Research On Technology Of Kubernetes For IIoT

Kubernetes has been widely used in resource management of various scenarios with its powerful cluster choreography management capability.However,the existing Kubernetes technology has not yet met the needs of IIoT scenarios.In view of the deficiency of Kubernetes load pressure prediction and resource scheduling technology in IIoT scenario,this thesis improves the existing technology,so as to predict the load state of the cluster in IIoT scenario,achieve accurate,flexible and efficient resource scheduling,and finally apply Kubernetes technology in IIoT scenario truly.Based on the above problems,first of all,this thesis proposes a multi-resource scheduling(MRS)strategy under the industrial Internet of Things,which comprehensively considers the resource requirements of the tasks to be deployed and the current resource usage of the cluster system,and dynamically weighs CPU resources.The importance of multiple resources such as memory resources,network bandwidth resources,and disk space resources has improved the scoring strategy of the default scheduler,so that the resources of the cluster can be reasonably scheduled in the IIoT environment.In addition,this thesis also focuses on the delay limitation after Pod deployment,so that after Pod deployment,it can meet the low-latency requirements in IIoT scenarios.Experiments prove the effectiveness of this method.Secondly,this thesis proposes a cluster load pressure prediction method based on EMA,which predicts the future load status of the cluster by learning the historical resource usage of the cluster,and guides the cluster scheduling according to the cluster status.The EMA feedforward neural network(EMA-FNN)proposed in this thesis finds a more compact bases set by estimating the maximum likelihood of parameters(bases),thereby obtaining a low-noise output and greatly reducing the computational complexity.On this basis,in order to improve the robustness of the prediction method,we fused EMA-FNN with RF and XGBoost.Experiments have shown that the prediction results obtained by this prediction method are more accurate.Finally,based on the above research,this thesis details the actual application of Kubernetes in the industrial Internet of Things scenario,including cluster monitoring solutions,load forecasting solutions,and resource scheduling solutions.This scheduling strategy first monitors,collects,stores and visualizes the resource usage information in the cluster,then performs cluster load prediction based on EMA based on the collected information,and finally performs multi-resource scheduling in the IIoT scenario.