Features
- Cover Type: Paperback with 368 pages
- Published by: O'Reilly Media, Inc. November 16, 2004
- Written in: English
- ISBN 10 Number: 0596005709
- ISBN 13 Number: 978-0596005702
-
Book Dimensions:
9.1 x 7 x 0.9 inches
- Weighs: 1.4 pounds
Product Description
To the outside world, a "supercomputer" appears to be a single system. In fact, it's a cluster of computers that share a local area network and have the ability to work together on a single problem as a team. Many businesses used to consider supercomputing beyond the reach of their budgets, but new Linux applications have made high-performance clusters more affordable than ever. These days, the promise of low-cost supercomputing is one of the main reasons many businesses choose Linux over other operating systems. This new guide covers everything a newcomer to clustering will need to plan, build, and deploy a high-performance Linux cluster. The book focuses on clustering for high-performance computation, although much of its information also applies to clustering for high-availability (failover and disaster recovery). The book discusses the key tools you'll need to get started, including good practices to use while exploring the tools and growing a system. You'll learn about planning, hardware choices, bulk installation of Linux on multiple systems, and other basic considerations. Then, you'll learn about
software options that can save you hours--or even weeks--of deployment time. Since a wide variety of options exist in each area of clustering software, the author discusses the pros and cons of the major free
software projects and chooses those that are most likely to be helpful to new cluster administrators and programmers. A few of the projects introduced in the book include:
- MPI, the most popular programming library for clusters. This book offers simple but realistic introductory examples along with some pointers for advanced use.
- OSCAR and Rocks, two comprehensive installation and administrative systems
- openMosix (a convenient tool for distributing jobs), Linux kernel extensions that migrate processes transparently for load balancing
- PVFS, one of the parallel filesystems that make clustering I/O easier
- C3, a set of commands for administering multiple systems
Ganglia, OpenPBS, and cloning tools (Kickstart, SIS and G4U) are also covered. The book looks at cluster installation packages (OSCAR & Rocks) and then considers the core packages individually for greater depth or for folks wishing to do a custom installation. Guidelines for debugging, profiling, performance tuning, and managing jobs from multiple users round out this immensely useful book.
About The Author
Joseph D. Sloan has been working with computers since the mid-1970s. He began using Unix as a graduate student in 1981, first as an applications programmer and later as a system programmer and system administrator. Since 1988 he has taught computer science, first at Lander University and more recently at Wofford College where he can be found using the
software described in this book.
Reader ReviewsThe free open source nature of Linux has driven its growth in general purpose client and server side usages. Here, Sloan takes linux into the rarefied context of high performance computing. Atop linux, he explains the merits of open source packages like Oscar and Rocks, to run your cluster. The basic motivation for him describing all this is the relatively low cost of using the machines. This can be a significant issue if your budget is limited or if you plan to have many machines in the cluster. The book is primarily about software. Though he also gives a chapter discussing mundane but important decisions regarding hardware. The software that is explained is mostly Oscar and Rocks, as explained above, and how these are to be run. Be aware that relatively little of the book is about linux, per se. Which is as it should be. The crucial starting assumption is that you are or will be using linux. But, roughly, linux on these machines is more or less the same as linux on a generic computer. The distinguishing feature is the next layer of software. On the programming side, Sloan points out that C and Fortran dominate, with C++ usage rising. There is no significant effort in Java, because of its performance penalty. Maybe on the cluster's lead computer that interfaces with the rest of the world, you can have a nice Java GUI program that controls the cluster. But the heavy lifting is done in the other languages.
Search for books at
