Resource Prediction And Scheduling Of Kubernetes Based On LSTM And Particle Swarm

With the rapid development of container technology,many Internet enterprises are gradually moving towards container technology.The resource scheduling module and resource allocation of Kubernetes based container scheduling tool can only adjust the CPU resources,which is difficult to solve the problem of continuous container restart caused by memory overflow caused by changing load.It is difficult to consider the problem of resource consumption and load balancing when deploying multiple containers.To solve these two problems,this paper will focus on the kubernetes resource prediction and scheduling method.By studying the scheduling algorithm of the container scheduling tool,we can effectively allocate the container cluster resources,improve the resource utilization rate,and minimize the resource consumption cost.The specific research work mainly includes the following three aspects:(1)A prediction method of kubernetes resources based on recurrent neural network is proposed.Aiming at the problems of slow training convergence and low precision in traditional prediction model,a PP-LSTM(Pictures Prediction Long Short Term Memory)method based on deep learning is proposed.In this method,attention gate is introduced to optimize forgetting gate and input gate,and the traditional LSTM model is improved to get PP-LSTM,which is used to predict the use of kubernetes resources.The experimental results show that the prediction accuracy of PP-LSTM is improved by nearly 8% on average,and PP-LSTM deep learning model can effectively predict the cluster load change caused by traffic change.(2)The kubernetes resource scheduling method based on particle swarm is proposed.For the resource consumption model of the original scheduling model,it is difficult to balance the service resources with multi Pod deployment scheduling and long running time.The meta heuristic algorithm is studied to solve the problem of cloud resource scheduling,and a new particle swarm optimization model is proposed.First,the scheduling model is established,then the adaptive rules of speed selection are adjusted,and finally the particle swarm optimization algorithm and its method for kubernetes resource scheduling are realized.The experimental results show that the improved algorithm improves the resource utilization by nearly 20% compared with other algorithms.(3)A hybrid scheduling method of kubernetes resources based on elastic scaling is proposed.By constructing resource monitoring and collection module,resource prediction module,auto scaling module and resource scheduling module respectively,kubernetes resource hybrid scheduling scheme is constructed.The monitoring data is transferred to the resource prediction module,and the result of the prediction module is obtained by the automatic expansion module,which will automatically expand,and the resource scheduling module will allocate the expanded Pod to each node.The experimental results show that this method can be used to improve the utilization of resources,and reduce the possibility of service breakdown and enhances the quality of service.