Deep Q-Network Based Cloud Service Tasks Scheduling and Resource Allocation

Since the beginning of the global pandemic, users’ demand for cloud services has increased rapidly, and cloud service providers are also facing the challenge of minimizing energy costs in data centers. This paper proposes a Double Deep Q-Network based Cloud Service Tasks Scheduling and Resource Allocation algorithm (DDQN-CSTSRA), which is suitable for massive servers with lots of requests provided by users. The two-stage Resource Provisioning-Tasks Scheduling (RP-TS) process based on the DDQN algorithm is to achieve the purpose of outputting long-term optimal decisions by continuously learning the feedback from the environment corresponding to different decisions in various states. DDQN-CSTSRA achieves high cost of the energy efficiency, low uptime and fast convergence. Compared with the DQN-based algorithm, DDQN-CSTSRA has a significant advantage in the energy efficiency when the time loss does not increase, and this advantage is further expanded with the increase in the number of servers.