Ningaui: A Linux Cluster for Business

Clusters of commodity PC hardware are very attractive as a basis for large scale computing. In fact, this style of system, commonly referred to as Beowulf systems, are well on their way to dominating the supercomputing arena, which is almost solely concerned with large scientific computing. In other domains, issues other than performance predominate.

This paper describes Ningaui, an architecture for computing on partitioned datasets using a cluster of loosely coupled computers. Although performance is a goal, it is dominated by the goals of availability, reliability, scalability and manageability. We describe the architecture, together with a large-scale application, and describe how we met our goals. We also discuss how well our platform fared, and the lessons we have learned. As our environment is jargon rich, we include a glossary.