Research And Implementation Of Cloud Resource Scheduling Algorithm For Kubernetes

With the advent of 5G and the Internet of Things era,cloud computing,as the technological foundation,is constantly iterating and updating.As a mainstream container choreography technology,kubernetes is used as the core framework of cloud computing data center by manufacturers.However,its scheduling algorithm still has some shortcomings.First,there is the problem of unbalanced cluster load in resource allocation.Secondly,the static threshold scaling capacity algorithm has a lag problem.The current algorithm triggers capacity expansion only when the load reaches the threshold.When services are idle,capacity expansion cannot be implemented to ensure resource utilization and service quality cannot be expanded to ensure service quality when service access volume increases rapidly.Aiming at the above problems,this paper proposes a cloud resource scheduling algorithm based on Kubernetes,which is helpful to solve the load imbalance problem of cluster and the lag problem of static threshold algorithm,and to improve resource utilization rate and user service experience.The main contents of this paper are as follows:(1)Aiming at the cluster load imbalancing problem,an improved random walk optimization scheduling method is proposed,which takes the cluster load balancing as the optimization objective and solves the optimal scheduling scheme.The Kubernetes cluster built in this paper realizes a custom scheduler based on random walk algorithm to complete the experiment.Experimental results show that compared with PSO and Kubernetes default scheduling algorithm,the load balancing index point of random walk algorithm is reduced by 5.56%,and the load difference index point decreased by about 8.74%,making the cluster have better load balancing effect.(2)Aiming at the lag caused by static threshold scaling of Kubemetes cluster,an Informer load prediction model based on Gaussian noise interference is proposed.By adding gaussian noise,the model can improve the generalization ability in production environment.Experimental results show that the MAE decreases about 15.9%,RMSE decreases about 11.38%,and Psocre increases about 15.67%in the prediction of CPU utilization by Informer model with gaussian noise interference.Elastic scaling based on the model can ensure the quality of service and the reliability of the cluster,and improve the utilization of cluster resources.(3)Combined with the resource scheduling model and load forecasting model proposed in this paper,a Kubemetes resource management and scheduling platform is designed and implemented.The system includes resource monitoring module,resource control module and user interaction module.The resource monitoring module periodically collects information of cluster resource objects.The resource control module can manage resource objects in a cluster and application services in a cluster.The system can also visually display the information of various resources and application services in the cluster.