The Research On Energy Efficiency Improvement Of Cloud Datacenter Based On Reinforcement Learning

Under cloud computing environment,the expansion of data center and the introduction of new technology constantly improves the flexibility of resources to meet more emerging high-performance computing needs,but it also makes the cloud data center more complex.The integration and efficient use of existing resources is the most noteworthy.The goal of resource scheduling in this thesis is to achieve the optimal scheduling of energy consumption for the tasks submitted by users,and make the resource utilization as high as possible on the premise of meeting the needs of users.Based on the research of the key energy improvement technologies that affect the workload of cloud data center system and the standard of evevaluating energy consumption of cloud data center,this thesis transforms the energy efficiency optimization problem of cloud data center as a Markov process model.Here design a scheduling strategy and optimization management system based on Q-Learning.The proposed algorithm can be solved without prior information of the system.Only calculate the energy consumption of the data center,then build the energy consumption model based on the utilization rate of CPU resources.The main service task collection and resource allocation of the system are operated to balance the resource allocation of workload.The acquired operations include adding/deleting virtual machines and closing/activating servers,so as to reduce the whole energy consumption of cloud data center.In this thesis,offering a software level of resource scheduling management way,combined with other technologies to provide server integration,virtual machine migration and so on.In the CloudSim simulation environment,we can simulate the dynamic changes of real cloud computing and resource utilization by setting dynamic random delay submission tasks.In the experiment,we observe the effect of scheduling strategy on the load of computing node in the time slice,and record the utitlization of each server and running task.Through the amount of computing resources consumed by the physical server running services in the time slice,it is determined how to allocate more virtual machine resources to perform the task operations.Through the experiment,it is found that the energy efficiency optimization strategy of reinforcement learning can reduce the total number of physical machines activated,so the energy consumption cost of corresponding state is relatively low.The feasibility and effectiveness of energy-saving resource scheduling method based on reinforcement learning are verified.It is proved that the proposed reinforcement learning mechanism can effectively improve the energy efficiency of cloud data center.