Research And Implementation Of Multi-Cluster Container Cloud Resource Scheduling Mechanism Based On Edge Compnutig

The continuous development of the new generation of information technology has brought profound changes to the Internet industry,and it has also put forward new requirements for the computing model.Centralized cloud computing cannot meet the computing resources and bandwidth requirements of massive edge data processing,nor can it meet the real-time and privacy requirements of applications.Due to the lack of cloud computing,the concept of edge computing has begun to receive widespread attention.The distributed edge computing model is considered as a supplement and extension of cloud computing.By sinking the computing power of the cloud to the vicinity of users,it can effectively alleviate the computing load of the cloud center and the bandwidth pressure of the core backbone network.Any computing model is inseparable from the support of the underlying virtualization technology.Container virtualization is considered to be the underlying technology foundation for future edge computing implementation due to its efficient performance.Considering the limited resources and heterogeneity in the edge environment,it is a major research direction of edge computing to study the scheduling problem of container-based applications in limited resources.Since edge computing does not yet have mature standards and systems,this paper combines container technology and edge computing to build an edge container cloud cluster.Based on practical applications,this paper studies the working framework of multiple edge container cloud clusters and their resource scheduling strategies.This paper first analyzes the spatial-temporal difference of edge container cloud load and the delay-sensitive demand difference of edge application in multi-cluster environment,and proposes a multi-cluster edge cloud framework adopting master-slave mode management.The framework is divided into two parts:the edge cloud cluster and the collaboration layer.The edge cloud cluster performs specific workloads,and the collaboration layer is responsible for edge cluster and edge application management.The framework divides edge applications into two categories,delay-sensitive applications and delay-insensitive applications,and distinguishes them by domain name.The delay-sensitive application requests are directly processed in the nearby edge cloud,and the delay-insensitive application request completes the request proxy and forwarding through the cooperation layer,thereby effectively solving the application positioning problem under the multi-cluster edge cloud coordinated scheduling.On this basis,this paper studies the resource scheduling problem of delay sensitive applications in the framework.Considering that delay-sensitive applications often require high quality of service,in the edge environment where application load dynamics change,responsive dynamic scaling strategy cannot provide effective service guarantee due to its lag.Aiming at this problem,this paper proposes an active dynamic scaling strategy based on gray model and weighted moving average model.This paper refers to it as GGHHA(Gray and Moving for Horizontal Pod Autoscaling strategy).GMHPA uses the gray model to predict the application load,and compares with the actual value to judge the load change trend.Then,according to the chancle trend,select different model prediction results to calculate the application copy number.Experiments show that compared with the existing responsive strategy,the GMHPA strategy can effectively realize the delay-sensitive application to expand capacity when the load rises,and to reduce the capacity when the load decreases,and better meet the service quality requirements of the application.Finally,in order to improve the overall resource utilization rate of the cluster container cloud load in multi-cluster environment,this paper proposes a cross-cluster resource scheduling strategy for delay-insensitive applications,referred to as DICCS(Delay Insensitive Cross-Cluster Scheduling).The strategy divides the scheduling process into three steps.First,the cluster is classified according to the load usage rate.Then,according to the classification result,the application to be scheduled is selected according to the scheduling trigger factor in the high-load cluster,and finally the destination cluster is allocated for the application to be scheduled.Experiments show that the DICCS strategy can effectively complete the cross-cluster scheduling function of delay-insensitive applications,improve the resource utilization rate of the cluster,and achieve load balancing between clusters.