This proposal funds the purchase of a cluster of processors on which to develop and test cluster management software that exploits database system technology to improve system manageability, usability, and scalability. One particularly important problem facing cluster management systems is the vast amounts of data such systems must handle - both system data and user data. The central goal of the proposed work is the invention, implementation, and evaluation of data-centric (rather than process-centric) cluster management techniques.
This project is intended to have an impact on users running their computations on the cluster managed by the prototype system, to have an impact on the design of future computational cluster management systems, and ultimately to contribute to the increased efficiency (both computational efficiency deriving from better use of hardware and human efficiency deriving from an easier-to-use system) provided to the users of future cluster computing systems.
