Utility-based Resource Management for Cluster Computing

The vision of utility computing is to offer computing services as a utility so that users only pay when they need to use. Hence, users define their service needs and expect them to be delivered by utility computing service providers. However, most current high performance computing resources which constitute clusters of computers do not consider user-centric service needs for resource management. They still adopt system-centric resource management approaches that focus on optimizing overall cluster performance. To address this problem, we investigate how market-based resource management can enable utility-based resource management for cluster computing and have: • developed a taxonomy to understand existing market-based resource management systems and analyze the research gap to support utility-based cluster resource management, • designed and evaluated three resource management policies: Libra+$ to provide commoditybased pricing of cluster resources, LibraSLA to manage penalties for Service Level Agreement (SLA) based resource management, and LibraRiskD to manage the risk of deadline delay for job admission control, • proposed the use of risk analysis by computing service providers to ensure that their essential objectives are achieved, and • demonstrated the need for a utility computing service to adopt autonomic metered pricing. This is to certify that (i) the thesis comprises only my original work, (ii) due acknowledgement has been made in the text to all other material used, (iii) the thesis is less than 100,000 words in length, exclusive of table, maps, bibliographies, appendices and footnotes.