The why portion of this series justified Linux clusters as an economical, stable solution for replacing large SMP systems for the proper class of applications and services. The what portion of building clusters involves understanding the integration of not only the physical, visible components of the cluster, but also the virtual software components. Our job is to make the proper design choices to ensure that we don't lose the hardware savings that a cluster can provide.
We need to avoid the consequences of Lucke's First Law: "Ignorance simplifies any problem." To be successful with a cluster project, it's important to understand the wide range of design choices and make appropriate decisions that reflect the organization's business needs. The bad news is that the wide range of design choices requires a lot of knowledge or help from a knowledgeable source. The good news is the wide range of choices available to implement a cluster: custom designs, preconfigured hardware, commercially available cluster software packages, cluster software toolkits, and completely integrated cluster solutions.
The final article in this series will cover the how of building clusters. This will include presenting a formal process for a cluster project.