Cloud Computing Resource Scheduling Algorithm Based On Reinforcement Learning

Cloud computing is a new computing model that provides computing services on demand,and it has been increasingly used in recent years.In the cloud computing model,users can quickly and easily expand the basic service architecture that changes with business needs at a small cost,without having to spend high hardware costs.In order to provide high-quality services,cloud service providers need to use cloud computing resource scheduling strategies to efficiently match their own computing resources with cloud user task requests to maximize their benefits.Therefore,how to design an efficient cloud computing resource scheduling algorithm is a research hotspot in the field of cloud computing.In order to solve the problems faced by cloud computing resource scheduling,this paper proposes a cloud computing resource scheduling algorithm DQNVMS based on deep reinforcement learning to reduce task response time and low energy consumption by modeling the cloud computing resource scheduling problem as an MDP problem..The algorithm first preprocesses tasks based on parameters such as task length,task deadline and task waiting time to divide task priorities.Then,according to the different virtual resources required by the task,we divide them into different queues,in turn The task is scheduled to the corresponding virtual resource.At this time,the corresponding virtual resource will be in a busy state.Finally,the scheduling module based on the DQNVMS algorithm will obtain the physical hardware facility status and the virtual machine resource status through the monitoring module,and give a scheduling plan,that is,create or Destroy different types of virtual machines.According to the different focus of scheduling goals,we can set different parameter values for different indicators to improve scheduling efficiency.The main work of this paper has the following three points.First,the cloud computing resource scheduling is modeled as a Markov decision process,and the DQNVMS algorithm based on deep reinforcement learning is further used to process cloud computing resource scheduling.Secondly,in order to make the scheduling decisions made by reinforcement learning meet the actual physical resource constraints,we added a penalty for resource exceeding the limit to the reinforcement learning reward to reduce unreasonable scheduling actions.At the same time,creating or destroying virtual machines too frequently will bring additional overhead,so we also reduce unreasonable scheduling behaviors by adding penalty items for scheduling actions.Finally,this paper conducts simulation experiments based on Deeplearning4j software package and CloudSim platform to verify the influence of algorithm parameters on the results,and compares with other algorithms to prove the effectiveness of this method.