Cost-aware real-time job scheduling for hybrid cloud using deep reinforcement learning

Hybrid cloud computing enables enterprises to get the best of both private and public cloud models. One of its primary benefits is to reduce operational costs, and the prerequisite is that jobs should be executed in an effective way in the hybrid environment. Although many job scheduling methods have been proposed for cloud in the past decade, most of them focus on handling batch jobs rather than real-time ones. Moreover, few of them have ever considered real-time jobs in hybrid cloud. Inspired by the recent success of using deep reinforcement learning (DRL) for solving complex optimization problems, in this paper, we propose a DRL-based approach for scheduling real-time jobs in hybrid cloud, with a focus on optimizing monetary cost for job executions while ensuring that high quality of service and low responsible time can be also achieved. Specifically, our method can learn to make appropriate decisions in selecting suitable virtual machines for incoming jobs in real-time over hybrid cloud, with the scheduling agent getting trained through rewards in its learning experiences. We give the detailed design of our approach, and our experimental results demonstrate that our method is more cost-efficient, compared to the current approaches.