Optimal scheduling in the hybrid-cloud

The emerging hybrid cloud architecture allows organizations to optimize their computation needs and costs by maintaining their private computational infrastructure at high utilization and meeting peak requirements by offloading selected tasks to the public cloud. Consequently, there is a need to devise efficient systems equipped with online task cloudbursting algorithms that optimize the overall cost while maintaining adequate quality of service. Such algorithms must take into account the difference in communication and computational requirements associated with different tasks. For example, it is clear that when two tasks have the same local computational requirements, the one with the lower cloudbursting cost is a better candidate to be off-loaded and sent to the cloud. In this paper, we address the case in which arriving tasks have the same computational cost but different communication costs. We design scheduling system based on online decision algorithms driven by the user's local infrastructure constraints. We model the online scheduling problem as Markov Decision Process (MDP) problem and provide optimal policies for scheduling tasks either locally or remotely. We further explore the usage of MDP in different scenarios and prove the structural properties of the optimal policies in order to incorporate them into the decision engine. The design of the practical scheduling system is supported by the analytical results and numerical evaluations. We demonstrate the practical computational advantage of threshold type policies and provide an insight into their dependence on system parameters.