is to make computers that are ever faster and more powerful--has been at the cutting edge of computer technology since the early 1960s. Until recently, research cost in the millions of dollars, and many of the companies that originally made supercomputers are now out of business.The early supercomputers used distributed computing and parallel processing to link processors together in a single machine, often called a mainframe. Exploiting the same technology, researchers are now using off-the-shelf PCs to produce computers with supercomputer performance. It is now possible to make a supercomputer for less than $40,000. Given this new affordability, a number of universities and research laboratories are experimenting with installing such Beowulf-type systems in their facilities.This how-to guide provides step-by-step instructions for building a Beowulf-type computer, including the physical elements that make up a clustered PC computing system, the software required (most of which is freely available), and insights on how to organize the code to exploit parallelism. The book also includes a list of potential pitfalls.

Editorial Reviews

Amazon.com Review

How to Build a Beowulf covers the essentials of today's "cheap supercomputing" that is available with off-the-shelf PC hardware running Linux. Filled with advice from the experts, this book is a working guide to the essentials of planning, installing, and running a Beowulf cluster.

After an introduction to Beowulf and parallel computing in general, the authors describe the advantages to and organization of a typical Beowulf setup. They next describe the basic PC hardware (which will be familiar to many Intel users). The do-it-yourself impulse in Beowulf supercomputing is strong, and the authors show how to choose everything from a CPU and memory to networking options (including TCP/IP basics and Fast Ethernet). They cover hardware and software installation and the basics of configuring Linux on Beowulf nodes (which do the work of parallel processing).

Next the book covers issues of security and system administration of a Beowulf cluster. (Here the authors strike a balance between accessibility and security with the concept of a "guarded Beowulf.") They cover a variety of Linux utilities for remote computing and administration.

An essential piece of Beowulf technology is the Message Passing Interface (MPI), a set of APIs that permit programmers to develop parallel programs in C/C++ and FORTRAN. With MPI, programs running on different CPUs can pass messages and share the same data. The samples that round out this book are excellent--a ray-tracing example, a parallel sorting algorithm, and a cellular automata program. The authors do a good job of explaining the issues of taking advantage of parallelism within Beowulf software. --Richard Dragan

About the Author

This how-to guide provides step-by-step instructions for building aBeowulf-type computer, including the physical elements that make up aclustered PC computing system, the software required (most of which isfreely available), and insights on how to organize the code to exploitparallelism.

John Salmon is Staff Scientist at the California Institute of Technology.

Donald J. Becker is Staff Scientist at the Center for Excellence in Space Data and Information Sciences.

Product Details

• Paperback: 238 pages
• Publisher: The MIT Press (May 28, 1999)
• Language: English
• ISBN-10: 026269218X
• ISBN-13: 978-0262692182
• Product Dimensions: 9 x 8 x 0.7 inches
• Shipping Weight: 1.4 pounds (View shipping rates and policies)
• Average Customer Review: 2.3 out of 5 stars  See all reviews (13 customer reviews)
• Amazon Best Sellers Rank: #1,415,514 in Books (See Top 100 in Books)

More About the Author

Discover books, learn about writers, read author blogs, and more.

Customer Reviews

Most Helpful Customer Reviews

26 of 27 people found the following review helpful:

5.0 out of 5 stars Excellent starting point, November 7, 1999

By 

Chester D. Fitch (Denver, CO) - See all my reviews
(REAL NAME)   

This review is from: How to Build a Beowulf: A Guide to the Implementation and Application of PC Clusters (Scientific and Engineering Computation) (Paperback)

I must take issue with some of the other reviewers of this text. I found this book to be a very good overview (and snapshot) of the current status of a rapidly evolving system architecture. It is written at a fairly high level (although it does dip periodically into details) - so the reader gets a good overview of what a Beowulf is, its' components, and many of the issues involved in designing and implementing one. In my view it balances detail and theory quite well.

IT IS NOT (nor do I believe it was intended as) a detailed roadmap of EXACTLY how to build one. The Beowulf architecture isn't so much a single type of implementation, but rather an approach to applying COTS technology to solving computational problems. The details of any single Beowulf implementation depend greatly on the specific computational problem being attacked. (Something that is pointed-out within the book.) The authors therefore took a different approach.

Some of the topics covered in the book WILL, eventually, be outdated: specifically, the section on the PCI bus, some of the material on network technology, and the section on available processors. As COTS technology advances, and as Beowulf architectures change to take advantage of those advances, some sections will become outdated. However, this is unavoidable for any text reviewing the current state-of-the-art. There is also a lot more here that is NOT likely to be outdated within the next several years..

There may also be sections in the text that the reader will already be familiar with, and can therefore skip. This is also inevitable considering the nature of the text and will obviously vary depending on the reader.

I can recomend this text highly as a starting point in learning what a Beowulf is, some of the ways they are put together, and for exploring many important design and implementation decisions. In my own case, it helped me resolve a number of design issues I was wrestling with about my own system. It does not, however, stand alone. After starting with this text, most readers will then certainly need to refer to online sources for further information.

14 of 16 people found the following review helpful:

2.0 out of 5 stars Very disappointing, June 11, 2000

By 

Poser - See all my reviews

This review is from: How to Build a Beowulf: A Guide to the Implementation and Application of PC Clusters (Scientific and Engineering Computation) (Paperback)

I was hoping for great things from this book, but as a lot of the other reviewers have said very disappointing. To much time was spent on linux, which if you are going to be building a beowulf you would know, and even if you didn't there is a endless supply of linux knowledge on the net which greatly exceeds what is in this book.

At the end of the day I DID build my own low-end beowulf, but sadly I can't say that any of the information that I needed was found in this book. It might have been there but the unorganized layout mad it impossible to find anything.

7 of 8 people found the following review helpful:

1.0 out of 5 stars Useless in 1999, a total waste of money in 2005, January 8, 2005

By 

Ripped Off (Izmir, Turkey) - See all my reviews

This review is from: How to Build a Beowulf: A Guide to the Implementation and Application of PC Clusters (Scientific and Engineering Computation) (Paperback)

As most other readers have elaborated, the book has little or no real substantive material concerning the actual building of a beowulf. Everyone knows Myrinet is better than gigabit ethernet, bu HOW MUCH better for a given class of problems? For any problem???

The book does not even consider addressing real issues and configuration choices, but babbles with obvious choices like ssh vs rsh etc, for pages and pages. I feel I was ripped off. These guys seem to be just resting on their laurels and apparently it doesn't seem to bother them that they are giving the intended audience of the book no real value whatsoever. I believe they should either research the matter thoroughly and completely rewrite the book, or just withdraw it from circulation. One can get orders of magnitude more relevant and reliable information just by reading HOWTOs and Googling around.

I just wish I had read the other reviews BEFORE buying the book.