Research On Autoscaling Algorithms For Microservices In Edge Clouds

With the development of cloud computing and microservices,the number of users in the cloud is increasing,and the number of load request peaks is greatly increased,which leads to load variation plus aggregation.To cope with the dynamically changing load requests,the microservices autoscaling(Autoscaling)technology has emerged.This technology can automatically scale containers in the cloud based on dynamic changes in load.However,the traditional central cloud microservice autoscaling approach is not applicable to edge cloud environments.Since devices in edge clouds generally have lower arithmetic power,they can easily lead to scaling timeout violations and poor resource handling capabilities when coping with dynamic changes in load.To address the above problems,this thesis introduces improvements and innovations in two aspects of the autoscaling algorithm,namely,machine learning prediction and load balancing joint scaling,and deploys an autoscaling monitoring system in the hardware environment to achieve performance monitoring.The specific research points are as follows.(1)To address the problem of excessive resource consumption of microservice autoscaling in edge cloud environment,this thesis proposes a hybrid autoscaling method(Predictively Horizontal and Vertical Pod Autoscaling,Pre-HVPA)based on load prediction.The method first selects the machine learning model with the lowest error to predict the load data features and obtains the load prediction results.The predicted load cases are then imported into the microservice hybrid autoscaling module proposed in this thesis.The simulation results show that based on this method,37.2% scaling jitter and 14.7% container usage number of microservice autoscaling in edge cloud can be reduced.(2)To address the problem of high latency of microservice autoscaling in edge cloud environments,this thesis proposes a Jointed Load Balancing for Horizontal and Vertical Pod Autoscaling(LB-HVPA)approach for microservice hybrid autoscaling.This approach avoids additional scheduling due to unknown load distribution scenarios by integrating microservice load balancing results into the hybrid auto-scaling module proposed above with centralized unmonitored load awareness.Simulation results show that the jointed load balancing algorithm can effectively reduce the average response time of microservice autoscaling by 41.9% and the number of timeout violations by 37.9%,allowing the auto-scaling algorithm to be better applied in edge cloud environments.(3)In order to better monitor the performance of clusters and scaling,this thesis actually builds a K8 S platform in an edge cloud environment and configures tools such as Prometheus and Grafana to achieve performance monitoring of clusters and automatic scaling processes.And through Locust load testing tool,some of the proposed scaling algorithms are compared and validated,and the experimental results prove that the scaling strategy proposed in this thesis achieves better results.