Editor's Note: Be sure to read the other articles in this series, Building a Linux Cluster, Part 1: Why Bother? and Building a Linux Cluster, Part 2: What's Involved?.
In the previous two articles in this series, we examined some of the whys and whats of building Linux clusters. This article concludes our series by concentrating on the hows of cluster building. We've seen that a clustered approach to certain computing solutions can save lots of money in hardware and support costs. Now our job is to produce a method of building clusters that's repeatable and predictable—we don't want to give back our hard-won savings in project cost overruns.
This means, to everyone's discomfort, that we'll have to use and discuss the "p word": process. Its mere mention can send a shudder through any creative person, but this reaction should not be automatic. Processes are necessary for repeatable steps leading to predictable outcomes, but how the particular process is defined and implemented becomes (and remains) the issue affecting our attitude toward that process.
To get started, this article presents a three-phase process for designing a
cluster, comprising design, installation, and testing
phases. Each phase produces necessary information or deliverable items required
to complete the cluster. The process can be adjusted for your particular
situation and requirements, of course; what I'll describe is a prototypical
process for building clusters. (Hey, we said in