Automatic Installation and Configuration for Large Scale Farms

Since the early appearance of commodity hardware, the utilization of computers rose rapidly, and they became essential in all areas of life. Soon it was realized that nodes are able to work cooperatively, in order to solve new, more complex tasks. This conception got materialized in coherent aggregations of computers called farms and clusters. Collective application of nodes, being efficient and economical, was adopted in education, research and industry before long. But maintainance, especially in large scale, appeared as a problem to be resolved. New challenges needed new methods and tools. Development work has been started to build farm management applications and frameworks. In the first part of the thesis, these systems are introduced. After a general description of the matter, a comparative analysis of different approaches and tools illustrates the practical aspects of the theoretical discussion. CERN, the European Organization of Nuclear Research is the largest Particle Physics laboratory in the world. High Energy Physics calculations performed here involves an enormous computing capacity, which is achieved by large commodity-hardware clusters. For an automatic installation and configuration at the CERN Computer Center, the quattor system administration toolsuite is used. This framework gets a stronger emphasis in the thesis, by having a more detailed architectural overview. Being a technical student at the CERN Information Technology department, in the Fabric Infrastructure and Operations group between 2003.02.11 and 2004.03.31, I had the opportunity to get involved in operations for large scale productions farms managed by quattor. Here I gained experience with using the framework, which got deepened by related software development. My tasks included both building plug-in modules, and re-engineering system internals. The results of my work are explained in the last part of the thesis.