Find in Library

A new kind of virtual machine instance named burstable instance is recently introduced in leading cloud providers, which can help cloud user handle fluctuating demands with reduced cost. There is an impending need to incorporate burst instances in common scheduling. Due to the time varying provisioning of burstable resources, the stochastic demand oriented scheduling can be a better choice where the demands are represented using distribution functions instead of mean values. However, little has been done to the scheduling of the burstable resources with stochastic demands. To address the problem, we present the first stochastic revenue model for the burstable resources, and propose a bounded delta-solution based differential evolution algorithm, which exploits inherent information of the problem to accelerate the searching of optimal solution. Experiments using simulated and real-world data indicate that the algorithm outperforms the existing representative algorithms and can increase the revenue significantly.