Optimizing Utility in Cloud Computing through Autonomic Workload Execution.PDF

Optimizing Utility in Cloud Computing through Autonomic Workload Execution Norman W. Paton, Marcelo A. T. de Arag˜ao, Kevin Lee, Alvaro A . A. Fernandes, Rizos Sakellariou School of Computer Science, University of Manchester, U.K. (norm,maragao,klee,rizos,alvaro)@cs.man.ac.uk Abstract Cloud computing provides services to potentially numerous remote users with diverse requirements. Al- though predictable performance can be obtained through the provision of carefully delimited services, it is straightforward to identify applications in which a cloud might usefully host services that support the composition of more primitive analysis services or the evaluation of complex data analysis requests. In such settings, a service provider must manage complex and unpredictable workloads. This paper describes how utility functions can be used to make explicit the desirability of different workload evalu- ation strategies, and how optimization can be used to select between such alternatives. The approach is illustrated for workloads consisting of workflows or queries. 1 Introduction Cloud computing essentially provides services; shared computational resources execute potentially diverse re- quests on behalf of users who may have widely differing expectations. In such a setting, someplace in the architecture, decisions have to be made as to which requests from which users are to be executed on which computational resources, and when. From the perspective of the service provider, such decision making may be eased through the provision of restrictive interfaces to cloud services, as discussed for cloud data services in the Claremont Report on Database Research [1]: Early cloud data services offer an API