Tuning Genetic Algorithms for resource provisioning and scheduling in uncertain cloud environments: Challenges and findings

Cloud computing allows users to devise cost-effectivesolutions for deploying their applications. Nevertheless, the deci-sions about resource provisioning are very challenging becauseworkloads are seriously affected by the uncertainty of cloudperformance and their characteristics vary. In this paper weaddress these issues by explicitly modeling workload and clouduncertainty in the decision process. For this purpose, we adopt aprobabilistic formulation of the optimization problem aimed atminimizing the expected cost for deploying a parallel applicationunder a deadline constraint. To find a sub-optimal solutionof the problem we apply a Genetic Algorithm. By tuning itsparameters we are able to assess their role and their impact onthe effectiveness and efficiency of the algorithm for provisioningand scheduling in uncertain cloud environments.